Integrating network analysis with in vivo–in situ manipulation to elucidate microbiome dynamics in Aedes albopictus

Abstract

Host-associated microbiomes have significant impacts on host biology and physiology, but the underlying processes governing their structure and assembly are not well understood. One approach to better understanding those process is the use of computationally driven modeling tools, such as network analysis to identify patterns of cooccurring taxa across microbiomes. Those patterns can then be tested to identify taxa that are potentially more important in the overall structuring and assembly processes. Here, we used network analysis to explore cooccurrence patterns within the microbiome of Aedes albopictus. We identified important nodes in the network using the centrality metrics of node degree and betweenness. Among the nodes with the highest centrality values, more ITS ASVs were present than 16S ASVs. We then tested the network analysis predictions in vivo/in situ in A. albopictus. A series of exclusion experiments were used to manipulate environmental microbiome source pools by filtering the source pool by cell size. Our results show that including microbial eukaryotes, such as fungi, in the source pool affects microbiome assembly and structure in A. albopictus, which aligns with the network analyses predictions of this system. To our knowledge, this is the first study to integrate microbial network centrality analysis with in vivo/in situ validation using filtration-based microbial community exclusion.

Interdomain associations in the Aedes albopictus microbiome shape microbiome assembly and structure.

Introduction

Microorganisms routinely enter symbiosis with metazoans to form microbial communities, often referred to as microbiomes. These host-associated microbiomes modulate many physiological processes of the host including development, metabolism, immunity, nutrient provisioning, and behavior (McFall-Ngai et al. 2013, Gould et al. 2018). Despite their importance to host biology, microbiomes can exhibit substantial diversity among individuals within a population (Falony et al. 2016, Frankel-Bicker et al. 2020). The origin of this diversity is not fully understood, but it has been hypothesized that environmental variation may be a significant driver (Mushegian et al. 2019). A considerable portion of the microbiome is environmentally acquired (Seabourn et al. 2023); thus, environmental factors can influence the structure of microbial communities on the landscape (Medeiros et al. 2022), ultimately shaping the regional source pool from which a host will draw (Amend et al. 2022).

Tests of variance in composition such as PERMANOVA and principal coordinates analysis, have provided significant insight into the taxonomic diversity of host-associated microbiomes, but these techniques often fall short of explaining the mechanisms underlying that diversity. Techniques like network analysis may provide the first steps toward uncovering those ecological mechanisms. Network analyses are powerful approaches for inferring correlations among the microorganisms within complex microbial communities (Layeghifard et al. 2017), which might relate to direct associations and interactions. These analyses have been particularly insightful in revealing cooccurrence patterns among microorganisms across a range of habitats (Faust et al. 2012, Williams et al. 2014). Such patterns of association are complex, driven by both abiotic, including preference for similar environmental conditions, and biotic factors like microbe–microbe interactions. Those interactions can be positive or negative, related, for example, to mutualism, competition, and predatory interactions, that likely contribute to shaping microbiome composition (Wright et al. 2013, Hassani et al. 2018). Associations of particular interest, though relatively poorly investigated in metazoan-associated microbiomes, are those between symbiotic bacteria and fungi. Bacterial–fungal associations (BFA) play an important role in microbial communities (Frey-Klett et al. 2011, Sam et al. 2017), and these associations could be potential drivers of diversity across host-associated microbiomes. For example, Tipton et al. (2018) demonstrated that microbiome networks are more robust to stressors when bacteria and fungi cooccur when compared to single-domain networks and that antagonistic interactions between bacteria can be offset when also cocultured with fungi.

While computer-generated models and simulations of networks serve as a valuable tool to explore the complex dynamics of microbial communities, it is also necessary to test the predictions of network analyses in biological systems to validate model inferences and to extend predictions about stability to real system properties like community composition. Some work has been done in this area using culturing techniques (Steinway et al. 2015, Das et al. 2018, Tipton et al. 2018), but no work has been done manipulating entire microbiomes in vivo and in situ based on network predictions within a host organism. Insects are an excellent model for this purpose because they have simpler microbiomes that are more tractable than those of other host systems (Muturi et al. 2017, Bennet et al. 2019).

Here, we use network analysis to inform in vivo host-level microbiome manipulations using a mosquito host, Aedes albopictus, as a model. We used a next generation sequencing dataset consisting of amplicon sequencing variants (ASVs) of the bacterial 16S rRNA gene and the fungal internal transcribed spacer (ITS) gene obtained from wild A. albopictus. We generated three networks from the sequencing data: a 16S-only network, an ITS-only network, and a cross-domain network consisting of 16S and ITS. We analysed each network’s attack robustness (Iyer et al. 2013), which is a measure of the network’s ability to remain connected when nodes are sequentially removed, as an indicator of network stability. We also performed a network keystone species analysis using betweenness centrality (the total number of shortest paths from node to node) and node degree (a measure of how many connections one node makes to another node) (Tipton et al. 2018), which are measures of node importance within a network. From the predictions of the network analyses, we hypothesized that the presence of fungi in the mosquito microbiome is an important driver of community assembly and structure of the bacterial component of the microbiome, and thus removing fungi and other eukaryotes would alter bacterial community assembly. We tested the hypothesis by (i) sourcing microbial communities from three different environments, (ii) manipulating the composition of those communities using stratified filtration regimes that sequentially removed larger cells (e.g. fungi and other eukaryotes) and particulates from aquatic substrate (similar to classic exclusion experiments in community ecology; Doherty and Sale 1986, Kempel et al. 2015), (iii) exposing mosquitoes to the microbial communities in the filtered substrates, and (iv) characterizing the mosquito microbiome across manipulations. Microbiome research often focuses only on the bacterial component of host-associated microbiomes and this work highlights why microbiome studies should, to the best extent possible, include fungal components as they can have major effects on the bacterial assembly within host-associated microbiomes.

Methods

Mosquito sample collection used for network analyses

Wild adult A. albopictus [n = 305, (F = 141; M = 164)] were collected in Mānoa Valley, Honolulu, Hawaiʻi using hand nets and handheld aspirators. The mosquitoes were transported from the field to the lab at 4°C. Upon arrival in the lab, they were sterilized with 75% ethanol for 1 min and rinsed twice in sterile 1X phosphate buffered saline (PBS). Midguts were immediately dissected and placed individually into a 1.5 ml centrifuge tube containing sterile 1X PBS. Midgut samples were stored at −80°C until DNA extraction. DNA was extracted and purified on a KingFisher Flex using a NucleoMag Tissue kit following the manufacturer’s protocol. SeqMatic (Fremont, CA, USA) sequenced the 16S library in a single run on an Illumina NovaSeq 6000 using a 250 bp paired-end NovaSeq 6000 SP Reagent Kit v1.5 500 cycles. The ITS library was sequenced in a single run on an Illumina MiSeq PE300 platform using a 250 bp paired end with a MiSeq reagent kit v3.

Bioinformatics of mosquitoes used for network analyses

Bioinformatics analysis was performed using the MetaFlow|mics pipeline for microbiome marker data (Arisdakessian et al. 2020). Reads were truncated at position 220 (190 for the reverse read) and were discarded if they contained at least one base below quality 20 (minimum quality to assign nucleotide identity with <1% error) or a number of expected errors above 3. Two ASVs were merged if all the three following conditions were satisfied: (i) they cooccur in every sample, (ii) one of the two ASVs has a lower relative abundance than the other in every sample, and (iii) they share a sequence similarity of at least 99%. All singletons (ASVs with reads in only one sample) and ASVs with no annotation at the kingdom level were removed. After quality control, sequencing resulted in 20.16 million 16S reads and 780 066 ITS reads. Further ASV processing was performed by merging the abundance data, taxonomic data, and sample data using the Phyloseq package (McMurdie and Holmes 2013) in R. 16S and ITS ASVs were included in the final analysis only if they appeared more than once in 97.5% of total samples.

Estimation of microbial association networks

We constructed our microbial association networks using the Sparse InversE Covariance estimation for Ecological Association and Statistical Inference (SPIEC-EASI) framework (Kurtz et al. 2015). Nodes in each network corresponded to ASVs and edges corresponded to inferred associations between ASVs. We ran SPIEC-EASI in neighborhood selection mode and model selection was performed based on the Stability Approach to Regularization Selection (Liu et al. 2010) using a variability threshold of 0.05%.

Evaluating and comparing networks

Networks were analysed using functions within the igraph R package version 1.3.5 (Csardi and Nepusz 2006). We evaluated node degree (the total edges a node has), betweenness centrality (the number of paths going through a node), and modularity (the number of subgraphs or communities within the network). We also measured attack robustness (the network’s ability to remain connected when nodes are removed) using the R package brainGraph version 3.0.0 (Watson 2020). We performed the robustness analyses by targeting nodes based on betweenness and node degree, which are integral to network stability.

Habitat design for in vivo validation of network analyses

Two large plastic buckets were filled with one tablespoon of field topsoil and one teaspoon of crushed leaf detritus collected from the University of Hawaiʻi at Mānoa (UH), 3 l of sterile water and one wooden shim (for oviposition). The buckets were placed outside, in a semishaded area, at UH and were left exposed to natural conditions for 7 days. The buckets were checked every 2 days to ensure water levels never fell below 2 l. The water level was restored to 3 l with sterile water when necessary. After 7 days, the water from the buckets was collected and returned to the laboratory for processing. Aedes albopictus larvae were present in the habitat when the water was collected. Two liters of water were also collected from the tanks of bromeliads (genus Neoregelia), at Lyon Arboretum (Lyon), which is ~4 miles north of UH. The bromeliad water was collected on the day of processing; no A. albopictus were found in it.

For both water sources, three inocula were prepared: a 30–50 μm filtration inoculum, a 10 μm filtration inoculum, and a 0.1 μm filtration inoculum. The 30–50 μm inoculum was prepared by filtering the collected water through a 30–50 μm coffee filter once. The 10 μm inoculum was prepared by filtering the collected water through a 30–50 μm coffee filter once, then filtering the filtrate through a 10 μm syringe-filter once. The 0.1 μm inoculum was prepared in the same way as the 10 μm inoculum followed by a single filtration through a 0.1 μm vacuum filtration cup. We chose these filtration sizes to allow for the microbial composition of the inoculum to vary by cell size. The 30–50 μm filter would allow larger eukaryotic cells, such as fungi and yeasts and bacterial cells to enter the filtrate. The 10 μm filter would prevent most eukaryotic cells from entering the filtrate but would allow some small yeasts and bacterial cells to pass. The 0.1 μm filter would prevent eukaryotic and bacterial cells from entering the filtrate but would allow metabolites and other molecules to pass through. An approximate total of 300 ml was obtained from each filtration level for both water sources. A control inoculum, which consisted of pure water from a Millipore Milli-Q water filtration system (Massachusetts, USA), was included to represent the larval habitat conditions mosquitoes experience in a typical laboratory environment.

Mesocosm design

Experimental mesocosms were created by pipetting 100 ml of a single inoculum type into a 150 mm × 15 mm Petri dish. There were seven treatment levels: Lyon 0.1 um, 10 um, 30–50 um, UH 0.1 um, 10 um, 30–50 um, and laboratory environment control. All treatments contained three replicates for a total of 21 mesocosm treatments. After pipetting each inoculum into the Petri dish, 100 A. albopictus eggs were placed into each mesocosm. The eggs were obtained from our lab colony that originated from wild A. albopictus collected from Mānoa Valley, Honolulu, Hawaiʻi and raised in the lab for 21 generations prior to the start of the experiment. To prevent environmental contamination from laboratory sources, each Petri dish was kept covered with the sterile Petri dish cover included from the manufacturer until food was provided, and all feeding events were conducted in a biosafety cabinet.

Mosquito rearing and handling

We reared A. albopictus with a 12 L:12 D photoperiod. Temperature in the light period was 28°C, whereas temperature in the dark period was 22°C. Humidity was kept constant at 60% for the entire photoperiod. A custom food preparation was made that consisted of a 1:1 ratio of liver powder to dried yeast extract mixed with pure water to a concentration of 18 mg per ml. To kill viable microorganisms, the food preparation was autoclaved for 30 min prior to being introduced into each mesocosm. Larval food water was administered daily at a concentration of 0.2 mg per larva per mesocosm. 8 and 10 days after the start of the experiment, L3–L4 larvae (n = 5) were removed from each mesocosm with a sterile pipette, placed in a 1.5 ml centrifuge tube and stored at −20°C until processed for microbiome analysis. Remaining larvae were monitored daily for pupation. Pupae were removed using a sterile pipette and transferred to a 50 ml sterile centrifuge tube. Pupae were housed individually after being collected from the mesocosm and were monitored daily for eclosion.

DNA extraction, amplification, and sequencing of mesocosm mosquitoes

Emerged adults were housed at 4°C for ~30 min for cold anaesthetization and transferred into 1.5 ml centrifuged tubes before being frozen at −20°C. Prior to DNA extraction, all adults (n = 168; 84 males, 84 females, 4 per mesocosm replicate) and larvae (n = 84; 4 per mesocosm replicate) that were used for microbiome analysis were subjected to a surface-sterilization protocol adapted from Muturi et al. (2018) consisting of one wash in 70% ethanol followed by two rinses in sterile 1X PBS. DNA was extracted and purified on a KingFisher Flex using a Qiagen MagAttract PowerSoil DNA KF Kit following the manufacturer’s protocol. The bacterial library preparation was done using a modified version of the Earth Microbiome Project 16S V4 protocol (Caporaso et al. 2011). Polymerase chain reaction (PCR) amplification targeted the V4 region of bacterial 16S rRNA using 515F and 806R primers from the Earth Microbiome Project. The ITS library preparation was also done using a modified version of the Earth Microbiome Project ITS Illumina Amplicon protocol (Caporaso et al. 2011). PCR amplification targeted the ITS region using ITS1f and ITS2 primers from the Earth Microbiome Project. Both libraries were checked for quality and quantity using a Bioanalyzer High Sensitivity chip (Agilent Technologies) and sequenced using Illumina MiSeq PE300 with a MiSeq Reagent Kit v3 (Illumina). Both 16S and ITS were sequenced in single runs; however, the single runs were performed separately for each library.

Bioinformatics of mesocosm mosquitoes

Bioinformatics analysis was performed with the same method used for the network analyses mosquitoes. Further ASV processing was performed by merging the abundance data, taxonomic data, and sample data using the Phyloseq package (McMurdie and Holmes 2013) in R. 16S ASVs were included only if they appeared more than once in 95% of total samples. A lower threshold was used to allow for more ASVs to be evaluated in the β-diversity analyses.

Community assembly analyses

To compare bacteriomes, β-diversity in the microbiome among individual mosquitoes was investigated among the mesocosms with different water filtration treatments. To assess β-diversity, changes in the composition of the bacteriome in response to the larval environment (control, 30–50 um, 10 um, and 0.1 um) were tested statistically with a generalized linear mixed model (GLMM) using the R package glmmTMB (Brooks et al. 2017) assuming a negative binomial error distribution, similar to the approach detailed by Sweeny et al. (2023). All models also included a formula that allows the dispersion parameter to vary among levels of a specified fixed effect, which included the filtration level, mosquito life stage, and individual ASV. All candidate models included an offset for sequencing reads per sample, fixed effects for mosquito life stage (larva, adult), filtration level, and a random intercept for ASV identity. Among specific models in the candidate set, the responses in relative abundance of every ASV to both mosquito life stage and filtration type were allowed to vary by specifying random interaction effects. The model grouping parameters and the hypotheses that generated them are defined in Table 1. The full set of candidate models are listed in Table S1. We used the Akaike information criterion (AIC) to select the most suitable model among the models included in this candidate set. To further explore which specific taxa are changing in response to filtration regimen using the conditional modes of random effect estimates similar to the approach used by Sweeny et al. (2023).

Table 1.

Definitions of model parameter groupings.

Model parameter	Grouping definition	Hypothesis
Treatment	Laboratory mesocosms were treated as individual replicates, replicates of the filtration mesocosms were grouped into a single group. Groupings included: Lab 1, Lab 2, Lab3, 0.1 μm filtration, 10 μm filtration, and 30–50 μm.	Each group has distinct environmental microbiomes
Treatment2	Laboratory mesocosms replicates were treated as a single group, filtration mesocosms replicates were treated as a single group by filtration size. Groupings included: Lab, 0.1 μm filtration, 10 μm filtration, and 30–50 μm.	Each group forms distinct microbiomes, but there is no variation across the lab environment
BacSource	Laboratory and 0.1 um filtration mesocosms were grouped together, 30–50 um filtration mesocosms were treated as a single group, and 10 um filtration treatments were treated as a single group. This model accounted for egg-associated bacteria effects.	Environmental microbiomes are distinct for 30–50 μm filtration, 10 μm filtration, but are the same for lab and 0.1 μm filtration
BacSource2	Laboratory and 0.1 um filtration mesocosms were grouped together, 30–50 um and 10 um filtration mesocosms were grouped together. This model accounted for egg-associated bacteria effects.	Environmental microbiomes the same for 30–50 μm filtration and 10 μm filtration, and are the same for lab and 0.1 μm filtration
BacSource3	Laboratory mesocosms were grouped together, 0.1 um filtration mesocosms were an individual group, 10 um and 30–50 um filtration mesocosms were grouped together. This model assumes that egg-associated bacteria are not impacting the mesocosms.	Environmental microbiomes are the same in 30–50 μm filtrations and 10 μm filtrations, but are distinct in lab and 0.1 μm filtrations
Substrate	Water is sourced from the laboratory or from the wild (bromeliads)	Environmental microbiomes are the same across all environments, but distinct from the lab

Estimation of Ascogregarina taiwanensis load

The presence of Ascogregarina taiwanensis was tested to rule out alternate explanations for differences in mosquito microbiomes following the methods of Seabourn et al. (2020). This parasite is known to influence the A. albopictus microbiome (Seabourn et al. 2020).

Results

Network analyses

The 16S single-domain network consisted of one main connected component [76 out of 84 ASVs (90.48%)] and eight singleton 16S ASVs unconnected to the main network. The average node degree of the network was 3.69 ± 4.44. The average betweenness centrality of the network was 122.06 ± 272.67. Based on their connectivity pattern, we clustered the 16S ASVs using modularity optimization (Clauset et al. 2004), resulting in 12 modules with a modularity score of 0.46. The fungal ITS single-domain network consisted of one main connected component [246 out of 276 ASVs (89.13%)], one dyad, and 28 singleton ASVs. The average node degree of the network was 2.96 ± 2.40. The average betweenness centrality was 446.97 ± 766.31. The fungal-only network consisted of 40 modules with a modularity score of 0.69. The 16S-ITS cross-domain network consisted of one main connected component [515 ASVs out of 518 ASVs (99.42%)] and three singleton fungal ASVs without connections to the main network (Fig. S2). The average node degree of the network was 6.52 ± 3.14. The average betweenness centrality was 702.03 ± 672.71. The cross-domain network consisted of 12 modules with a modularity score of 0.48.

Attack robustness is a measure of the network’s overall ability to withstand the failure of its constituent parts (Iyer et al. 2013) and can be measured by sequentially removing nodes within the network and measuring the size of the largest remaining component after each removal (Ruiz et al. 2017). Following the methods of Tipton et al. (2018), nodes were removed in order of decreasing betweenness centrality (Fig. 1A) and in order of decreasing node degree (Fig. 1B). In each measure, the attack robustness of the cross-domain network was considerably increased compared to those of the 16S-only network and the ITS-only network. Cross-domain networks can inherently be more robust because of increased species richness, so to ensure that the observed robustness patterns were not an effect of increased species richness, we down-sampled the richness so that each network contained an equal numbers of nodes (represented by a unique ASV). The down-sampled robustness analysis showed the same pattern as both the cross-domain and single-domain networks, with the cross-domain network being the most robust. This suggests that cross-domain associations do have a real effect on network stability.

Figure 1.

Robustness curves for all networks by (A) betweenness centrality and (B) node degree. Each network is expressed as a line on the graph. A more robust network is characterized by having a larger area under the curve.

The keystone species analysis of the cross-domain network revealed that fungal ASVs represent a higher proportion of important nodes with the greatest centrality values (Fig. 2). High betweenness centrality indicates important connector taxa within the network and a high node degree indicates those taxa that are important hubs within the network. Together, both metrics highlight the importance of those taxa to the overall structure of the network. Biologically, those important taxa could represent organisms that have substantial influence on community structure and assembly akin to the classical ecological concept of keystone species (Jordán 2009, Hale and Koprowski 2018)

Figure 2.

Keystone species analysis. Potential fungal and bacterial keystone ASVs were identified by comparing betweenness centrality and node degree. Nodes with high betweenness centrality represent taxa important to the connectivity of the network. Nodes with high node degree represent taxa that are hubs in the network. Taxa are identified at the domain level.

Sequencing data summary of in vivo microbiome manipulations

A total of 10 655 249 16S reads passed through quality filtering and were allocated among 1959 ASVs. For the Lyon dataset, 530 16S ASVs were prevalent in >5% across samples and were included in the subsequent analysis of the microbiome. For the UH dataset, 449 16S ASVs were prevalent in >5% across samples and were included in the subsequent analysis of the microbiome. Among the top 20 16S ASVs, there were three phyla (Proteobacteria, Actinobacteriota, and Bacteroidota) represented by 16 genera (Fig. S1). Wolbachia, a common symbiont of A. albopictus, was removed from the analysis because it is not environmentally acquired. A total of 960 796 ITS reads passed through quality filtering and were allocated among 234 ASVs. Of these 234 ASVs, the top 20 accounted for 98.99% of reads; there were two phyla (Ascomycota and Basidiomycota) represented by 13 genera. Notably, there were 27 ITS ASVs that were unique to the Lyon 30–50 μm filtrations (11.54% of total ASVs) and 13 ITS ASVs that were unique to the UH 30–50 μm filtrations (5.56% of total ASVs).

In vivo microbiome analyses

For the treatments derived from Lyon, the model that best fitted the data (AICc weight = 1; Table 1) included fixed effects of (i) mosquito life stage (larva and adult), (ii) microbial source (natural or lab environment), (iii) an interaction between mosquito life stage and microbial source, and random effects of (iv) ASV identity, (v) mesocosm identity, (vi) interaction between ASV identity and mosquito life stage, (vii) interaction between ASV identity and mesocosm identity, (viii) interaction between ASV identity and microbial source, and (ix) three-way interaction between ASV identity, mosquito life stage, and microbial source. In this model, the microbial source variable was defined by grouping the filtration treatments into four distinct groups: lab (Millipore water), environmental water filtered at 0.1 μm, environmental water filtered at 10 μm, and environmental water filtered at 30–50 μm (Fig. 3A). See Table S1 for the full set of candidate models and Table S2 for their AICc results. For the treatments derived from UH, the model that best fitted the data (AICc weight = 1.0; Table 2) was identical to the best-fit model from the Lyon treatments (Fig. 3B). See Table S3 for their AICc results. Using the conditional modes of the random effects estimates, taxa specific differential abundances showed strong ASV shifts between the 30–50 μm and 10 μm filtration levels (Fig. 4). For example, ASVs assigned to Actinobacteriota and Verrucomicrobiota were enriched in the 30–50 μm filtration compared to the 10 μm filtration in larvae in the UH group. ASVs assigned to Bacteroidota were enriched in the 30–50 μm filtration compared to the 10 μm filtration in larvae in the Lyon group. We also saw similar patterns of differential abundance in adults from both locations. Most notably, ASVs assigned to Bacteroidota were enriched adults in the UH 30–50 μm filtrations, but in the Lyon adults Bacteroidota were enriched in the 10 μm filtrations, suggesting that microbiome assembly is influenced by the presence of eukaryotic microorganisms.

Figure 3.

Nonmetric multidimensional scaling plots demonstrating the composition of the bacteriome that assembled A. albopictus mosquitoes by filtration level and mosquito life stage in the (A) Lyon treatments and (B) UH treatments. Distances are based on the Morisita–Horn index.

Table 2.

Top performing GLMMs that partitioned variance in the composition of the bacteriome among different hypothetical groupings of the Site 1 and Site 2 filtration regimens.

Fixed effects	Random effects	Hypothesis
Offset(log(sum)), mosquito-stage, treatment2, mosquito-stage*treatment2	ASV, mesocosm, ASV*mosquito-stage, ASV*mesocosm, ASV*treatment, ASV*mosquito-stage*treatment2	Mosquito bacteriome assembly is modulated by the inclusion of large microbial cells in the inoculate. Treatment2 groups filtrations into individual groups: lab, 0.1 μm, 10 μm, and 30–50 μm.
Offset(log(sum)), mosquito-stage, BacSource3, mosquito-stage*BacSource3	ASV, mesocosm, ASV*BacSource3, ASV*mesocosm, ASV*BacSource3, ASV*mosquito-stage*BacSource3	Mosquito bacteriome assembly is not modulated by the inclusion of large microbial cells in the environment but is modulated by bacteria source (natural environment versus lab environment).
Offset(log(sum)), mosquito-stage, treatment, mosquito-stage*treatment	ASV, mesocosm, ASV*mosquito-stage, ASV*mesocosm, ASV*treatment, ASV*mosquito-stage*treatment	Mosquito bacteriome assembly is modulated by the inclusion of larger microbial cells in the inoculate. Treatment groups filtrations into individual modules: lab (replicate 1), lab (replicate 2), lab (replicate 3), 0.1 μm, 10 μm, and 30–50 μm.

Figure 4.

Differential abundances of ASVs comparing 30–50 μm to 10 μm filtrations using the conditional modes of the random effect for the three way interaction (ASV identity × mosquito life stage × microbial source) from the best-fitted GLMMs in (A) larvae from UH, (B) adults from UH, (C) larvae from Lyon, and (D) adults from Lyon. Samples with positive values indicating higher relative taxonomic abundance in the 30–50 μm filtrations. Samples with negative values indicate higher relative taxonomic abundance in the 10 μm filtrations. Each point represents an individual ASV, and each plot shows which ASVs are driving differences in beta diversity across filtrations.

Presence of A. taiwanensis

None of the 232 A. albopictus samples tested positive for the presence of A. taiwanensis.

Discussion

Network construction is a powerful computational tool for analysing ecological associations. Networks provide a more integrated picture of ecological complexity and allow for a holistic analysis of the associations and interactions within communities (Fath et al. 2007). Based on sequences of bacterial 16S rRNA and fungal ITS genes, we inferred single-domain and cross-domain microbial-association networks of the A. albopictus microbiome. The inferred cross-domain network had higher connectivity than both single-domain networks. However, the cross-domain network showed higher modularity than the 16S single-domain network, but lower modularity than the ITS single-domain network. The cross-domain network also showed higher attack robustness than both single-domain networks.

Several studies have shown that higher robustness has a stabilizing effect on the network when the network is under attack (Mahana et al. 2016, Ruiz et al. 2017, Lee et al. 2022). Since our cross-domain network showed the highest robustness, we then investigated which components of the network were driving the higher robustness. We performed a keystone species analysis by looking at the distribution of nodes (ASVs) in the network and found that the nodes with the highest centrality measures primarily consisted of fungal ASVs, indicating that fungi may highly influence the structure and assembly of the A. albopictus microbiome (Fig. 2).

Based on the results of the network and keystone species analyses, we hypothesized that fungi have a prominent influence on the assembly of the A. albopictus microbiome. Therefore, we designed our filtration regimens to alter the compositionality of the microbial community in our mesocosm inocula based on microbial cell size. Our hypothesis was supported for the mesocosms derived from both Lyon and UH. Our best-fit model for the Lyon treatments (Table 2) suggested that the presence of large cells (i.e. fungi and other eukaryotes) is an important driver of bacteriome assembly in mosquitoes. We saw 27 unique fungal ASVs in our 30–50 μm filtration treatment from this environment that were not present in the other filtration treatments. However, it is likely that fungal identity is more important than fungal presence per se. For example, ASV 345 was identified as belonging to Penicillium, a large genus from which the antibacterial penicillin was derived. It is possible that this fungal ASV could be influencing the bacterial community via secretion of penicillin, which would suppress susceptible bacteria. Another example, ASVs 174 and 215, belonging to Komagataella, have been reported as having potential probiotic effects (Birman et al 2023) that may contribute microbiome stability by promoting growth of beneficial bacteria. Further investigation into the mechanisms of fungi-mediated (especially using fungi with a functionally validated capacity to exert those effects) microbiome assembly is an area for future study.

For the UH treatments, our best-fit model also suggested that bacterial community assembly was impacted when fungi are present in the environment. There were 13 ITS ASVs present only in the 30–50 μm filtrations. We also found two ASVs (160 and 133) belonging to Komagataella, which, again, may have probiotic properties that influence bacterial community structure. We also identified another potential groups of ASVs (205 and 347) belonging to Candida that could potentially impact bacterial community assembly. For example, Candida albicans can alter oral microbiome community structure by promoting the growth of some bacteria over others (Janus et al 2017, Di Martino et al 2022). Our findings underscore the complex nature of BFA, their potential role in bacterial community assembly, and highlight the need for future studies to focus on elucidating the interactions mechanisms underlying BFA.

Another potential driver of microbiome assembly in A. albopictus is infection with the apicomplexan parasite A. taiwanensis. This parasite is present in A. albopictus populations and infection with A. taiwanenis is associated with microbiome diversity in A. albopictus hosts (Seabourn et al. 2020). Since all of our mosquito samples were negative for A. taiwanensis infection, we conclude that the patterns of bacterial assembly we saw in our results were not caused by this eukaryotic parasite.

Both best-fit models also showed a statistically significant effect on bacterial assembly of the interaction between mosquito life stage and filtration. Aedes albopictus larvae were distinct from adults in the Lyon and UH 30–50 μm filtration, 10 μm filtration, 0.1 μm filtration, and lab groupings. It is unlikely that the difference between the 30–50 μm and 10 μm filtrations was because of different bacterial source pools as all bacterial ASVs tested were found in both filtrates. It is also unlikely that the difference between the laboratory mesocosm and the 0.1 μm filtration is because of different environmental bacterial source pools. Since the eggs used in this experiment were not sterilized, and the 0.1 μm filtration would have prevented viable cells in the source pool from entering the filtrate used to inoculate the mesocosm, those two treatments probably had a similar source pool of viable bacteria at the onset of the experiment. The assembly of the mosquito bacteriome could have been influenced by any metabolites that passed through the 0.1 μm filter into the inoculum as secondary metabolite cross-feeding can promote taxonomic diversity within microbial communities (Embree et al. 2015, Douglas 2020). Alternatively, the distinctiveness of the 0.1 μm filtration group could be due to sequencing of 16S rRNA fragments present in the mesocosm postfiltration that were ingested by larvae during foraging. Lastly, both scenarios could also occur simultaneously; thus, a future experiment eliminating environmental DNA from the 0.1 μm inoculum while leaving secondary metabolites, verified through metabolomics, could clarify the role of secondary metabolites on bacteriome assembly in A. albopictus.

It is unsurprising that larvae and adults would have different bacteriomes because microorganisms were introduced during the larval stage only. Thus, larvae were exposed to a greater diversity of microbes than the adults. Mosquito larvae also evacuate their gut contents shortly before pupating, which eliminates a substantial proportion of the microbial commmunity prior to becoming adults (Duguma et al. 2015). Furthermore, the larval midgut undergoes extensive remodeling during metamorphosis (Fernandes et al. 2014), which impacts bacteria present in the gut at eclosion (Moll et al. 2001). However, this is an area that needs research because no study has explored how the remodeling process mechanistically affects microbiome composition in mosquitoes.

Host-associated microbiomes exhibit considerable diversity within and among hosts. Yet, the origin of this diversity is not well understood. To the best of our knowledge, this is the first study to pair in silico modeling of microbial associations with in vivo community-level manipulations of whole environmental and host-associated microbiomes to describe how microbial interactions contribute to microbiome diversity. The results from our in vivo experiments imply that interdomain interactions could be an important mechanism shaping microbiome diversity in A. albopictus. These results demonstrate that the often overlooked components of microbial communities, such as fungi, can have substantial ecological roles in host-associated microbiomes. Thus, future microbiome studies, not only in mosquitoes but other systems, should account for, and include, fungi and other eukaryotic microorganisms to the best extent possible.

Conflict of interest

None declared.

Funding

Research reported in this publication was supported by the National Institute of General Medical Sciences of the National Institute of Health under Award Number P20GM125508. This center grant funded the Medeiros laboratory directly, the University of Hawaiʻi at Mānoa Microbial Genomics and Analytical Laboratory, and the Insectary for Scientific Training and Advances in Research. This content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Health.

Data availability

The obtained raw sequences were submitted to the National Center for Biotechnology Information (NCBI) as Sequence Read Archive (SRA) under accession numbers PRJNA1266722 for the 16S raw sequences and for PRJNA1269049 ITS raw sequences.