Historical mosquito colonization dynamics are associated with patterns of microbial community assembly in aboveground aquatic habitats

Abstract

Mosquito larvae develop in aquatic habitats that harbor highly variable communities of bacteria and other microorganisms, which have been well demonstrated to shape individual fitness outcomes in laboratory settings. However, relatively little is known about how this microbial variation contributes to or is influenced by mosquito population dynamics in the field. To investigate potential associations between mosquito population dynamics and microbial community assembly, we characterized bacterial communities in naturally occurring larval habitats with variable historical mosquito productivity using amplicon sequencing. We then applied a null model approach to quantify the relative importance of selection, dispersal, and drift processes in bacterial community assembly. Habitat microbiota clustered into two distinct biotypes: Biotype 1 communities were dominated by Proteobacteria, while Biotype 2 communities were dominated by Cyanobacteria. Both biotypes were shaped by a combination of selection and neutral (i.e., dispersal and drift) processes. However, selection played a more prominent role in habitats with Biotype 1 communities, whereas drift was more influential in Biotype 2 habitats. Variation partitioning further identified historical mosquito productivity and the spatial aggregation of sites with similar productivity histories as key drivers of selection. These findings suggest that mosquito population dynamics are associated with differences in microbial community structure, potentially through feedbacks between mosquito activity and habitat conditions. This study lays the foundation for future work to disentangle causal relationships and to integrate patterns of microbiota diversity and mosquito occurrence into vectorial capacity models for improved prediction of mosquito-borne disease transmission dynamics in the field.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40793-025-00831-1.

Introduction

Freshwater ecosystems are vital to biodiversity and human well-being, serving as habitats for a myriad of organisms, including diverse microbial communities [1, 2]. These microbial assemblages play crucial roles in nutrient cycling, water quality, and ecosystem health, influencing the fitness of both aquatic invertebrate and vertebrate species [3]. However, the composition and function of these communities are not solely determined by contemporary environmental conditions. Historical events, such as colonization by specific invertebrate or vertebrate species, can leave lasting impacts on microbial assemblages, a phenomenon known as legacy effects or historical contingencies [4, 5]. Despite the recognized importance of legacy effects in ecology, there remains a gap in our knowledge regarding their specific influence on microbial assemblages in freshwater habitats.

Aquatic invertebrates, such as insects, crustaceans, and mollusks, play a pivotal role in shaping freshwater microbial communities [6]. These organisms contribute to nutrient cycling by breaking down organic matter, which in turn provides a rich substrate for microbial growth [6]. For example, the feeding activities of detritivorous invertebrates, like certain insect larvae and amphipods, fragment organic material, increasing its surface area and making it more accessible to microbial colonization [7]. Additionally, the excretion of waste products by these invertebrates releases nutrients such as nitrogen and phosphorus into the water, further stimulating microbial activity [8, 9]. Predatory invertebrates also influence microbial populations directly via grazing (i.e., bacterivory) or indirectly by controlling the abundance of other invertebrates that graze on microbial biofilms [6]. However, while it is well-established that aquatic invertebrates play a crucial role in shaping microbial assemblages through their feeding activities and nutrient cycling, the long-term effects of past population changes on present-day microbial communities remain underexplored.

Mosquito larvae represent a particularly abundant group of detritivorous invertebrate organisms within standing freshwater ecosystems [10], where they feed on decaying organic matter and environmental microbes [11]. These feeding activities can significantly impact microbial diversity and alter the composition of microbial communities, at least over short time scales [12, 13]. Microbes present in the water column and ingested by mosquito larvae also play a crucial role in their development and fitness [14–18]. Certain microbial taxa provide essential nutrients and metabolic functions that enhance larval growth, survival, and ultimately, the competence of adult females of some mosquito species to transmit human pathogens [15, 17–23]. However, the long-term impacts of mosquito population dynamics on the distribution of these and other microbial taxa across aquatic habitats with variable larval colonization histories have yet to be investigated.

Here, we used high throughput amplicon sequencing to characterize bacterial communities in continuously monitored aboveground aquatic habitats with different historical mosquito population dynamics. A null model approach was then employed to quantify the relative importance of selection, dispersal, and drift processes in bacterial community assembly. Finally, variation partitioning was used to investigate the relative contribution of different historical mosquito productivity measures, environmental indices, and spatial factors in driving observed patterns in contemporary bacterial communities.

Results

Aquatic habitats vary in historical mosquito productivity

Since 2007, Public Health Madison & Dane County (PHMDC) has monitored and controlled the breeding activity of Culex pipiens, a primary vector of West Nile Virus, and other mosquito species throughout Dane County, WI, USA. Potential oviposition (i.e., egg-laying) sites are monitored from late May to September to track sites producing large numbers of larvae, with sites exhibiting moderate to high densities of Culex larvae being flagged for treatment with a larvicide comprised of spores and other derivatives of the soil bacterium Bacillus thuringiensis israelensis (Bti), which kills mosquito larvae and larvae of other closely related flies (Diptera: Nematocera) via the production of crystalline proteins that disrupt digestion [24]. To assess variation in mosquito productivity across naturally occurring larval habitats, we analyzed monitoring records from 877 oviposition sites collected by the PHMDC between 2007 and 2018. The dataset included site observations and larval dip counts for each site, though sampling was not uniform across sites or years. Some sites were visited more frequently or consistently than others, reflecting local public health priorities and resource constraints. A preliminary exploration of this dataset revealed three important observations: (i) sites varied in the frequency at which larvae were detected, with the majority of sites never containing larvae and only a few sites reporting the presence of larvae every year (Fig. S1A); (ii) sites varied in larval density, with sites in which larvae were more frequently detected producing greater numbers of larvae (Fig. S1B); and (iii) sites varied in larval diversity, with most sites on average reporting the presence of larvae of only one or two mosquito species across the entire monitoring period (Fig. S1C).

Habitat microbiota group into two clusters

Of the 877 PHMDC sites, 86 sites across the full range of high to low historical mosquito productivity levels were selected for sampling in June and July 2019 (Fig. 1A). Productivity levels were defined based on two criteria: the frequency at which larvae were detected across all site visits during the monitoring period (range: 0–100%), and the average larval density (range: 0–20 larvae per dip). Sites were chosen to ensure representation across this gradient and were limited to those with complete data available for the entire monitoring period. At each site, 50–200 ml of water was collected in sterile conical tubes and immediately transported to the laboratory on ice, where it was centrifuged at maximum speed for 20 min to form a cell pellet for DNA isolation. Bacterial density in water samples was also estimated by counts of colony forming units (CFUs) on LB agar plates using an aliquot of each sample prior to centrifugation.

Fig. 1.

Microbiota diversity in aquatic habitats with variable historical mosquito productivity. A Locations of the collection sites in Dane County, WI USA. Dots in the right panel are color-coded by biotype classification (Biotype 1, blue; Biotype 2, orange). B Relative abundance of bacterial phyla in water from 86 naturally occurring larval habitats in the Madison, WI, USA area. Each bar presents the proportion of sequencing reads assigned to a given phylum. Only categories > 1% are presented. C Assortment of habitat microbiota into biotypes. Shown are Between-Class Analysis (BCA) visualizations of biotypes (ellipses, color-coded by biotype as in panel A) as identified by Partitioning Around Medoids (PAM) clustering, with black dots representing abundance distributions of bacterial phyla from an individual site and numbered white rectangles marking the center of each biotype. Bacterial phyla overrepresented in the corresponding biotypes are listed. D Relative abundances of the six bacterial phyla identified by BCA as principally contributing to the separation of habitat microbiota biotypes. Shown are means, ranges and first and third quartiles. Color coding of biotypes follows that in panels A and C

High throughput sequencing of 16S rRNA gene amplicons from all water samples generated a total of 1,733,696 sequences (median = 14,344 per sample) that were assigned to 19,752 unique Amplicon Sequence Variants (ASVs) after quality control filtering. A total of 62 bacterial phyla were identified across all samples, but seven accounted for ~ 95% of sequences: Proteobacteria (57%), Cyanobacteria (13%), Bacteroidetes (10%), Firmicutes (5%), Actinobacteria (4%), Verrucomicrobia (4%), and Planctomycetes (2%) (Fig. 1B). At the ASV level, sites exhibited substantial variation in the total number of unique ASVs detected in a given sample and their relative abundance. Individual ASVs also exhibited substantial variation in frequency as measured by the total number of samples in which they were detected and their relative abundance in each sample, which ranged from 1 to 76 and ~ 0.00094 to 79%, respectively. Across all ASVs detected in a given sample, the vast majority were rare. In ~ 70% of the samples we sequenced (60 out of 86), > 50% of detected ASVs were present at < 0.1% relative abundance. These rare ASVs accounted for up to ~ 68% of the total number of sequences within a given sample. In contrast, dominant ASVs (i.e., those present at > 1% relative abundance) comprised only ~ 0.085 to 34% of the total number of ASVs detected per sample, but accounted for up to ~ 79% of the total sequences in the same sample.

Partitioning Around Medoids (PAM) analysis revealed two distinct clusters, which we designated as “Biotype 1” and “Biotype 2” (Fig. 1C). These biotypes were identifiable by variation in the relative abundance of six of the seven phyla to which the majority of our sequences were assigned (Actinobacteria, Cyanobacteria, Firmicutes, Planctomycetes, Proteobacteria, Verrucomicrobia) (Fig. 1D) and sample grouping by biotype explained up to ~ 36% of variation in microbiota composition among sites (Fig. S2). The enrichment in the relative abundance of Proteobacteria and Firmicutes in Biotype 1 sites was further accompanied by an increase in both proteobacterial and firmicutal ASV diversity, while the enrichment in the relative abundance of Actinobacteria, Cyanobacteria, and Planctomycetes in Biotype 2 sites was accompanied by a corresponding increase in actinobacterial, cyanobacterial, and planctomycetal ASV diversity (Fig. S3A). Total ASV richness and culturable bacterial densities were also overall higher in Biotype 1 sites than in Biotype 2 sites (Fig. S3A, B).

We also quantified the relative importance of different community assembly processes in shaping microbiota composition in Biotype 1 versus Biotype 2 sites using the null model approach described by Stegen et al. [25]. For both Biotype 1 and Biotype 2 sites, community structure was shaped primarily by neutral processes, such as stochastic dispersal, ecological drift, and random colonization events (Fig. 2). However, communities in Biotype 1 sites exhibited significantly lower Beta Nearest Taxon Index (βNTI) values—an indicator of phylogenetic turnover between communities—than those in Biotype 2 sites (Fig. 2A). This suggests that Biotype 1 communities were more consistently structured by homogenous selection, meaning that similar environmental conditions or biological pressures (e.g., high mosquito productivity) repeatedly favored closely related taxa across sites, leading to reduced phylogenetic variability (Fig. 2B). Biotype 1 communities were also more strongly shaped by selection—both homogenous and variable—than Biotype 2 communities (Fig. 2B). In contrast to homogenous selection, which reflects consistent pressures favoring similar taxa across sites, variable selection arises from differing local conditions that promote distinct community compositions.

Fig. 2.

Community assembly processes in habitat microbiota biotypes. In both panels, shades of blue represent selection processes while shades of red represent neutral processes. A βNTI distributions for Biotype 1 and Biotype 2 sites. Phylogenetic turnover that is less than null expectations (i.e., βNTI < −2) indicates homogenous selection, phylogenetic turnover that is greater than null expectations (i.e., βNTI > 2) indicates variable selection, and phylogenetic turnover that does not vary from null expectations (|βNTI| < 2) indicates neutral processes. B Proportions of community assembly processes between Biotype 1 and Biotype 2 sites. Asterisks represent statistically greater proportions (Z-tests) at the following significance levels: *p < 0.05, **p < 0.01, ***p < 0.001. Selection overall (i.e., homogenous + variable) was greater in Biotype 1 communities than in Biotype 2 communities (p < 0.0001)

Historical mosquito productivity associates with microbiota assembly

We finally investigated the contribution of different mosquito productivity measures, environmental indices, and spatial factors to microbial community assembly. For each site, we estimated the average frequency and density of larvae detected over the entire PHMDC monitoring period leading up to microbiota sampling and sequencing (2007–2018). The total number of times a given site was treated was also estimated in order to test the effect of Bti on associated communities. We used MODIS satellite data to estimate the following environmental indices for each site averaged over the same monitoring period: (i) Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), day- and nighttime Land Surface Temperatures (LST_day, LST_night), Gross Primary Productivity (GPP), and Net Primary Production (NPP). NDVI and EVI are well-established proxies for forage/plant biomass [26], while GPP and NPP measure the total amount of organic carbon fixed by plants in a given area through photosynthesis and the energy left over for plant growth and/or consumption by detritivores and herbivores, respectively [27]. Principal Coordinate of Neighbor Matrices (PCNM) were used to quantify the spatial relationships of microbiota sampled between sites.

Variation partitioning identified spatial structure and specific productivity measures as key drivers of selection across sites (Table 1). Productivity measures accounted for a relatively greater proportion of the variation in βNTI than spatial factors alone (Table 1), suggesting that differences in productivity and the aggregation of sites with similar productivity histories may have played a more prominent role in driving the selective pressures observed in Biotype 1 communities compared to Biotype 2 communities. Indeed, the frequencies of Aedes spp. larvae were overall higher in Biotype 1 sites than in Biotype 2 sites (Fig. S4). Biotype 1 sites were also historically colonized by overall higher densities of Culex spp. larvae that resulted in them being treated with Bti more often over the monitoring period (Fig. S4). In contrast, Biotype 2 sites were historically colonized by higher densities of Anopheles spp., which were rarely detected in sites harboring Biotype 1 communities (Fig. S4).

Table 1.

Partitions of variation in βNTI accounted for by mosquito productivity measures and spatial factors

Partition	Adj. R2	p-value	Significant variables
Prod	0.23	0.001	TotTreat, PercAeFnd, AvgAnDens, AvgCxDens
Space	0.14	0.001	PCNM3, PCNM11
Prod + Space	0.31	0.001	TotTreat, PercAeFnd, AvgAnDens, AvgCxDens, PCNM33
Prod | Space	0.18	0.001	TotTreat, AvgAnDens, AvgCxDens
Space | Prod	0.09	0.003	PCNM4, PCNM11, PCNM33
Prod ∩ Space	0.05	–
Residuals	0.69	–

The significance of each partition was tested using distance-based redundancy analysis (dbRDA). Prod + Space represents the total variation explained by both components; Prod | Space (Space | Prod) represents the marginal fraction of variation explained by each component after controlling for the other; Prod ∩ Space represents the fraction of explained variation shared between both components (note that the significance of Prod ∩ Space cannot be tested). The significance of individual variables within each variation partition was determined using permutation tests (anova.cca function, "vegan" package) following dbRDA. For productivity measures, abbreviations are as follows: TotTreat, total number of times Bti was administered to a given site over the entire monitoring period; PercAeFnd, percent total site visits with Aedes spp. larvae present; AvgAnDens and AvgCxDens, total number of Anopheles or Culex spp. larvae, respectively, detected in a single dip across all site visits. Spatial factors (denoted as “Space” in the table) are represented by Principal Coordinates of Neighborhood Matrix (PCNM) scores generated using the geographic coordinates of each site, with low-order PCNM vectors representing large-scale spatial associations between sites and higher-order vectors representing more fine-scale spatial associations. A total of 35 PCNM vectors were generated

Discussion

In the study area, mosquito larvae were uncommon among sampled habitats, with highest densities in most frequently occupied habitats. Prior surveys of larval habitats have reported incidences up to 83% [28–30], with different habitat types, such as canals and concrete tanks, differing in frequency of larval presence [31]. The habitat types in the present study may have higher representation of low-usage habitat types compared to studies targeting confirmed larval habitats, since these sites encompass the full range of drainage structures identified with maps of surface water inventory and stormwater runoff control structures [32]. Co-occurrence of different mosquito species was low, with fewer than two species detected at the vast majority of sites at any given time. This contrasts with findings from a study in the peri-Iquitos region of Amazonian Peru, where two or more species were present at 70% of sites [33]. Comparatively low species overlap may be a product of low habitat occupancy, with denser and more frequently used habitats more likely to contain multiple species.

Bacterial communities in the study area were predominantly comprised of taxa that have been reported from other mosquito larval habitats, including Proteobacteria, Cyanobacteria, Bacteroidetes, Firmicutes, Actinobacteria, and Verrucomicrobia [14]. ASV diversity levels per sample were also comparable to those previously found in water containers [34]. Although ASV richness is not directly comparable with OTU richness, which the majority of existing mosquito microbiota surveys report, diversity at higher taxonomic levels may be compared. We find that habitat microbiota are highly variable between sites, consistent with previous surveys which have found most OTUs or genera present at less than 1% relative abundance [35, 36] and very few OTUs present in all sites [37]. However, the sites in the present study recovered much higher diversity at higher taxonomic levels than previous studies, with 62 phyla recovered in contrast with the 9–15 phyla from previous studies with comparable or greater numbers of samples profiled [35–37]. This elevated diversity likely reflects deeper sequencing in our study, which enabled detection of rare taxa that may have been missed in previous surveys. Ecological factors, such as variation in habitat disturbance and bacterial residence time, may also contribute to the accumulation of taxonomic diversity [38]. Despite the overall sequencing depth, our dataset exhibited substantial variation in coverage and saturation across samples. To ensure comparability, we rarefied the data to a threshold below saturation for some samples, striking a balance between sequencing depth and sample retention. Although deeper rarefaction can improve detection of rare taxa, it risks excluding low-read samples that may capture ecologically meaningful variation—particularly in environmental datasets where biomass and DNA yield vary widely. To address this trade-off, our analyses focused on dominant taxa and excluded low-abundance ASVs, thereby reducing concerns about underrepresentation of rare taxa in diversity estimates. When all samples are rarefied to the same depth, comparisons of alpha and beta diversity remain valid and interpretable, even if richness is somewhat underestimated. Consequently, the observed variability in community composition and diversity across the sites we sampled likely reflects true ecological differences rather than artifacts of sequencing depth.

Clustering of habitat microbiota in sites with variable mosquito productivity histories identified two distinct biotypes, with Biotype 1 communities being dominated by members of the Proteobacteria and Biotype 2 communities being dominated by members of the Cyanobacteria. Null modeling of assembly processes further demonstrated that while both Biotype 1 and Biotype 2 communities were predominately structured by dispersal limitation, selective processes were overall more important in structuring Biotype 1 communities than Biotype 2 communities, which were more structured by homogenizing dispersal and drift. Dispersal, or the movement and establishment of organisms in space, is well known to be an important regulator of microbial community assembly [39, 40], and the observation that dispersal limitation was the dominant process shaping both Biotype 1 and Biotype 2 communities is consistent with work in soils and other aquatic systems demonstrating the dominance of dispersal limitation in shaping the random distribution of bacteria at local scales [41–43]. That drift (i.e., stochastic changes attributable to reproduction and death) was more important in structuring Biotype 2 communities than Biotype 1 communities is also consistent with expectations of small populations being more vulnerable to dispersal limitation and thus subject to drift and the observation that Biotype 1 communities exhibited overall higher levels of ASV diversity and culturable cell density than Biotype 2 communities [44, 45]. The greater influence of homogenizing dispersal (i.e., heightened transport between sites) in Biotype 2 communities may be attributed to greater age and stability of these sites and the homogenization of environmental conditions that often accompanies the passive migration of aquatic communities such as in lakes, which are also dominated by Cyanobacteria [46].

Variation partitioning identified drivers of selection to be specific mosquito productivity measures and spatial factors, indicating that changes in mosquito population dynamics and spatial aggregation of sites with similar mosquito productivity histories accounted for the greater influences of both homogenous and variable selection we observed in Biotype 1 communities as compared to Biotype 2 communities. Indeed, sites harboring Biotype 1 communities exhibited overall higher productivity measures than Biotype 2 communities. The greater influence of homogenous selection on Biotype 1 communities could reflect consistent bacterial responses to environmental variables associated with the presence and/or abundance of larvae, which are well known to shape water pH, nutrient levels, and bacterial metabolism [13], as well as management interventions such as treatment with Bti. Notably, Biotype 1 sites were treated with Bti more frequently over the monitoring period, suggesting that Bti could act as an additional environmental filter by altering microbial communities directly or indirectly through its effects on mosquito population dynamics. The greater influence of variable selection suggests the existence of distinct bacterial niches in Biotype 1 sites but not Biotype 2 sites, which could be the result of microscale spatial heterogeneity [47] or the mosquitoes themselves. As detritivores, mosquito larvae ingest bacteria and other microorganisms while filter feeding, and a portion of these bacteria colonize the gut to form a microbiota that is in part transstadially transmitted to the adult stage [48]. Bacteria that do not persist in the gut may be digested or egested with undigested food [48] or may experience population bottlenecks in response to unfavorable conditions in the mosquito gut (e.g., high pH, immune factors). Interestingly, members of the Proteobacteria are well recognized as dominant members of the larval and adult gut microbiota of most (if not all) mosquito species [48], where they have been demonstrated to impact not only larval but adult fitness [48]. Adult female mosquitoes are also known to be attracted to oviposition sites containing bacteria, including members of the Proteobacteria [48]. In this way, selection in Biotype 1 communities may be reinforced by dispersal of microbiota that are known to colonize and persist and/or confer fitness benefits in mosquitoes. In contrast, the predominance of Cyanobacteria in Biotype 2 communities may serve as a deterrent to adult female mosquitoes during oviposition and/or reduce mosquito fitness. While these patterns suggest plausible ecological mechanisms linking mosquito productivity and bacterial community assembly, a more direct, experimental approach would be necessary to confirm the causal relationships underlying these observations. Controlled manipulations of mosquito presence, productivity, Bti treatment, and bacterial community composition across spatially replicated sites could help disentangle the relative contributions of mosquito-mediated selection, environmental filtering, and dispersal processes.

A surprising outcome of the present study was the observation that environmental indices, including temperature and forage biomass, did not independently account for any variation in community assembly. We recognize that the application of the MODIS satellite data and associated environmental indices used in this study has not been thoroughly vetted in aquatic systems, although studies do suggest that the metrics we used can be relatively good predictors of in situ measurements in freshwater ecosystems [49, 50] and the vast majority of the sites we sampled represent transitional habitats that alternate between terrestrial and aquatic over time. Our decision to average environmental indices over the entire period for which we obtained monitoring records for mosquito populations also likely impacted our ability to detect an environmental signal. Future studies would greatly benefit from including in situ measurements of environmental variables known to affect microbial assembly and mosquitoes, as well as repeated measures analyses to account for the inherent structure present in time series data. Future studies could also include longitudinal sampling of bacterial communities, given that the relative strength of different assembly processes are likely to change during community succession. Finally, we recognize that our community profiling data are derived from samples that may contain “relic” DNA from inactive (i.e., dead or dormant) bacteria, which is a constituent of total genomic DNA extracted from water, soils, and other natural material [51]. However, some work suggests that relic DNA does not strongly influence richness estimates [52, 53]. Approaches to measure protein synthesis potential using rRNA, which is likely higher in active organisms, can also be problematic in diverse communities and may still capture dormant organisms [54].

Overall, our results reveal clear associations between mosquito population dynamics and bacterial community assembly in their associated larval habitats, with bacterial communities in sites with elevated historical productivity exhibiting stronger signals of selection than those in sites with lower productivity histories. While future work is certainly warranted to assess whether these patterns are specific to our study area, this study casts new light on the mechanisms that drive changes in microbial communities in aboveground aquatic habitats and identify bacterial taxa that may serve as bioindicators of sites with improved or reduced population fitness outcomes in mosquitoes.

Materials and methods

Study sites

A total of 877 potential oviposition sites, including creeks, detention ponds, ditches, rain gardens, and retention ponds in and around Madison, WI, USA, were identified by the PHMDC in 2007 and have since been monitored each year from May through September to limit the spread of West Nile Virus. The following monitoring records were collected at each site visit: (i) presence/absence of water, (ii) presence/absence of larvae, (iii) genus identification of larvae (if present), and (iv) larval density by genus (measured as the number of larvae per 10 standard 350 ml dips) [55]. Sites where Culex species larvae were detected at a density ≥ 3 larvae per dip were flagged for treatment with a commercial formulation of a Bti-based larvicide (VectoBac, Valent BioSciences, Libertyville, IL, USA), to be administered at the next site visit. A summary of the key metrics used in the analyses presented in this study is provided in Table S1.

Microbiota sample collection and bacterial density measurements

A total of 86 larval sites were selected for microbiota sampling. At each site, 50–500 ml of water was collected in sterile conical tubes (Thermo Fisher Scientific, Waltham, MA, USA) and immediately centrifuged at high speed (21,130×g) for 20 min, supernatant removed, and pellet stored at −20 °C for downstream DNA isolation and sequencing to characterize microbiota diversity. Cell density was estimated by counting colony-forming units (CFUs) on LB agar plates incubated at 30 °C for 72 h. In brief, aliquots of fresh (uncentrifuged) water from each site were diluted to 10^− 4 and 50 µl of the diluted suspensions were used for plating in triplicate. Cell densities for each site were then estimated using the average CFU count among replicate plates.

16S microbial community profiling

DNA extractions from water samples were performed by a phenol-chloroform-isoamyl alcohol (25:24:1), bead-beating procedure [56]. Sample DNA was amplified in 25 µl reactions using dual-indexed versions of the universal bacterial primers 515 F and 806R targeting the V4 region of the 16S rRNA gene, as previously described [57]. PCR amplification was confirmed by visualizing all 25 µl of products on 1% low-melt agarose gels (National Diagnostics, Atlanta, GA, USA) prior to purification using a ZR-96 Zymoclean Gel DNA Recovery Kit (Zymo Research, Irvine, CA, USA). Purified amplicons from each sample were then pooled and paired-end sequenced (2 × 250 bp) on an Illumina MiSeq at the DNA Sequencing Facility at the University of Wisconsin-Madison. PCR reactions conducted with material from reagent-only DNA extractions or sterile water served as negative controls.

16S data processing

Raw sequence data were processed in QIIME2 [58] as follows: multi-sample fastqs were demultiplexed and adaptor and primer sequences were removed using “Cutadapt” [59]. The resulting sequences were then quality-filtered, error-corrected, chimera-cleared filtered, and pair-end reads were merged using “DADA2” to generate ASVs [60]. Suspected contaminants and sequencing artifacts were identified and removed from the ASV table using the “decontam” (frequency method) and "PERFect" R packages [61]. The remaining merged paired-end sequences were taxonomically assigned using the Greengenes reference database (13.8) [62, 63], and ASVs assigned to chloroplast or mitochondrial sequences, or not assigned to Bacteria, were removed. Taxonomy-filtered sequences were aligned using “MAFFT” [64], and phylogenies were generated with “FastTree” [65]. The resulting QIIME2-generated taxonomy table, ASV table (Table S2A), and phylogenetic tree were then imported into R (https://www.r-project.org/) and combined with sample metadata as phyloseq objects [66]. Rarefaction curves for most samples saturated between ~ 1000–3000 sequences (Fig. S5A).

Microbiome composition based on 16S ASVs

Bacterial community biotypes were identified using PAM clustering via the R package “BiotypeR” [67], based on Jensen-Shannon distances calculated from phylum-level data. The optimal number of clusters was determined using the Calinski-Harabasz index and silhouette coefficients [68, 69]. To visualize the clusters in two-dimensional space and identify taxa that differentiated biotypes, we performed Between-Class Analysis (BCA) using the “ade4” R package [70]. BCA indices were then used to assess the enrichment or depletion of phyla within each cluster. To evaluate shifts in microbial diversity between biotypes, we compared total ASV diversity and ASV richness across the phyla most responsible for biotype separation using Kruskal-Wallis tests implemented in the “PMCMR” R package [71]. Microbial diversity estimates were calculated using the “phyloseq” R package [66]. Finally, to quantify differences in microbiome composition among samples, beta diversity was assessed using Bray-Curtis and Jaccard dissimilarity indices, as well as weighted and unweighted UniFrac distances. Compositional differences were visualized through Principal Coordinate Analysis (PCoA) ordinations. The influence of biotype assignment on community structure was tested using Permutational Multivariate Analysis of Variance (PERMANOVA) with 999 permutations, and differences in group dispersion were evaluated using the betadisper function in the “vegan” R package [72]. To minimize the influence of low-abundance taxa on microbial diversity estimates, ASVs not reaching 0.5% relative abundance in at least two samples were removed (Table S2B). Samples were then rarefied to a depth of 958 reads—the minimum read count across all samples and a level at which most samples reached saturation following filtering (see Fig. S5B)—prior to downstream alpha and beta diversity analyses. Note: When included as a factor in a binomial model predicting biotype classification, neither the sampling date nor the DNA extraction date had a significant effect on classification outcomes across sites (binomial GLM, p >0.05).

Microbiota community assembly processes

To quantify community assembly processes in bacterial communities in Biotype 1 versus Biotype 2 sites, we used the null model approach described by Stegen et al. [25], which employs the “vegan” and “iCAMP” packages in R [72, 73] and has been described in detail elsewhere [74]. This method rests on the assumption that closely related taxa are also ecologically similar, which we confirmed for our data by applying Pagel’s lambda to measure phylogenetic signals for the environmental preferences of taxa using the “phytools” package in R (environmental variables are described below) [75, 76]. This null model method distinguishes between selection and neutral processes by calculation of standardized phylogenetic turnover between communities (i.e., βNTI), where phylogenetic turnover less than or greater than null expectations indicates homogeneous or variable selection, respectively, and turnover that does not differ from null expectations indicates neutral processes. Standardized compositional turnover (i.e., RC_Bray) then distinguishes between specific neutral processes, where compositional turnover less than or greater than null expectations indicates homogenizing dispersal or dispersal limitation + drift, respectively, and turnover not differing from null indicates drift alone.

Spatial analysis and variation partitioning

We assessed potential drivers of selection by conducting variation partitioning on the βNTI matrix generated using the null model method described above (varpart function, “vegan” package), following an approach similar to Fillinger et al. [77]. We considered a variety of mosquito productivity measures, environmental indices, and spatial factors as candidate driver variables. In brief, monitoring records collected from 2007 to 2018 were used to calculate the following mosquito productivity measures for each site: (i) percent total site visits with larvae from a specific genus present (i.e., Aedes, Anopheles, Culex), (ii) average larval density (measured as the total number of larvae of a specific genus detected across the entire monitoring period divided by the total number of dips taken), and (iii) total number of times Bti was administered over the entire monitoring period. Environmental indices were estimated using four MODIS products acquired using the NASA Application for Extraction and Exploring Analysis Ready Samples (AppEEARS) website (https://appeears.earthdatacloud.nasa.gov/). The first product was the 16-day composite Aqua MODIS vegetation index product MYD13Q1 (Collection 6), which has a spatial resolution of 250 m for both the Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI). The second product was the daily Land Surface Temperature (LST) product (MYD11A1, 1 km, Collection 6), from which we extracted both daytime (LST_day) and nighttime (LST_night) temperatures. The third and fourth products were the 8-day gap-filled gross primary productivity product (MYD17A2HGF, 500 m, Collection 6) and yearly gap-filled net primary productivity product (MYD17A3HGF, 500 m, Collection 6), which were used to assess the impact of Gross Primary Productivity (GPP) and Net Primary Production (NPP) on habitat microbiota assembly processes, respectively. For all products, we extracted values averaged across the entire spatial area of each site using vector polygons. Means for each index (i.e., NDVI, EVI, LST_day, LST_night, GPP, and NPP) were then calculated for each site using MODIS data acquired from across the entire monitoring period. Finally, spatial factors were represented by Principal Coordinates of Neighborhood Matrix (PCNM) scores generated using the geographic coordinates of each site (pcnm function, “vegan” package).

We selected specific variables for variation partitioning by constructing a distance-based redundancy analysis (dbRDA) model (capscale function, “vegan” package) for each variable category containing the full set of variables in that category and then used the ordistep model selection procedure (“vegan” package) to select the best-supported dbRDA model for each category. Only variables in the best-supported dbRDA model were included in the final variation partitioning analysis. We determined statistical significance of individual driver variables using a permutation test (anova.cca function, “vegan” package), while the statistical significance of each variance partition was determined using dbRDA. Because dbRDA requires positive distance values, we scaled βNTI values to range between 0 and 1 prior to analysis. Environmental variables were also scaled between 0 and 1 prior to analyses. However, we ultimately omitted environmental variables from the final variation partitioning analyses because they did not independently account for any variation in βNTI.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

S.Y.Z. and K.L.C. conceived of the study. S.Y.Z., J.H., and K.L.C. contributed to the study design. S.Y.Z. and K.L.C. collected the study data and carried out the data analysis. S.Y.Z. wrote the initial manuscript, and K.L.C. contributed to revisions.

Data availability

Raw Illumina reads are available in the NCBI Sequence Read Archive (https://www.ncbi.nlm.nih.gov/sra) under BioProject ID PRJNA1180561. Raw data files and scripts used for analysis and figure generation are available in the Coon laboratory’s GitHub repository (https://github.com/kcoonlab/dane-microbes).

Declarations

Competing interests

The authors declare no competing financial interests.