Hologenomic structure of bacterial and fungal community composition in the West Nile virus vector Culex tarsalis

Abstract

Background

Microbiota play a crucial role in determining the ability for arthropod disease vectors to transmit pathogens. Microbial community structure can be heavily influenced by microbe-microbe interactions, host genetics and environmental factors. Here, we characterize the host population genetic structure, and bacterial and fungal communities in natural populations of the West Nile virus mosquito vector Culex tarsalis. Mosquitoes were collected and analyzed across the species range of the mosquito in the United States, where we used PoolRADSeq to quantify population genetic structure. Microbial community composition was characterized using bacterial 16S rRNA gene sequencing (V3-V4 region) and fungal ITS sequencing (ITS1 region).

Results

PoolRADSeq identified four broad genetic clusters of mosquito populations, which corresponded to previous clusters identified by microsatellite analysis and RADSeq on individual mosquitoes. Microbiome diversity grouped mosquito populations into three broad clusters, with each cluster distinctively represented by diagnostic abundant bacteria (Ralstonia, Pseudomonas, or Zymobacter/Providencia, respectively). Clustering for fungal taxa was less pronounced. Geographic distance between populations was positively correlated with microbiome community dissimilarity, and multiple environmental factors were significantly correlated with microbial species richness and diversity.

Conclusions

These results suggest that bacterial and fungal communities are geographically structured in Cx. tarsalis, interact with important environmental factors, and are partially correlated with host genetic structure. As microbiota can affect the ability for mosquitoes to transmit pathogens, understanding the factors underpinning microbiome variation across space and time has important implications for the spread of vector-borne pathogens such as WNV.

Introduction

Culex tarsalis is one of the most important native North American mosquito arbovirus vectors [1]. The species is widely distributed in western North America, feeds primarily on avian hosts which contributes to virus amplification [2], and can move relatively long distances in search of a blood meal or oviposition sites [1,2,3]. Historically Cx. tarsalis is an important vector of Western equine encephalitis virus (WEEV) and Saint Louis encephalitis virus (SLEV) [4]. When West Nile virus (WNV) was introduced into the Americas in 1999, Cx. tarsalis was demonstrated to be one of the most competent vectors of this novel invasive pathogen [5].

Previously, population genetic data was used to implicate a role for Cx. tarsalis in transporting WNV across the western United States. An analysis of microsatellite loci suggested that across the western USA, Cx. tarsalis populations are structured into approximately three broad metapopulation clusters (“Midwest”, Sonoran [or “Southwest”], and “Pacific”), which closely matched the observed pattern of WNV geographic spread in the early 2000s [1,3]. A more recent population genetic analysis, using RADSeq on approximately 300 individual mosquitoes, identified a broadly similar pattern (with increased resolution and a new identified cluster [“Northwest”] due to the denser marker set) [6].

Knowledge of the factors influencing the spread of arboviruses in mosquito populations is critical for understanding dynamics of disease incidence, development of risk assessment strategies for novel virus introductions and development of virus transmission biomarkers that can be used to efficiently target control efforts [1,7]. In general, there are three main barriers that an arbovirus must overcome to be transmitted by a mosquito vector. The virus particles must 1) establish an infection of the midgut epithelium and replicate, 2) escape the midgut through the basal lamina and infect other body tissues, and 3) infect the salivary glands, escape into the salivary gland lumen and be transmitted during salivation [8,9,10]. These barriers can be modulated by the intrinsic host genetics of the mosquito, components of the endogenous mosquito microbiota, and environmental factors, which has been demonstrated for multiple mosquito species in both the lab and the field [11,12,13,14,15,16,17]. It has become increasingly clear that while pathogen transmission can be affected by host genetics and environmental factors, this is, at best, an incomplete representation of the factors influencing pathogen transmission phenotypes. Recent research on arthropod holobiomes (the host plus all associated microorganisms) has demonstrated the potential impact of mosquito-associated microorganisms on transmission-related phenotypes of a variety of pathogens [15,16,17,18,19,20,21,22, 23,24,25,26,27,28].

Laboratory studies have demonstrated that components of the mosquito holobiome such as bacteria and fungi can modulate pathogen vector competence phenotypes. For example, bacteria in the genera Chromobacterium, Enterobacter, Serratia, Pseudomonas, and Delftia can decrease Plasmodium oocyst levels in Anopheles mosquitoes [17,20,21,23], while bacteria in the genera Serratia and Aeromonas can increase mosquito susceptibility to arboviruses [22,24]. Less is known about how fungi alter pathogen vector competence in mosquitoes, but examples exist. Specific isolates of Penicillium enhanced Anopheles infection with Plasmodium [25], while Metarhizium anisopliae infection rendered Aedes mosquitoes less susceptible to dengue virus [26]. These data suggest that components of the mosquito microbiome can be potent modulators of pathogen transmission, but despite their importance there is very little data on what role these microorganisms play in determining vector competence variation in the field.

Little is known about the natural microbiome of Cx. tarsalis or its effects on pathogen infection and transmission. Current studies are limited to characterizing bacterial communities in Cx. tarsalis mosquitoes reared in experimental mesocosms, where communities were dominated by the aquatic bacteria Thorsellia [29, 30]; a common microbe found in freshwater aquatic environments [30]. Nothing is currently known about fungal communities in this species. To address this knowledge gap, we examined the bacterial and fungal communities in wild-caught Cx. tarsalis mosquitoes from populations across the species range in the western USA. Combined with host RADSeq data, we found signatures of population structuring in microbial communities that were partially correlated with mosquito population genetic structure, where mosquitoes clustered into three broad groups based on their microbial composition. Microbial structure was more pronounced for bacteria than fungi, possibly reflecting differences in microbe acquisition mechanisms. Geographic distance between populations was positively correlated with microbial community dissimilarity. Environmental factors such as longitude, latitude, temperature, and precipitation were significantly correlated with species richness and/or diversity. These results suggest that bacterial and fungal communities in this mosquito are geographically structured, are potentially shaped by host genetics, and interact with or are affected by environmental factors. These results have critical significance for understanding patterns of microbial ecology in this mosquito species, as well as having implications for the epidemiology of vector-borne pathogens such as West Nile virus.

Methods

PoolRADSeq

2b-RADseq on pooled mosquitoes was used to characterize Cx. tarsalis population structure. Wild Cx. tarsalis mosquitoes were collected from 15 populations in 2021 and 18 populations in 2022 using CDC traps (Fig 1, Table 1). Mosquitoes were shipped overnight on dry ice to Penn State and stored at −80 °C until processing. For each population, we created pools of 10 randomly selected mosquitoes. DNA was extracted from individuals using Qiagen spin columns, quantified with Qubit, and equal amounts of DNA from each individual were combined to form each pool. After DNA QC, standard 5’ –NNN- 3’ adaptors were used to ligate restriction tags for concatenated Isolength restriction site-associated tags sequencing. Genomic DNA was digested with the type IIB restriction enzyme BsaXI to produce restriction fragments of uniform length, followed by adaptor ligation, amplification, purification, concatenating and barcoding to incorporate sample-specific barcodes. Samples were then pooled, QC assessed, and sequenced on the Illumina Hiseq Xten/NovaSeq platform using paired end 150bp sequencing. Reads were quality filtered. Reads containing the restriction endonuclease recognition site were extracted from the sequencing data and clustering was performed with ustacks software (version 1.34) from the Stacks package to construct a reference sequence. Sequencing data was aligned to the reference sequence using SOAP software (version 2.21), and SNP calling was performed using Maximum Likelihood (ML) using the RADtyping package [31]. A total of 9,626 quality-filtered SNP markers were obtained across all samples. Phylogenetic trees were constructed using Neighbor Joining. Library construction, sequencing, and analysis were conducted by Novogene.

Fig. 1.

Map of collection locations for mosquitoes assayed for PoolRADSeq analysis. Red = 2021; Blue = 2022.

Table 1.

Collection information for PoolRADSeq samples

Collection year	Sample ID	Location	State	Latitude	Longitude
2021	P1_2021	Lake point	UT	40.669406	−112.305263
2021	P2_2021	Stansbury Park	UT	40.651018	−112.299081
2021	P3_2021	Stansbury Park	UT	40.647411	−112.314970
2021	P4_2021	Yakima/Grandview	WA	46.193421	−119.899400
2021	P5_2021	Yakima/Grandview	WA	46.224977	−119.901053
2021	P6_2021	Yakima/Grandview	WA	46.205611	−119.892026
2021	P7_2021	Lubbock	TX	33.462700	−101.891400
2021	P9_2021	Loveland	CO	40.403933	−105.071421
2021	P10_2021	Loveland	CO	40.401394	−105.108170
2021	P12_2021	Cottonwood, Shasta County	CA	40.414286	−122.216945
2021	P13_2021	Anderson, Shasta County	CA	40.445232	−122.243092
2021	P14_2021	Cottonwood, Shasta County	CA	40.396727	−122.281994
2021	P15_2021	Yolo County	CA	38.610336	−121.711942
2021	P16_2021	Sacramento County	CA	38.107311	−121.649758
2021	P17_2021	Yolo County	CA	38.810503	−121.808285
2022	Z1	Riverside county, Mecca	CA	33.53271	−116.03503
2022	Z2	Riverside county, Thermal	CA	33.46238	−116.06125
2022	Z3	Yolo county, Conaway Ranch	CA	38.64849	−121.71102
2022	Z4	Yolo county, Vic Fazio	CA	38.56179	−121.62924
2022	Z5	Adams County	CO	39.90015	−104.91565
2022	Z6	Broomfield	CO	39.89499	−105.13878
2022	Z7	Larimer county, Loveland	CO	40.43432	−105.10429
2022	Z8	Larimer county, Fort Collins	CO	40.48614	−104.97362
2022	Z9	Mapleton	ND	46.89102	−97.04281
2022	Z10	Mapleton	ND	46.88508	−97.06925
2022	Z11	Grand Forks county, Grand Forks	ND	46.94475	−97.06344
2022	Z12	Grand Forks county, Grand Forks	ND	47.88686	−97.09869
2022	Z13	Cascade county, Sunriver	MT	47.51992	−111.77331
2022	Z14	Cascade, Vaughan	MT	47.53991	−111.63364
2022	Z15	Yakima county, Grandview	WA	46.22498	−119.90105
2022	Z16	Yakima county, Mabton	WA	46.20787	−119.99583
2022	Z17	Lubbock, Lubbock	TX	33.58318	−101.89314
2022	Z18	Lubbock, Lubbock	TX	33.58254	−101.82794

Microbiome mosquito collection and DNA extraction

Cx. tarsalis mosquitoes were collected from 17 locations in five states (CA, CO, UT, TX, and WA) in 2021 (Fig 2, Table 2). Mosquitoes were collected using CO₂ baited EVS (encephalitis vector surveillance) or CDC traps overnight and morphologically identified at species level by the collecting laboratory. Mosquito samples were stored in DNA/RNA Shield (Zymo Research, Irvine, USA) at −80C until they were shipped to Penn State University. Morphological species identification of Cx. tarsalis was confirmed, and individual mosquitoes were surface sterilized in 500 μl 70% ethanol for 5 minutes followed by 3 washes in PBS. All tools used for handling mosquitoes were sterilized using absolute ethanol to minimize cross contaminations. DNA extractions were performed using ZymoBiomics DNA/RNA miniprep kit (Zymo Research, Irvine, USA) following the manufacture’s protocol with some modifications. In brief, individual whole mosquitoes were homogenized in 2.0mm Zymo BeadBeating Lysis tubes (Zymo Research, Irvine, USA) for 30 second at 6m/s in 750ml Zymo DNA/RNA Shield lysis buffer (Zymo Research, Irvine, USA) using a Fisher brand Beadmill 24 (Fisher Scientific, Waltham, MA) followed by additional mechanical homogenization in 0.1–0.5mm BeadBeating Lysis tubes for two 30 second cycles at 6m/s with a 45 second dwell. The homogenization protocol was optimized to ensure extraction of both bacterial and fungal DNA from a single mosquito while minimizing DNA degradation by excessive shearing. To minimize batch effects, two randomly selected populations were processed per day by two personnel, with a blank (PBS) extraction added as negative control for each batch (total 16 negative controls). All negative controls were determined to be negative for 16S PCR and did not pass QC for library preparation. For positive control, two units of ZymoBIOMICS Microbial Community Standard (Zymo Research, Irvine, USA) were used to validate our extraction method by confirming detection of known bacterial and fungal species in the standard mixture. Quantity and quality of DNA were checked using Qubit 4.0 fluorometer (Fisher Scientific, Waltham, USA) and nanodrop, respectively. DNA samples were stored at −80C until down-stream processing. Sample sizes and collection locations are listed in Table 2.

Fig. 2.

Map of collection locations for mosquitoes assayed for microbiome analysis. Colors indicate States.

Table 2.

Collection information for microbiome samples

Population ID	State	Number of samples		City/County	Latitude	Longitude
		16S	ITS1
CA1	CA	17	0	Cottonwood, Shasta County	40.41429	−122.21695
CA2		19	2	Anderson, Shasta County	40.44523	−122.24309
CA3		7	1	Cottonwood, Shasta County	40.39673	−122.28199
CA4		14	8	Yolo County	38.61034	−121.71194
CA5		10	7	Sacramento County	38.10731	−121.64976
CA6		10	11	Yolo County	38.8105	−121.80829
CO1	CO	15	3	Loveland	40.40393	−105.07142
CO2		16	4	Loveland	40.40139	−105.10817
CO3		3	0	Fort Collins	40.52993	−105.05145
TX1	TX	8	8	Lubbock	33.4627	−101.8914
TX2		3	0	Lubbock	33.59046	−101.93866
UT1	UT	14	11	Lake point	40.66941	−112.30526
UT2		13	4	Stansbury Park	40.65102	−112.29908
UT3		13	1	Stansbury Park	40.64741	−112.31497
WA1	WA	16	6	Yakima/Grandview	46.19342	−119.8994
WA2		10	2	Yakima/Grandview	46.22498	−119.90105
WA3		9	4	Yakima/Grandview	46.20561	−119.89203

Microbiome sequencing and bioinformatics analysis

Bacterial and fungal communities were characterized by splitting the DNA from same individual mosquito. For bacterial characterization, the V3-V4 hypervariable region of 16S rRNA gene was amplified using primers 341F (5’- CCTAYGGGRBGCASCAG-3’) and 806R (5’-GGACTACNNGGGTATCTAAT-3’) with expected amplicon size of approximately 450–550bp. For fungal characterization, the Internal Transcribed Spacer (ITS) 1 region was amplified using primers ITS1F (5’-CTTGGTCATTTAGAGGAAGTAA-3’) and ITS2 (5’-GCTGCGTTCTTCATCGATGC-3’) with expected amplicon size of approximately 200–400bp. PCR was carried out in 30 μL reactions with 15 μL of Phusion High-Fidelity PCR Master Mix (New England Biolabs), 0.6 mM of forward and reverse primers, and 5~10 ng of template DNA (10 μL for negative controls). Thermal cycling for bacterial characterization consisted of initial denaturation at 98°C for 1 min, followed by 30 cycles of denaturation at 98°C for 10 sec, annealing at 50°C for 30 sec, and extension at 72°C for 30 sec, and lastly by 72°C for 5 min. The same cycle was used for fungal characterization, except for the annealing temperature of 55°C for 30 sec. PCR products within the expected size were selected by 2% agarose gel electrophoresis. The same amount of PCR product from each sample was pooled, end-repaired, A-tailed and further ligated with Illumina adapters. PCR products were purified using the Qiagen Gel Extraction Kit (Qiagen, Germany). Sequencing libraries were generated using NEBNext Ultra DNA Library Pre ^®Kit for Illumina. Libraries were commercially sequenced (Novogene) to generate 250 bp paired-end raw reads. The library was checked by Qubit and real-time PCR for quantification, and bioanalyzer for size distribution detection. Quantified libraries were pooled and sequenced on the NovaSeq PE250 platform (Illumina, San Diego, CA, USA).For sequencing and data processing, paired-end reads were assigned to samples and truncated by cutting off the barcode and primer sequences. After truncation, FLASH (v1.2.11, ccb.jhu.edu/software/FLASH/) was used to merge reads to get raw tags [2]. Then, fastp software was used to perform quality control of raw tags, and high-quality clean tags were obtained. Finally, chimeras were removed using VSEARCH. For denoising and species annotation of ASVs (Amplicon Sequencing Variants), DADA2 was used to reduce noise and sequences with abundance less than 5 were filtered out to obtain the final ASV [32]. Representative sequences for each ASV were annotated to obtain the corresponding species identification using QIIME2. By applying QIIME2’s classify-sklearn algorithm [33,34], a pre-trained Naive Bayes classifier was used for species annotation of each ASV. PCR, library preparation, sequencing and data processing were performed by Novogene.

Microbiome data and statistical analysis

Data and statistical analyses were conducted using MicrobiomeAnalyst 2.0 [35], GraphPad Prism (v8.0), and R. For data filtering prior to downstream analyses, features that were constant across all samples and singletons (those with only one total count) were removed. To eliminate low-quality or uninformative features, the low-count filter was set to zero to avoid random selection bias for tied low-count data, and the low-variance filter was set to 10% interquartile range, followed by a data normalization process. Filtered data were used for all downstream comparative analyses unless otherwise specified. To understand overall bacterial and fungal community structure, we first constructed relative abundance and heat trees of representative taxa. Based on visual inspection of these data, we hypothesized a clustering of mosquito populations and validated this grouping through a series of comparative analyses including Alpha and Beta diversity, Random Forest classification, Core microbiome analysis, dendrogram construction, Linear Discriminant Analysis Effect Size (LEfSe), correlation network analysis, and correlations between environmental factors, community structure, and geographic distance. For Alpha diversity analysis, non-filtered original data were used. For Random Forest classification, five consecutive Out-of-Bag (OOB) error tables were generated to estimate the mean OOB error. Correlation network analysis employed Sparse Estimation of Correlations among Microbiomes (SECOM) with Pearson1, which efficiently handles both linear and non-linear relationships among microbes while maintaining sparsity. Correlation analyses between geographical distance and community dissimilarity indices were performed using the Mantel test (vegan package) in R. All other analyses performed in MicrobiomeAnalyst 2.0 followed the default settings as recommended [35]

Wolbachia verification

Wolbachia was detected by 16S amplicon sequencing in a few (12/197) samples. For these samples we attempted to verify this finding by amplifying and sequencing the Wolbachia surface protein (wsp) gene using primers 81F and 691R [36]. We also amplified and sequenced the mosquito mitochondrial ND4 gene [37] to verify mosquito species identification. PCR and Sanger sequencing for these genes was carried out as described in the respective references.

Results

RADSeq population genetics

A total of 9,626 SNP markers that passed quality control were obtained across all samples, which were distributed across UTRs, introns, exons, and intergenic regions. Results from both years were combined for analysis to obtain a more complete spatial sampling across the species range of the mosquito. Clustering neighbor-joining results from PoolRADSeq were qualitatively similar to previous results from RADSeq on individual mosquitoes [6] and microsatellites [1,3], resolving the same metapopulation clusters (Pacific, Sonoran/Southwest, Northwest, and Midwest) that previous analyses [1,3,6] identified (Fig 3).

Fig. 3.

Neighbor-joining tree generated from PoolRADSeq analysis. Clusters previously identified by Venkatesan and colleagues (microsatellites) [1] and Liao and colleagues (individual RADSeq) [6] are highlighted.

Microbiome composition profile

For processing bacterial sequence data, total read count was 27,771,776, and the average count per sample was 140,973 (Maximum = 502,316; Minimum = 10,132) from 197 samples that passed QC. A total 107 ASVs were left of 66396 by removing singletons (one total count) and low abundance ASVs (less than 4 read count). For fungal sequence data, total read count was 16,402,771 and the average count per sample was 227,816 (Maximum = 792,580; Minimum = 23,624) from 72 samples that passed QC. A total of 42 ASVs were left of 8,261 by removing singletons and low abundance ASVs. For bacterial and fungal combined sequence data, total read count was 26,776,978 and the average count per sample was 377,140 (Maximum = 1,001,576; Minimum = 76,642) from 71 samples. 163 ASVs were left of 74,657 by removing singletons and low abundance ASVs.

We first investigated overall community composition in bacteria and fungi in across all Cx. tarsalis populations. For bacterial communities, Proteobacteria were the most abundant at the phylum level (89.8%) which was mainly composed of Gammaproteobacteria (85.7% of total abundance) (Fig 4; Additional file 1). Within Gammaproteobacteria, Burkhoderiales was most abundant followed by Pseudomonadales, Enterobacterales and Oceanospirillales, representing over 85% of total abundance at the order level. Within Gammaproteaobacteria, Ralstonia, Pseudomonas, Zymobacter, and Providencia were the most abundant taxa representing 75% of total abundance at the genus level. While the top three abundant taxa were found in 74% to 96% of all 197 samples, Providencia was relatively less prevalent (25%).

Fig. 4.

Bacterial communities in Culex tarsalis mosquitoes. A: Percent relative abundance of the top 10 identified taxa. B: Heat tree showing hierarchical structure of representative taxa. Taxa with 50% prevalence (% ASV detected in all samples) or higher are presented (family names are not labeled for better visibility). Genera within the top 10 relative abundance (see Fig. 1A) are in red text. Color and node indicate prevalence and number of ASV, respectively. See Additional file 1 for heat tree with all taxa represented.

We detected Wolbachia in approximately 6% (12/197) of samples. In previous surveys, Cx. tarsalis was found to be uninfected with Wolbachia [39], so we attempted to further validate this finding. From these 12 samples, we amplified and sequenced the Wolbachia wsp gene to strain type the infection. PCR was successful in only 2 samples, where the infection was identified as wPip (the canonical Wolbachia strain infecting Cx. pipiens/quinquefasciatus; [40]); the remaining 10 samples either did not amplify or amplified random mosquito DNA. We then amplified and sequenced the mosquito mitochondrial ND4 gene to molecularly identify the mosquitoes. The two verified Wolbachia-positive samples were identified as Cx. quinquefasciatus. The other 10 samples were identified as Cx. tarsalis. These data indicate that Wolbachia infection is not present in Cx. tarsalis mosquitoes and the observed rare positive samples were due to sample misidentification or potentially contamination. The two Cx. quinquefasciatus samples were removed from further downstream analysis.

For fungal communities, Ascomycota and Basidiomycota were the two most abundant phyla representing over 79% of total abundance (Fig 5, Additional file 1). While Basidomycota was mainly composed of Agaricomyctetes (21.4%), Ascomycota was mostly represented by Dothideomycetes (53.1%) at the class level. Within Dothidemycetes, Capnodiales and Pleosporales were the two most abundant taxa at the order level representing over 53% of total abundance. Mycosphaerella (28.4%) and Cladosporium (16%) were the most abundant within Dothideomycetes while Alternaria was most abundant (3.8%) within Pleosporales at the genus level, and these three taxa were detected in 83 to 94% of all 72 samples. While Ganoderma was the second most abundant taxa (21.4%) belonging to Basidiomycota, its prevalence was relatively lower (36%).

Fig. 5.

Fungal communities in Culex tarsalis mosquitoes. A: Percent relative abundance of the top 10 taxa. B: Heat tree showing hierarchical structure of representative taxa. Taxa with 50% prevalence or higher are labeled. Genera within the top 10 relative abundance (see Fig 2 A) are in red text. Color and node size indicate prevalence (% sequence read detected in all samples) and number of ASV, respectively. See Additional file 1 for heat tree with all taxa presented.

Clustering of mosquito populations by microbial community structure

We identified a pattern where specific bacterial taxa at the genus level were more abundant in certain clusters of populations. Based on visual inspection and geographic proximity, we hypothesized three clustering groups of mosquito populations; CA1, CA2, CO1, CO2, CO3, TX1, and TX2 for group CCT (initials of California, Colorado and Texas), CA3, CA4, CA5, and CA6 for group CA (California), and UT1, UT2, UT3, WA1, WA2, and WA3 in group UW (initials of Utah and Washington) (Fig. 6A and Additional file 2). With this grouping, Ralstonia and Pseudomonas were the most abundant taxa in group CCT (70.2% relative abundance) and UW (66%), respectively, while Zymobacter (35.6%) and Providencia (31.3%) were most abundant in group CA (Additional file 2, 3).

Fig. 6.

Hypothesized clustering pattern of mosquito populations based on community structure of A: bacteria, B: fungi, and C: bacteria/fungi combined at the population (left) or group level (right). Numbers within each bar graph indicate the number of mosquito samples assayed.

Clustering of fungal community structure according to this grouping was less pronounced (Fig. 6B; Additional file 2, 4). For example, while Ganoderma was most abundant in UW (36.5% relative abundance) and Cladosporium was relatively more abundant in CA (37.8%) than CCT (17.8%) or UW (6.3%), Mycosphaerella were found in all three groups with similar abundance ranging from 18.3 to 33.1%. Similarly to the case of bacteria, grouping was more pronounced at order or lower taxonomic levels (Additional file 4). When community structure of bacteria and fungi were combined, clustering was moderate, similarly to the case of fungal community (Fig 6C; Additional file 2; Additional file 5) as relative abundance of major taxa identified above (i.e., Pseudomonas, Zymobacter, Providencia, Mycosphaerella, Cladosporium and Ganoderma) followed the hypothesized grouping model.

Alpha and Beta diversity, and Random Forest classification

Alpha diversity analyses, Chao1 (species richness) and Shannon (species diversity) indices were compared for community structure of bacteria, fungi, or bacterial/fungi combined (Fig. 7A). Chao1 indices were significantly different among the three groups (P < 0.05, Kruskal-Wallis) with higher index value in group CCT relative to CA (P = 0.006, Bonferroni corrected after pair-wise comparison among three groups) while Shannon indices were not different among three groups (P > 0.05, Kruskal-Wallis). For fungal communities, Chao1 indices were similar among three groups while a difference was observed for Shannon indices among the groups (P < 0.01, Kruskal-Wallis). When bacteria/fungi were combined, no difference was observed for Chao1 indices among the groups (P > 0.05, Kruskal-Wallis), but Shannon index was lowest for UW (P < 0.001, Kruskal-Wallis; Bonferroni corrected after pairwise comparison). Rarefaction curve analyses and comparison of sequencing depth among three groups indicated overall high sequencing depth across samples with minimal heteroskedasticity regardless of data filtering (Additional file 6). Beta diversity was next explored using Nonmetric Multidimensional Scaling (NMDS), revealing a significant differences in community structure among the three groups (PERMANOVA, P < 0.01) for bacterial, fungal, and bacterial/fungal combined (Fig. 7B).

Fig. 7.

Alpha and Beta diversity analysis. A: Chao1 (species richness) and Shannon index (species diversity) were compared for Alpha diversity analysis. Pair-wise Mann-Witney tests were conducted between groups followed by Benjamini-Hochberg correction. B: Nonmetric Multidimensional Scaling (NMDS) with Bray-Curtis distance matric was estimated followed by Permutational Multivariate Analysis of Variance (PERMANOVA) for Beta diversity analysis.

We then performed machine learning based Random Forest classification analyses to validate our grouping model (Fig. 8). For bacterial communities, the mean Out-of-bag (OOB) error was less than 0.1 in five consecutive runs (Fig. 8A; Additional file 7) indicating that about 90% of predictions were correct by our grouping model, and Ralstonia was identified to have the greatest influence on the classification model followed by Providencia, Pseudomonas, Collinsella and Zymobacter (Fig. 8B). Similarly, dendrogram analysis largely supported our clustering model regardless of algorithms applied (Additional file 8). Classification was less clear for fungal data with slightly higher OOB error (0.228) (Fig. 8A; Additional file 7, 9). The top five fungal genera that had greatest influence on the classification were Sporobolomyces, Sarocladium, Ganoderma, Cladosporium, and Aspergillus; Mycosphaerella (the most abundant fungus) had relatively little influence (Fig. 8B). For combined bacterial/fungal data, intermediate OOB error (0.152) was observed (Fig. 8A; Additional file 7, 10), and Zymobacter had greatest influence on classification followed by Providencia, Sporobolomyces, Pseudomonas, and Saccharibacter (Fig. 8B).

Fig. 8.

Random Forest classification analysis with A: representative tree plots, and B: top 15 taxa at genus level that had greatest influence on classification models. Mean OOB (Out Of Bag) errors were calculated using five consecutively generated OOB tables (see Additional file 7) (*,Burkholderia, Caballeronia, Paraburkholderia).

Core microbiome, taxon differentiation, and correlation network

Prevalence (% detected in samples) profiles were characterized using Core microbiome analysis (Fig. 9). For bacterial data, Ralstonia was most prevalent followed by Pseudomonas, Zymobacter, Entomospria and Escherichia in all samples. When each group was analyzed separately, Ralstonia and Pseudomonas were most prevalent in group CCT and UW, respectively while Zymobacter and Providencia were most prevalent in group CA. For fungal data, Cladosporium and Mycosphaerella were widely found in all three groups while Alternaria and Ganoderma were found in CCT and UW respectively. A similar pattern was observed when bacteria/fungal data were combined.

Fig. 9.

Core microbiome analysis. Taxa with 20% or higher prevalence are presented at genus level. The detection threshold was set at 0.01% relative abundance (*, Burkholderia, Caballeronia, Paraburkholderia).

The major taxa identified in Core microbiome analyses were consistently observed in Linear Discriminant Analysis with effect size (LEfSe) for bacterial data (Additional File 11). Ralstonia and Pseudomonas were significantly more abundant in CCT and UW, respectively while Zymobacter and Providencia were more abundant in group CA. However, for fungal data, Cladosporium and Ganoderma were more abundant in group CA and UW, respectively, but Mycospharella was not significantly enriched in either of three groups. A similar pattern was observed for combined bacterial/fungal data.

To explore whether interactions among taxa had influence on our clustering model, we investigated correlation network for the major taxa identified above. For bacteria, Ralstonia, Pseudomonas, and Zymobacter were uncorrelated (Additional file 12; Additional file 13) while for fungi, Mycosphaerella and Cladosporium were positively correlated, while Mycosphaerella was negatively correlated with Alternaria. In total, 33 significant correlations were identified between bacteria and fungi, and 85% (28 cases) of them were negative. The two most abundant fungi, Mycosphaerella and Cladosporium, accounted for 50% (14 cases) of total negative correlations (Additional file 14A). Consistent with this result, species richness (Observed ASV/Chao1) was negatively correlated between bacteria and fungi (R = −0.3, P = 0.0119) (Additional file 14B). These results indicate a possible exclusive interaction between bacteria and fungi within the mosquito microbiota.

Relationship between environmental factors and community structure

Because Cx. tarsalis population genetic structure was recently observed to be correlated with environmental factors [6], we explored whether environmental factors were correlated with microbial community structure or species richness/diversity in this mosquito. We first examined the relationship between environmental factors (e.g., longitude, latitude, temperature, and precipitation) and Alpha diversity indices (Fig. 10A). For bacteria, ASV/Chao1 and ACE were positively correlated with longitude and precipitation while these two indices were negatively correlated with temperature. For fungi, ASV/Chao1 and ACE were negatively correlated with longitude while these two indices together with Shannon and Simpson indices were positively correlated with temperature. Latitude was also positively correlated with Shannon and Simpson indices. For combined bacterial/fungal data, no correlation was found with longitude for any indices, but all five indices were positively correlated with temperature and negatively correlated with latitude. These results suggest species richness/diversity is shaped by multiple environmental factors. Particularly, species richness of bacteria and fungi were negatively correlated along longitude.

Fig. 10.

Correlation analyses between environmental factors and species richness/diversity or geographic structure. A: Correlation between abiotic environmental factors and Alpha diversity indices using linear regression. Positive and negative numbers indicate positive (red) and negative (blue) correlation, respectively (***, P < 0.001; **, P < 0.01; *, P < 0.05). B: Correlation between geographic distance and Bray-Curtis dissimilarity index using Mantel test. Line plots with shaded areas indicate linear regression with 95% CI.

We noted a significant positive correlation between geographic distance and community dissimilarity for bacterial, fungal and combined bacteria/fungal data (Mantel test, P < 0.01; Fig. 10B), indicating that the community structure is more dissimilar as distance between collection sites increases for both bacteria and fungi and suggesting possible phylogenomic signal of microbial diversity in Cx. tarsalis.

Discussion

Our population genetic analysis of Cx. tarsalis across the western USA identified essentially the same population structure that previous microsatellite and individual RADSeq analyses. In total, these analyses demonstrate that gene flow is substantial between Cx. tarsalis populations across wide geographic distances, and that the metapopulation structure can be defined in terms of 3–4 large metapopulation clusters encompassing the Midwest, the Pacific Coast, the Sonoran desert, and potentially the Northwest [1,3,6]. Although it does not resolve individuals, PoolRADSeq produced similar population-level results at significantly lower cost.

There are no previous studies on the diversity of fungi colonizing Cx. tarsalis, and previous bacterial microbiome studies in this species were limited to analysis of mosquitoes that were reared in artificial field mesocosms [29]. These mesocosms were constructed in an attempt to mimic the natural habitat of this mosquito; however, the resulting microbial composition data differed significantly compared to our results using true wild-caught Cx. tarsalis. The bacterial microbiomes of mesocosm-reared mosquitoes were dominated by Thorsellia [29] a common aquatic bacterial taxon that has been previously associated with mosquitoes and other aquatic species [30, 41], and likely reflects the unnatural microbial colonization of the mesocosm habitats. In our study, we observed that distinct microbial taxa were enriched in wild-reared mosquitoes, and these taxa differed depending on the geographic region the mosquitoes were collected from. These data suggest that microbiome studies should, if at all possible, be conducted on individuals that were collected from natural habitats, as artificial habitats may not reflect the microbial diversity observed in nature and can give misleading results.

Wolbachia is a maternally inherited alphaproteobacterial symbiont that is of interest as an agent for control of arbovirus transmission in mosquitoes [27,28,29,40,43,44]. Previous studies suggested that Cx. tarsalis is not naturally infected with Wolbachia [39]. Artificial transfection with Wolbachia in Cx. tarsalis was shown to increase, rather than decrease, their susceptibility to WNV, thus detection of Wolbachia in this species has significant implication for the epidemiology of WNV transmission (and potentially other arboviruses) [45]. Therefore, when we detected Wolbachia by 16S amplicon sequencing in a small subset of individuals we attempted to independently validate the finding. Out of 12 samples where Wolbachia was detected by amplicon sequencing, we were able to independently validate the infection in only two individuals, which were ultimately molecularly identified as Cx. quinquefasciatus; a species which is known to be natively infected with Wolbachia [39,40]. In the remaining 10 samples, Wolbachia infection could not be verified, and these samples were molecularly identified as Cx. tarsalis. In total, these data affirm previous findings that Cx. tarsalis is not naturally infected with Wolbachia and highlight the importance of independently verifying low frequency Wolbachia infections detected solely on the basis of 16S amplicon microbiome sequencing.

Recent studies identified correlations between Cx. tarsalis population genetics and environmental variables such as temperature, humidity, and vegetation [6]. We therefore examined potential links between microbiome composition and environment. We found that bacterial diversity was positively correlated with longitude and precipitation, and negatively correlated with temperature. Fungal diversity was negatively correlated with longitude and positively correlated with temperature. The mechanisms underlying these relationships are at present unclear, but the data suggest that correlations between environmental parameters and mosquito genetics are confounded by microbial interactions. Future studies are indicated to tease apart the mechanisms underlying these interactions.

Our data suggests that microbiome structure in Cx. tarsalis is, to some extent, correlated with population genetic structure (Fig 11). While this relationship is not completely one-to-one with the genetic clusters defined by RADSeq, it is significant and partially overlaps with these clusters. Mantel regression demonstrates significant isolation by distance in microbial structure. Phylogenomic signal is more pronounced for bacterial data compared to fungi. Generally, mosquitoes acquire bacterial microbiome associates from the aquatic larval environment [19,46]. Less is known about how mosquitoes acquire fungal associates. Differences between bacteria and fungi could reflect different modes of acquisition between these two groups, differences in retention between bacteria vs. fungi during mosquito development, or differences in mosquito immune response to bacteria vs. fungi in response to colonization.

Fig. 11.

Comparison of A: Cx. tarsalis population genetic structure and B: Microbiome community structure. Colors denote grouping identities as described in the text.

Conclusions

Microbial associates have been demonstrated to affect pathogen infection and transmission in a variety of mosquito-pathogen systems [15,16,17,18,19,20,21,22, 23,24,25,26,27,28]. This is the first study to examine diversity of microbial association in natural populations of Cx. tarsalis across its species range. Identification of microbial taxa that can affect WNV or other pathogens in this mosquito has significant implications for understanding the spread and invasion of the viral pathogens transmitted by this mosquito across space and time.

Availability of data and materials

Raw bacterial microbiome data, fungal microbiome data, and PoolRADseq data are available at zenodo.org/records/14081783, zenodo.org/records/14080098, and zenodo.org/records/14081214.

Code availability

Code for analyses is available at zenodo.org/records/17014950.