Analysis of trapped mosquito excreta as a noninvasive method to reveal biodiversity and arbovirus circulation

Abstract

Emerging and endemic mosquito‐borne viruses can be difficult to detect and monitor because they often cause asymptomatic infections in human or vertebrate animals or cause nonspecific febrile illness with a short recovery waiting period. Some of these pathogens circulate into complex cryptic cycles involving several animal species as reservoir or amplifying hosts. Detection of cases in vertebrate hosts can be complemented by entomological surveillance, but this method is not adapted to low infection rates in mosquito populations that typically occur in low or nonendemic areas. We identified West Nile virus circulation in Camargue, a wetland area in South of France, using a cost‐effective xenomonitoring method based on the molecular detection of virus in excreta from trapped mosquitoes. We also succeeded at identifying the mosquito species community on several sampling sites, together with the vertebrate hosts on which they fed prior to being captured using amplicon‐based metabarcoding on mosquito excreta without processing any mosquitoes. Mosquito excreta‐based virus surveillance can complement standard surveillance methods because it is cost‐effective and does not require personnel with a strong background in entomology. This strategy can also be used to noninvasively explore the ecological network underlying arbovirus circulation.

1. INTRODUCTION

West Nile viruses (WNVs) are mosquito‐borne flaviviruses (from the Flaviviridae family, Flavivirus genus and belonging to the Japanese encephalitis [JE] serogroup) that circulate through several lineages into complex transmission cycles involving mosquitoes from the Culex genus and several vertebrate host species, with wild birds as amplifying hosts (Weissenböck et al., 2010). WNVs can cause severe and potentially fatal neurological disease in vertebrate hosts, including birds, equids and humans (Campbell et al., 2002). Native to Africa, WNVs from lineage 1 were first introduced in Europe presumably through migratory birds in the 1960s (Hubálek & Halouzka, 1999) where they caused mainly occasional self‐limited outbreaks and sporadic cases (Calistri et al., 2010) essentially restricted to rural areas. Major European outbreaks occurred in Camargue in 1962–1963 (Joubert et al., 1970) and Romania in 1996 (Tsai et al., 1998). The emergence of lineage 2 in southeastern Hungary in 2004 was associated with a significant upsurge of human and animal cases in several European countries. The largest outbreak was recorded in 2018 with 11 countries reporting 1548 locally acquired WNV infections, a number that exceeded the cumulative number of all reported infections between 2010 and 2017 (Bakonyi & Haussig, 2020). WNV lineage 1 and 2 are now circulating in endemic cycles in Europe (Calistri et al., 2010), causing disease incidence in human every year. Ecological interactions among bird species that led to WNVs introduction, amplification, and maintenance in local reservoirs, a prerequisite before the virus spillover in humans, are not yet fully understood (Jourdain, Toussaint, et al., 2007; Marzal et al., 2022).

Standard entomological surveillance provides the opportunity to monitor arboviruses circulation in their enzootic cycles based on the processing of thousands of mosquitoes that is expensive and time consuming. This surveillance method also poses logistical constraints and is thus not adapted to low infection rates in mosquito populations that typically occur in low or nonendemic areas. It has been demonstrated that mosquitoes infected with different arboviruses can excrete large amount of virus genomic RNA (Fontaine et al., 2016; Meyer et al., 2019; Ramírez et al., 2018). The detection of viruses in trapped mosquito excreta has been shown to be effective to monitor the presence of different mosquito‐borne pathogens in the field (Meyer et al., 2019; Minetti et al., 2020). We developed a 3D printed lodging that can be adapted on main standard mosquito traps that offer shelter and food for trapped mosquitoes and facilitate the collection of their excreta. Our xenomonitoring strategy succeeded at detecting two WNVs in excreta from the mosquito fauna sampled over a 2 month longitudinal study in summer 2020 in Camargue, a wetland area (Rhône delta) in the South of France. Using a single amplicon‐based metabarcoding approach, we succeeded to monitor the mosquito fauna diversity across different sampling sites and times in France and Mali without processing any mosquitoes. Importantly, we also succeed at identifying vertebrate hosts on which mosquitoes fed prior to be captured. Altogether, we showed that the use of mosquito excreta is efficient to trace arbovirus circulation at low cost and can give precious clues to characterize the ecosystem underlying arbovirus emergence.

2. MATERIALS AND METHODS

2.1. 3D‐printed BG sentinel mosquito trap modification

We created a 3D‐printed adapter (MX adapter) to increase mosquito longevity in BG sentinel (BGS, Biogents AG) traps, inspired from the device designed by Timmins et al. (2018). Briefly, our system is composed of a 120 mm high and 87/90 mm (inside/outside) diameter cylinder with a removable cap that is inserted into the depressurized BGS catch pipe. The MX adapter is attached beneath the intake funnel and is replacing the original catch bag. The original catch bag was cut at the bottom to house the MX adapter (Figure 1a). The MX adapter is traversed longitudinally by airflow piping to allow air depressurization inside the catch pipe, and thereby allowing aspiration at the top of the BGS intake funnel via the fan located underneath the system. The airflow piping is cut with its upward pointing section sealed with mesh fabric to prevent mosquito escape from the device while letting the air flow through. A mosquito feeder, filled with a cotton ball soaked in 10% sugar water, is attached to the inner side of the cylinder. The MX adapter has a removal bottom on which a filter paper (Whatman, grade 3, ref. 1003‐917) can be placed when the system is set in mosquito catching mode. A vertical sliding gate can be used to seal the system during transport when the bottom has been removed (optional; Figure 1b). The MX adapter provide a safe and moisturized enclosure for trapped mosquito with an easy access to sugar. Mosquito excreta can be easily collected on the filter paper at the bottom of the adapter. First version of BGS traps were used in this study but our device can be adapted to BG sentinel 2 and CDC light traps (Figure S1). The MX adapter was created on Fusion 360 (AutoDesk) and 3D printed in either PLA or PETG on a Sigmax R19 (BCN 3D). MX adapter 3D files (.stl format) are provided in File S3 under the Creative Commons (CC) licence BY‐NC‐SA.

FIGURE 1.

3D representations of the MX adapter designed to increase trapped mosquitoes' longevity and to collect their excreta for a virus surveillance purpose. (a) Schematic of the MX adapter inside the BG sentinel trap (BGS) in catching mode. (b) Different views of the adapter MX. All components are visible in the cross‐sectional and exploded views of the MX adapter: The cap, the mosquito feeder, the ring covered with mesh fabric, the cylinder, and the removable bottom on which can be placed a filter paper to collect mosquito excreta. In transport mode, the sliding gate covers the open slot in absence of the removable bottom. The MX adapter was created on fusion 360 (AutoDesk). MX adapter is under the creative commons (CC) licence BY‐NC‐SA (licensees may copy, distribute, display and make derivatives only for noncommercial purposes and by giving credits to the authors). All printable files are provided in File S3.

2.2. Study area and samples collection

The study was carried out in a ~130 km² area in Camargue (Figure 2), a large wetland in the Southeast of France with a temperate Mediterranean climate located inside the Rhone river delta. The Camargue is a nature reserve that hosts large populations of resident and migratory birds and is famous for its white horses and black bulls. West Nile virus transmission has been repeatedly reported in Camargue for several years (Bahuon et al., 2016; Murgue, Murri, Triki, et al., 2001). BGS traps were placed at the ground level at eight sites (one per locality) nearby the “Marais du Charnier”, each separated by few kilometres. Adult mosquitoes were captured over 6 weeks from 11 September 2020 to 23 October 2020, at the end of the mosquito season, using BGS modified with the MX adapter, as described previously. Mosquito captures were conducted over 3 to 4 consecutive days, with trap collection and reconditioning for a new mosquito trapping session occurring twice a week. Carbon dioxide was provided as a mosquito attractant for every BGS traps. Pressurized carbon dioxide bottles and 12 V power packs were used to operate over time. During trap collection, the MX adapter was removed from the BGS and maintained for 5 min on a sealed plastic box filled with dry ice to knock down mosquitoes. Knock down mosquitoes were then transferred to 50 ml tube and transported alive to the laboratory. This mosquito collection procedure avoids mosquitoes escaping from the device and was implemented late in our mosquito collection session (after the end of September) in response to mosquito escape issues coming from the fields. Filter paper covered with mosquito excreta were placed into annotated sealed plastic bags and transported at ambient temperature to the laboratory. Once in the laboratory, mosquitoes and filter papers were stored at −20°C until the extraction procedure. Adult mosquitoes were identified morphologically in the laboratory at the genus level and sorted by genus, trap, and collection date. One trap was implemented in Gao, Mali, from 26 August 2020 to 5 October 2020 in continuous operation with six samplings over this period. No mosquito was brought back to the laboratory and only mosquito excreta were analysed for these samples.

FIGURE 2.

Study map with the geolocalization of the eight sampling sites (a to h) in the Camargue wetland in south of France. The map was created using the free and open source QGIS geographic information system using satellite imagery from the ESRI.

2.3. RNA/DNA extraction

2.3.1. RNA/DNA extraction from filter papers with mosquito excreta

Virus detection was first conducted on mosquito excreta collected on filter papers. Briefly, filter papers were coiled and placed individually at the bottom of 14 ml plastic tubes (Falcon, ref: 352059) before being soaked in 3 ml of lysis buffer RAV1 (NucleoSpin 96 virus core kit, Macherey‐Nagel) for 5 min. Then, 10 μg of phage MS2 (RNA) was added to each tube as an internal extraction control. Filter papers were then manually grinded with a 5 ml pipette. Several 1 mm diameter holes were drilled in the transparent caps of the 14 ml tubes to create a colander, and then clipped on each tube. Closed tubes were placed upside down on a larger 50 ml tube (with the colander cap placed downward, at the bottom of the 50 ml tube) and centrifuged for 5 min at 1600 g. Flow throughs were collected and mixed with 3 ml 96%–100% ethanol before being loaded on NucleoSpin virus columns in several steps. RNA/DNA extraction was performed as indicated by the manufacturer's instructions. Eluates were kept at −20°C until use.

2.3.2. RNA/DNA extraction from captured mosquitoes

Virus detection in large number of samples is time consuming and cost prohibitive, especially when low infection rates are expected in mosquito populations. Here, we implemented a strategy based on the simultaneous grinding of 96 individual mosquitoes in a 96‐well plate inspired from Holleley and Sutcliffe (2009) MR4 protocol, followed by an extraction and detection in pool of 12 mosquitoes (one lane of a plate). This strategy drastically reduces labor and costs associated to virus screening while still providing the opportunity to sequence or isolate virus in single mosquito in a second intention, and to obtain accurate prevalence estimates by testing mosquitoes one by one in positive pools. Briefly, Culex mosquitoes were placed individually in each well of a shallow conical bottom 96‐well plate with 80 μl of buffer composed of 1.3 ml of proteinase K resuspended in proteinase buffer according to the manufacturer's instructions (NucleoSpin 96 virus core kit, Macherey‐Nagel) in 9 ml PBS. A male Ae. albopictus infected with the insect specific Aedes flavivirus (AEFV) was added on well position H12 of each plate as an internal extraction control. All mosquitoes from a plate were simultaneously grinded for 5 min by moving up and down and rocking vigorously a disposable bacterial colony replicator tool. Plates were then incubated at 70°C for 1 h with the bacterial colony replicator tool left on the top of the plate. After a short grinding step post‐proteinase K digestion, the replicator tool was discarded and 15 μl of each well from the same plate line were pooled. Plates with individually grinded mosquitoes were stored at −20°C. Then, 500 μl of AVL lysis buffer (QIAamp Viral RNA kit, Qiagen) was added to each pool with 5 μl of phage MS2 as an additional internal extraction control. After 10 min of incubation at room temperature, 500 μl of 100% ethanol was added to each pool and RNA was extracted according to the manufacturer's protocol (QIAamp Viral RNA kit, Qiagen).

2.4. Viruses and plasmodium spp. detection

Detection of WNV and USUV genomic RNA was performed with a one‐step reverse transcription quantitative polymerase chain reaction (RT‐qPCR) assay. The multiplexed RT‐qPCR was performed using the SuperScript one‐step RT‐PCR Platinum Taq Mastermix (Invitrogen), MS2 primers and probe (Ninove et al., 2011), or either WNV or USUV specific primers and probe (Table S1) using 10 μl of buffer 2×, 0.8 μl of forward and reverse primers, 0.3 μl of probe, 0.8 μl of enzyme and 2.3 μl of RNase‐free water for one reaction, with the following cycling protocol: 48°C for 30 min, 95°C for 10 min, 50 cycles at 95°C for 15 s, 60°C for 1 min. Amplification was performed on a CFX real‐time thermocycler (Bio‐Rad). A WNV from lineage 2 was used as a control. Concerning the detection of AEFV in infected male Ae. albopictus internal controls (extraction procedure for trapped mosquitoes in pools), virus genomic RNA was first reverse transcribed to complementary DNA (cDNA) with random hexamers using M‐MLV reverse transcriptase (Life Technologies, Inc.) according to the manufacturer's instructions. cDNA was amplified by 35 cycles of PCR using the corresponding set of primers described in Table S1 and amplicons were visualized by electrophoresis on a 1% agarose gel. Plasmodium spp. detection was performed using 10 μl of 2× SYBR Green PCR Master Mix (Thermofisher Scientific), 0.5 μl of forward and reverse 16 μM primers (Table S1), 4 μl of RNase‐free water and 5 μl of DNA extract for one reaction using the following cycling protocol: 95°C for 5 min, 45 cycles at 95°C for 15 s, 60°C for 1 min. Amplification was performed on a LightCycler (Roche Diagnostics).

2.5. Virus sequencing

We sequenced WNV genome based on the tiling amplicon‐based sequencing (PrimalSeq) method developed by Quick et al. (2017), and Grubaugh et al. (2019) and colleagues. Briefly, PrimalSeq protocol generates overlapping amplicons of ~400 base pairs from two multiplexed PCR reactions to generate sufficient templates for subsequent high‐throughput sequencing. We used a modified subset of multiplex primer scheme developed for WNV lineage 1 (Grubaugh et al., 2019) that covers the envelope gene with five overlapping amplicons with a set of primers degenerated to amplify WNV lineages 1 and 2. Illumina Nextera universal tail sequences were added to the 5′ end of each of these primers to facilitate the library preparation by a two‐step PCR approach (WNV lin.1 and 2, Table S1). Purified RNAs from excreta samples positive for WNV in RT‐qPCR were first reverse transcribed to complementary DNAs (cDNAs) with random hexamers using M‐MLV reverse transcriptase (Life Technologies) according to the manufacturer's instructions. For the multiplex PCR reactions, 5 μl of cDNA was used in a 20 μl reaction mixture made of 5 μl of Hot START 5X BIOamp DNA polymerase mix (Biofidal), 4 μl of forward and reverse primer mix at 10 μM, and 11 μl of water. The thermal programme was: 10 min of polymerase activation at 96°C followed by 35 cycles of (i) 30 s denaturing at 96°C, (ii) 30 s annealing at 62°C and (iii) 1 min extension at 72°C, followed by a final incubation step at 72°C for 7 min to complete synthesis of all PCR products. A 15 cycle PCR was then performed using Nextera Index Kit – PCR primers, that adds the P5 and P7 termini that bind to the flow cell and the dual 8 bp index tags. Resulting amplicons were purified with magnetic beads (SPRIselect, Beckman Coulter), quantified by fluorometric quantification (QuantiFluor dsDNA System, Promega) and visualized on QIAxcel capillary electrophoresis system (Qiagen). Libraries were sequenced on a MiSeq run (Illumina) using MiSeq version 3 chemistry with 300 bp paired‐end sequencing.

After demultiplexing, trimmomatic version 0.33 was used to discard reads shorter than 32 nucleotides, filter out Illumina adaptor sequences, remove leading and trailing low‐quality bases and trim reads when the average quality per base dropped below 15 on a 4‐base‐wide sliding window. Reads were aligned to the WNV lineage 1 reference genome (RefSeq entry: NC_009942) with bowtie2 version 2.1.0 (Langmead & Salzberg, 2012). The alignment file was converted, sorted and indexed using Samtools version 0.1.19 (Li et al., 2009). Coverage and sequencing depth were assessed using bedtools version 2.17.0. The consensus sequence was generated with Ivar (Grubaugh et al., 2019) with default options.

2.6. Phylogenetic analyses

A background set of 36 full‐length WNV genome sequences across different lineages was obtained from GenBank. One full‐length genome sequence of Japanese encephalitis virus was also downloaded to be used as an outgroup in the phylogenetic tree. Genome sequences were aligned using Clustal Omega and curated by gblocks software implemented in the seaview version 5.0.4 interface (Gouy et al., 2021) by allowing gap positions within the final block and less strict flanking positions. The best‐scoring maximum‐likelihood (ML) tree was then generated with 100 bootstrap replicates with phyml (Guindon et al., 2005). The GTR + I + G nucleotide substitution model was chosen based on the lowest Akaike information criterion (AIC) value using the Smart Model Selection (SMS) in phyml software (Lefort et al., 2017). Phylogenetic trees were visualized using the ggtree r package (Yu, 2020). All sequences were trimmed to the length of our sequenced amplicon (341 bp) before being imported into the popart program (Leigh & Bryant, 2015) to create a TCS haplotype network (Clement et al., 2000).

2.7. Amplicon‐based metabarcoding

The amplicon‐based metabarcoding method implemented here was performed on mosquito excreta eluates post NucleoSpin Virus RNA/DNA extraction. Even though the NucleoSpin Virus Columns kit is optimized to extract virus RNA/DNA, the eluate also contained nonviral RNA/DNA. Virus detection, sequencing and metabarcoding approaches were performed on the same eluates to facilitate the procedure. A ~460 bp mitochondrial DNA section corresponding to a subfragment of the classical Folmer cytochrome c oxidase subunit I (COI) fragments (Folmer et al., 1994), was amplified with universal primers BF2/BR2 (Table S1) that are routinely used for macroinvertebrate monitoring (Hajibabaei et al., 2019; Mechai et al., 2021). Illumina Nextera universal tails sequences were added to the 5′ end of each of these primers to facilitate library preparation, as described above for the WNV tiling amplicon‐based sequencing procedure. PCR amplifications, library preparation and sequencing were made according to the same protocol described above for WNV sequencing excepting for the annealing temperature that was set to 50°C for 30 s. All raw sequences have been deposited in the NCBI database under the NCBI BioProject number PRJNA768434.

2.8. Taxonomic identification and diversity indices

Demultiplexed fastq sequences were imported to qiime2 version 2021.4 for bioinformatic analyses. The qiime2‐dada2 pipeline (Callahan et al., 2016) was used for turning paired‐end fastq files into merged reads, filtering out Illumina adapters, denoising and removal of chimeras and filtering out replicates (File S4). Taxonomic assignment was first carried out for the amplicon sequence variants (ASVs) using the qiime2‐feature‐classifier classify‐consensus‐blast plugin using a database of 1,176,764 sequences gathering Fungi, Protist, and Animal COI records, recovered from the Barcode of Life Database Systems 7 March 2021. The database was built according to the step‐by‐step tutorial provided by O'Rourke et al. (2020). Only dereplicated sequences within our primer coordinates were retained in the final database. A percentage identity threshold of 98% over a query alignment coverage of 90% was used to assign a taxonomy to an ASV. A consensus taxonomy was assigned to an ASV by determining the taxonomic lineage on which at least 55% (six out of the 10 top blast hits) of the aligned sequences agreed. Phylogenetic distances between species from the Culicidae family and Chordata phylum were calculated with IQ‐TREE (Nguyen et al., 2015) with the ultrafast bootstrap algorithm and automatic model selection, as implemented in the qiime2 phylogeny iqtree‐ultrafast‐bootstrap plugin. Taxonomy assignment was performed using a Poisson tree processes (PTP) model based on nonultrametric phylogenetic trees inferred with IQ‐TREE. This similarity threshold‐free method can delimitate species using number of substitutions to identify branching rate transition points with the assumption that the number of substitutions between species is significantly higher than the number of substitutions within species (Zhang et al., 2013). Differential relative ASV abundances between sampling sites were compared on a nonrarefied data set using the q2‐aldex2 differential abundance package, by modelling the data as a log‐ratio transformed probability distribution rather than as counts (Fernandes et al., 2014). The R package phyloseq (McMurdie & Holmes, 2013) and seaview version 5.0.4 (Gouy et al., 2021) were used to represent phylogenetic relationship among ASVs assigned to Culex pipiens mosquitoes and 12 additional reference sequences of Culex quinquefasciatus and Culex pipiens taken randomly from the database (File S4). The GTR + G nucleotide substitution model was chosen based on the lowest AIC value using SMS in phyml software (Lefort et al., 2017). All further diversity metrics implemented in qiime2 can be accessed by running qiime2 diversity commands on data provided in File S4.

Matching coefficient of Sokal and Michener index (symmetrical) and Jaccard index (asymmetrical) were used to compare the Culicidae fauna composition revealed based on trapped mosquitoes identified morphologically and amplicon‐based metabarcoding on the corresponding mosquito excreta for all sample collections from Camargue at the genus level. Considering a contingency table of binary data such as n11 = a, n10 = b, n01 = c and n00 = d (0 and 1 referring to the species absence or presence in each observation group so that n11 means that the species was identified both morphologically and in amplicon‐based metabarcoding), the Sokal and Michener index was computed as follow: s = (a + d)/(a + b + c + d), and Jaccard index as follow: s = a/(a + b + c). The latter is an asymmetrical index that does not treat double zero (d) in the same way as double presences (a) as a reason to consider samples similar.

All statistical analyses were performed in the statistical environment R. Figures were made using the package ggplot2 (Wickham, 2016). The Map was created using the Free and Open Source QGIS Geographic Information System using satellite imagery from the Environmental Systems Research Institute (ESRI).

3. RESULTS

3.1. Virus screening in mosquito excreta

A total of 86 excreta samples were analysed in this study (80 excreta samples from Camargue, France and 6 from Gao, Mali). These excreta were collected at the end of summer 2020 from trapped mosquitoes contained into the 3D printed MX adapter that has been specifically designed to increase trapped mosquitoes' survival and to facilitate the recovery of their excreta over several days (Figure 1). A total of 6845 mosquitoes were recovered from traps in Camargue, comprising 2819 Culex spp., 3922 Ochlerotatus/Aedes spp., 91 Anopheles spp. and 13 Culiseta spp. mosquitoes (File S1). Anopheles spp. were morphologically identified as An. hyrcanus sensu lato and as species from the An. maculipennis complex (which includes An. atroparvus). Specimens not belonging to the Culicidae family were also collected in traps but were not considered in this study. Trapped fauna from Mali were not brought back to the Marseille laboratory for analysis. WNV genome was detected by one‐step reverse transcription quantitative polymerase chain reaction (RT‐qPCR) on excreta samples from two traps (2.3%) in Camargue, collected on location F on 15 September 2020 (cycle threshold [Ct] of 31) and on location G on 22 September 2020 (cycle threshold [Ct] of 33; Figure 2). No excreta samples were found positive for WNV in Mali, nor Usutu virus (USUV) in both Camargue and Mali. Excreta samples from Mali also tested negative for Plasmodium spp., the causative agents of malaria in Mali.

In Camargue wetland, a total of one Culex, 17 Ochlerotatus/Aedes and one Anopheles mosquitoes were collected in the first positive trap (F 15 September 2020) while five Culex, 31 Ochlerotatus/Aedes and one Anopheles mosquitoes were identified in the second trap (G 22 September 2020). No WNV viruses were detected in these mosquitoes by RT‐qPCR after an individual extraction. The amount of excreta recovered on these traps was higher than it would be expected by considering that they would be expelled by these mosquitoes only, suggesting that a mosquito escape issue could have occurred for these traps. Culex and Culiseta mosquitoes from traps tested as negative for WNV and USUV based on excreta, were screened for these viruses based on an individual grinding and a pooled extraction and detection procedure. These viruses were not detected in any of the 241 pools, each made of 12 mosquitoes.

We succeeded at sequencing a 341 bp fragment of the WNV genome (GenBank accession no. OK489805) located on the envelope gene in one excreta sample that tested positive by RT‐qPCR with the lowest Ct (Ct of 31, sample F 15 September 2020). Phylogenetic analysis revealed that the WNV identified in our study belonged to lineage 1 and was closely genetically related to a WNV isolated from a symptomatic horse on 3 October 2015 (Beck et al., 2020; GenBank accession no. MT863559; Figure 3a). Our WNV small genomic section diverged from the 2015 WNV (GenBank accession no. MT863559) by five single nucleotide polymorphisms (SNPs) and from a WNV isolated in Spain on a horse in 2010 (Sotelo et al., 2011; GenBank accession no. JF719069) by 4 SNPs (Figure 3b).

FIGURE 3.

Molecular relationship at the intraspecific level of West Nile virus (WNV) genomes from lineages 1, 2 and 3, with the virus identified in this study in mosquito excreta. (a) Molecular phylogenetic tree of 37 WNV genomes, including the 341 bp genome section identified in this work. The phylogenetic analysis was performed on a curated alignment of 10,302 bp by allowing gap positions within the final block. One sequence of JEV (AF075723) was used as an outgroup in the phylogenetic tree. The evolutionary history was inferred using a maximum likelihood method using phyml. Bootstrap values obtained after 100 replicates are shown at major nodes (percentage of replicate trees in which the associated taxa clustered together). Evolutionary distances, as represented by length of branches, are expressed in number of base substitutions per site. (b) Haplotype network inferred by the TCS method using the same sequences as above trimmed to the length of our sequenced amplicon (341 bp). The size of each circle represents the frequencies of the haplotype. Mutations are shown as perpendicular bars along the branches and grey small circles represent inferred unsampled haplotypes. The WNV genome section identified in this study is indicated in red.

3.2. Amplicon‐based metabarcoding analysis on trapped mosquito excreta

A fragment of ~460 bp corresponding to a fragment of the cytochrome c oxidase subunit I (COI) was amplified from the DNA mixture extracted from mosquito excreta. This corresponds to a subfragment of the classical Folmer cytochrome c oxidase subunit I (COI) fragment (Folmer et al., 1994) routinely used in metabarcoding analysis. The sequencing generated a total of 4,300,448 demultiplexed sequences across all samples with a mean of 44,796 (first quartile: 41,019; third quartile: 51,171) reads per sample. All paired‐end reads were turned into chimera‐free ASVs after a trimming, quality‐based filtering, merging, dereplicating and denoising steps (Callahan et al., 2016, p. 2). The total number of reads was reduced to 2536 ASVs with a median abundance of 28,381 (first quartile: 23,712; third quartile: 33,434) per sample. Taxonomy was next assigned to each ASV using blast on a database gathering Fungi, Protist, and animal COI records using stringent search parameters. A total of 467 (7.14%), 18 (0.27%), and eight (0.12%) ASVs were assigned to the Animalia, Fungi, and Protist kingdoms, respectively, and 6043 ASVs (92.46%) were unassigned. Unassigned ASVs corresponded to assignments falling below the required 98% identity threshold. After discarding unassigned ASVs, 97% of the total ASVs count were assigned to the Animalia kingdom (File S2). Fungi from the genera Rhodotorula (phylum: Basidiomycota) and Neurospora (class: Sordariomycetes, phylum: Ascomycota) were overrepresented, with 85 and 12% of the total ASV count in the Fungi kingdoms, respectively. The Arthropoda phylum was the most represented inside the Animalia kingdom (93% of ASV abundance) with most ASV counts (98%) assigned to the Insecta class (File S2).

3.3. Culicidae species composition based on trapped mosquito excreta

Both Arachnida and Insecta classes were represented in the pool of DNA recovered inside the adapter MX. The Diptera order represented 91% of ASV abundance inside the Insecta class and taxon from the Culicidae family represented 60% of ASV counts inside the Diptera order (File S2). This was consistent with the fact that mosquitoes (Culicidae) were not the only specimen recovered from the traps. ASVs could be assigned at the species level for 12 mosquito species, that is, Uranotaenia unguiculata (one ASV), Ochlerotatus detritus (six ASVs), Ochlerotatus caspius (72 ASVs), Culiseta subochrea (three ASVs), Culex theileri (seven ASVs), Cx. pipiens sensu lato (eight ASVs), Cx. modestus (21 ASVs), Anopheles melanoon (one ASV), Anopheles hyrcanus (eight features), Aedes vexans (five ASV), Ae. albopictus (two features), and Ae. aegypti (12 features; Figure 4a). Two ASVs were grouped into the genus Anopheles. The first (ASV: 6a7ed1f43df048aa35269038d8717433) was phylogenetically placed with Anopheles melanoon and was only present in Camargue (Figures 4 and S2). The second (ASV: d082c59e9821021f7047c16bc6536b35) was identified as a distinct species using a PTP model (Figure S2A). This ASV was only detected in Mali and shared 100% identity over its full length with An. gambiae sensu lato (BLASTN algorithm using the GenBank nonredundant nucleotide collection, 06/27/2022, NCBI). The ASV assigned to the Cx. pipiens s.l. species (ASV: e8c12e96fda22f7f88badd812979f9cc) was significatively associated to community composition differences between Camargue and Mali, accounting for the differential abundance of features between these study areas (Aldex2, p = .0002, expected p‐value of Welch's t test, Figure S3A). Features assigned to Ae. albopictus and Ae. aegypti were detected in Camargue. These species were not identified in trapped mosquitoes in this study area and are not likely to occur in Camargue, suggesting spurious ASV assignments for these species or contamination from the laboratory.

FIGURE 4.

Amplicon‐based metabarcoding analysis of the Culicidae diversity through the sequencing of trapped mosquito excreta. Heatmap representing the total number of ASVs (log10 scale) attributed to a taxon according to sampling sites and sampling times. The number of ASVs assigned to each taxon is represented with horizontal black bars at the right‐hand side of the graph for Camargue and Mali study sites.

An amplicon‐based genomic approach was further used to assess the genetic diversity inside the Cx. pipiens s.l. mosquito species complex. We revealed several Cx. pipiens s.l. cytochrome oxidase c subunit I (COI) haplotypes circulating in Camargue (Figures S2 and S3B). It was not possible to discriminate Cx. pipiens and Cx. quinquefasciatus species in the Cx. pipiens s.l. complex based on our COI sequence. The intra Cx. pipiens s.l. complex genetic diversity was not obviously associated to Cx. pipiens pipiens and Cx. pipiens molestus cryptic forms, as revealed by the tree topology incorporating representative sequences from all forms (Figure S3B). The haplotype (ASV: e8c12e96fda22f7f88badd812979f9cc) over‐represented in the Mali study site was, however, closed to Cx. quinquefasciatus sequences on the tree topology.

Mosquito species composition revealed on DNA present in mosquito excreta (amplicon‐based metabarcoding) was compared with the species composition revealed by morphological identification of trapped mosquitoes at the genus level for all sample collections from Camargue. The Aedes and Ochlerotatus genera were grouped for simplicity in this analysis. Mosquito species composition shared 70% similarity between both identification methods when considering the symmetrical Sokal & Michener similarity index, with 118 (35.7%) and 114 (34.5%) matches for the presence revealed by both methods and absence revealed by both methods, respectively. A total of 43 (13%) and 55 (16.7%) mismatches was revealed for the presence in metabarcoding/absence in trapped mosquitoes and absence in metabarcoding/presence in trapped mosquitoes, respectively. The similarity dropped to 54.6% when using the asymmetric Jaccard index that does not consider species absences in both methods. Importantly, the Jaccard similarity index was much lower when considering under‐represented mosquito species; that is, Anopheles (Jaccard index: 20%) or Culiseta (Jaccard index: 7%) genera, as compared to the highly represented Aedes/Ochlerotatus (Jaccard index: 67%) or Culex (Jaccard index: 73%) genera.

3.4. Chordata species community revealed by amplicon‐based metabarcoding based on trapped mosquito excreta

The Chordata phylum represented 7% of ASV abundance inside the Animalia kingdom, with most ASVs assigned to the Homo sapiens species (83% of Chordata), followed by species from the Artiodactyla order (Suidae [pigs] and Bovidae, 10% of Chordata) and birds (2% of Chordata). At the qualitative level, Homo sapiens, Rattus norvegicus and pigs (Sus scrofa) species were identified in both Camargue and Mali study sites. ASVs assigned to a frog species (Hoplobatrachus occipitalis) was uniquely identified in Mali. Four fish species were also detected in this study. ASVs assigned to these species of fish (class: Actinopterygii) were scarce (2% of the Chordata phylum) with Salmo salar (salmon) being the most abundant and present in three study sites in both Camargue and Mali. Less than 10 ASV counts were observed for all fish species excepting for Salmo salar in Mali with 88 and 193 counts for the 26 August and 02 September sampling dates, respectively. An overall lower intraspecies diversity was observed for invertebrates as compared to mosquitoes (Figure S2B).

Vertebrates typically associated to the Camargue regions were detected, such as Bos taurus, Equus or passerine birds. While accounting for only 0.2% of ASV counts inside the Animalia kingdom, ASVs assigned to the Aves class were the most diverse inside the Chordata phylum with six birds identified as the species level (i.e., Sylvia atricapilla, Motacilla flava, Hirundo rustica, Cettia cetti and Pica pica from the Passeriforms order, Gallus gallus; Figure 5).

FIGURE 5.

Vertebrate (Chordata) diversity identified based on digested blood through the sequencing of trapped mosquito excreta. Heatmap representing the total number of features (log10 scale) attributed to a taxon according to sampling sites and sampling times. The number of features assigned to each taxon is represented with horizontal black bars at the right‐hand side of the graph for Camargue and Mali study sites. *Samples might have been exposed to human DNA all along the procedure. Presence of ASVs assigned to human might be interpreted with caution because they might arise from contamination with human DNA. We also cannot totally rule out a possibility of DNA contamination concerning domestic species widely used in food industry (i.e., fish, pork, and beef).

4. DISCUSSION

WNV surveillance is usually achieved through the monitoring of symptomatic cases in human and animals (Autorino et al., 2002; Nagy et al., 2019), seroprevalence studies (Faggioni et al., 2018), dead birds (Ziegler et al., 2015), blood donor (O'Rourke et al., 2020) or virus screening in trapped mosquitoes (Calzolari et al., 2020). High rates of asymptomatic arbovirus infections, the broad immunological cross‐reactivity across flaviviruses, difficulties in reporting and collecting dead birds (Eidson et al., 2001), and low arbovirus infection prevalence in mosquitoes are hindering the use of these methods for the early detection of arbovirus circulation with an accurate dating of infection events. The discovery that mosquitoes with a systemic arbovirus infection can massively excrete virus RNA (Fontaine et al., 2016; Ramírez et al., 2018) has opened new perspectives in entomological surveillance of viruses (Meyer et al., 2019). Indeed, virus identification in trapped mosquito excreta considerably alleviates the processing charge and costs associated with entomological surveillance by extending the period between mosquito trap collections and making optional the steps of mosquito species identification, pooling, and testing. This noninvasive method that does not require strong entomological knowledge can also be extended to monitor mosquito‐borne parasites (Pilotte et al., 2016; Ramírez, van den Hurk, et al., 2019).

Here, we succeeded at identifying WNV circulation in Camargue over a short 6‐week period without the need to process any mosquitoes. This is only the third time that WNV have been detected in mosquitoes in this region since 1964 (Beck et al., 2020; Murgue, Murri, Zientara, et al., 2001; Rageau & Mouchet, 1967) despite important efforts to search for infected vectors, including after epidemics. The sequencing of a short genomic region on the envelope gene indicated its affiliation to WNVs from lineage 1 (Western Mediterranean clade) that circulated in Southern Europe in the last 20 years (Beck et al., 2020). These viruses can be maintained locally from 1 year to another in enzootic cycles that involve resident bird species and ornithophilic mosquitoes by overwintering in hibernating infected adult mosquitoes (Nasci et al., 2001; Rudolf et al., 2017) or in bird species with chronic infection (Wheeler et al., 2012). Another hypothesis would be recurrent introductions through migratory birds that would carry the viruses northward during spring migration from African wintering places where these viruses are endemic, or from West Europe in late summer when birds return to Africa. Unfortunately, no infected mosquitoes were recovered from the traps positive for WNV based on detection on mosquito excreta due to mosquito escape issues that occurred at the beginning of this study. Complete genome virus sequencing and WNV infection prevalence would have been easier to achieve by only processing mosquitoes in traps where viral RNA was detected in their excreta. More viral RNA genomes can be expected to be recovered in mosquitoes as compared to their excreta, which would increase the success of whole genome sequencing. In case of a mosquito escape due to a trap failure or a faulty handling, arbovirus RNA and trapped mosquito DNA can still be recovered in the excreta with our system. For instance, we could assess the mosquito species diversity and composition in Mali without an access to the trapped mosquitoes. WNV RNA has been shown to be well preserved during a prolonged (14 days) mosquito excreta storage on Flinders Associate Technologies (FTA) nucleic acid preservation cards (Ramírez, Hall‐Mendelin, et al., 2019). It has also been demonstrated that FTA and untreated filter paper can bind and preserve virus RNA equally over 28 days at 23°C (Hall‐Mendelin et al., 2010). Mosquito excreta collected on filter papers are thus also easy to transport and store from a remote field site to the laboratory.

We succeeded in having a global view of the mosquito species composition and abundance in different sampling sites. Mosquito species emblematic to Camargue such as Ochlerotatus caspius and Culex pipiens were identified via DNA contained in their excreta in Camargue. A low similarity index was observed when comparing under‐represented mosquito species between metabarcoding on mosquito excreta and directly on trapped mosquito. This might be explained by mosquito escape issues and absence of excreta from freshly trapped mosquitoes. Species composition data obtained directly from the trapped mosquito community would always outperform metabarcoding data obtained from their excreta. However, excreta‐based method alleviates logistical constraints and the need of trained operators in the field. It can allow a “bottom‐up” approach, where nontrained volunteers can collect and easily send filter‐papers impregnated with mosquito excreta to a remote laboratory. This can complement standard “top‐down” approaches, where trained operators are sent into the field to conduct the survey, to extend a study in space and/or time at low costs. Our COI haplotypes could not discriminate species from the Cx. pipiens s.l. complex which comprises Cx. pipiens quinquefasciatus and Cx. pipiens pipiens. Cx. p. pipiens further has two recognized forms—pipiens and molestus—that exhibit behavioural and physiological differences. Culex pipiens pipiens feed preferentially on birds, and Cx. pipiens molestus on humans. All these species can form hybrids and transmit WNV (Andreadis, 2012; Mattingly et al., 2009; Yurchenko et al., 2020). Our method revealed genetic diversity inside the species complex level but could not discriminate these species based on our COI haplotypes. Failure to discriminate these forms inside the complex using mitochondrial DNA has previously been described (Gunay et al., 2015; Tahir et al., 2016). A distinct haplotype was, however, significantly overrepresented in Mali and can probably be attributed to the Cx. quinquefasciatus species. Additional nuclear genomic regions could be targeted in the metabarcoding scheme to further discriminate these species inside the complex (Yurchenko et al., 2020) or to noninvasively screen for known molecular signatures (e.g., insecticide resistance).

Taxonomic identification is limited with our method to the genetic diversity that can be captured and amplify over a short mitochondrial sequence. Here, taxonomic identifications are biased toward arthropods because we used a degenerated primer set specifically designed to amplify arthropods (Mechai et al., 2021). The use of primers targeting the mitochondrial DNA of vertebrates on the same target might further increase our ability to identify the diversity of vertebrate species on which mosquitoes fed prior to being trapped (Reeves et al., 2018). In addition, it has been demonstrated that the success of host species identification decreases with the advanced stage of mosquito blood meal digestion (Martínez‐de la Puente et al., 2013). DNA degradation induced by the digestion process and exposure to the environment might affect the detection probability of host species in the excreta. These limitations might have explained the spurious or unexpected species identifications in this study. Aedes albopictus and Ae. aegypti COI sequences might share a strong homology on our target region with other Aedes species in Camargue that are not present in our database. The identification of fish species in this study might have resulted from any DNA contamination from the environment (e.g., food contamination), laboratory reagents or kits. ASVs assigned to these fish species were scarce and filtering out low‐abundant and/or low‐prevalent features might improve false‐identifications, with the trade‐off of losing species identification. Further studies would be needed to assess if species misassignments are associated to mosquito excreta‐based meta‐barcoding. Advancing sequencing technologies might soon allow to sequence at high quality larger DNA regions, thereby improving taxonomic assignments.

Potential WNV bird reservoir species were also noninvasively identified on trapped mosquito excreta. Determining the reservoir role of local bird species for WNV is not straightforward due to the challenging task of obtaining blood samples from animal for diagnoses purposes. It usually involves wild bird captures that are difficult to operationalize or blood‐meal analysis of trapped mosquitoes prior to implement experimental infection studies (Grubaugh et al., 2015). Blood‐fed mosquitoes can be accidentally captured in CO₂ baited traps or be trapped during their quest to complete their blood‐meals. It is also possible to specifically target blood‐engorged mosquitoes in their resting sites, but this method implies an active prospection and is difficult to implement for exophilic species. Our method captures vertebrate blood as it is incrementally digested by trapped mosquitoes without the need to process mosquitoes during their digestion stage. Blackcaps (Sylvia atricapilla), Barn swallows (Hirundo rustica) and Yellow wagtails (Motacilla flava) identified in this study were previously listed as bird species potentially involved in the introduction, amplification and spread of WNV in Camargue (Jourdain, Toussaint, et al., 2007). They are all bird species present in Camargue during the mosquito season that migrate in West Africa during winter. WNV antibodies have been detected in most of these species (Buckley et al., 2003). WNV has also been isolated from common magpies (Pica pica; Jourdain, Schuffenecker, et al., 2007) in 2004 in Camargue. Corvidae are sedentary birds that are potential hosts to amplify the virus and bring it closer to the urban areas. Our method cannot directly link a virus collected in trapped mosquitoes to a vertebrate host. However, it can reveal trophic preferences and the range of potential hosts involved in WNV introduction and amplification in different areas. Ecological factors associated with WNV outbreaks are not yet fully understood. Increased host diversity can either reduce the risk of virus emergence by diluting the abundance of a vector or a reservoir host inside a community (Ezenwa et al., 2006; Swaddle & Calos, 2008) or increase the risk if the abundance of a vector is a function of vertebrate host diversity (Dobson, 2004; Ferraguti et al., 2021). Extending the entomological survey over several diverse locations and seasons might help to assess if bird or mosquito species diversity and composition can influence the risk of WNV emergence.

Costs and logistic constraints are impeding the implementation of nationwide entomological surveillance programmes worldwide. Here, we used both the RNA and DNA genomic materials contained into excreta from trapped mosquitoes to (i) noninvasively survey the emergence and spread of WNV, (ii) monitor mosquito species diversity without the processing of mosquitoes and (iii) to invade ecological network underlying WNV emergence by identifying vertebrate animals on which trapped mosquitoes fed before to be captured, using a single and easy to implement amplicon‐based metabarcoding procedure. Trapped mosquito excreta analysis can be considered as a complementary cost‐effective strategy to detect the circulation of mosquito‐borne pathogens affecting human health or to noninvasively survey the presence of these pathogens in remote areas without health‐care system.

AUTHOR CONTRIBUTIONS

Grégory L'Ambert, Mathieu Gendrot, Agnès Nguyen and Albin Fontaine designed research. Grégory L'Ambert, Sylvain Pages and Hélène Savini contributed to the sample collection on the field. Albin Fontaine, Sébastien Briolant, Nicolas Gomez performed research. Laurent Bosio, Vincent Palomo, Nicolas Benoit, Bruno Pradines, Guillaume André Durand, Isabelle Leparc‐Goffart and Gilda Grard performed pathogens detection and contributed analytic tools. Albin Fontaine analysed data. Grégory L'Ambert and Albin Fontaine wrote the manuscript with input from all authors.

FUNDING INFORMATION

This study received funding from the Direction Générale de l'Armement (grant no. PDH‐2‐NRBC‐2‐B‐2113) and from the Direction de la Formation de la Recherche et de l'Innovation (Grant MX, DFRI). The contents of this publication are the sole responsibility of the authors. The funders had no role in study design, data collection, and interpretation, or the decision to submit the work for publication.

CONFLICT OF INTEREST

The authors declare that there is no conflict of interest regarding the publication of this article.

OPEN RESEARCH BADGES

This article has earned an Open Data badge for making publicly available the digitally‐shareable data necessary to reproduce the reported results. Data are available under the NCBI BioProject number PRJNA768434, GenBank accession number OK489805 and Dryad at DOI: https://doi.org/10.5061/dryad.37pvmcvph

BENEFIT‐SHARING STATEMENT

All data have been shared with the broader public via appropriate biological databases as described above.

Supporting information

Figures S1–S3

Table S1

File S1

File S2

File S3

File S4

L’Ambert, G. , Gendrot, M. , Briolant, S. , Nguyen, A. , Pages, S. , Bosio, L. , Palomo, V. , Gomez, N. , Benoit, N. , Savini, H. , Pradines, B. , Durand, G. A. , Leparc‐Goffart, I. , Grard, G. , & Fontaine, A. (2023). Analysis of trapped mosquito excreta as a noninvasive method to reveal biodiversity and arbovirus circulation. Molecular Ecology Resources, 23, 410–423. 10.1111/1755-0998.13716

DATA AVAILABILITY STATEMENT

Amplicon‐based metabarcoding raw sequencing data are accessible under the NCBI BioProject number PRJNA768434. The WNV genomic section sequenced in this project is accessible under the GenBank accession number OK489805. qiime2 artefacts with all amplicon‐based data are available in File S4.