Insect traps and slides were deployed in wooded areas at one urban and one rural site located <15 km apart in each of three Canadian cities: Sault Ste. Marie, Ontario; Fredericton, New Brunswick; and Halifax, Nova Scotia. Within each city, urban sites were located relatively close (< 2 km) to industrial parks whereas rural sites were located 6–15 km distant from these areas. At each of the six sites, we set up two insect traps and two slide collectors in a paired design with traps and slides < 5m apart within pairs and replicate pairs separated by about 30 m. Preservative fluids and slides were collected weekly or biweekly (3−4 times) from 1–28 August 2018 in New Brunswick and Nova Scotia, and weekly from 16 August–13 September 2018 in Ontario. The samples were shipped with icepacks and kept frozen at −20°C until extraction. A total of 46 slides and 44 trap preservative fluids were collected.
Black Lindgren 12-funnel traps (Synergy Semiochemical Corp., Burnaby, BC, Canada) were suspended from rope tied between two trees such that the trap was at least 1m from the nearest tree and the collection cup 30–50 cm above the ground. Traps were treated with 50% Fluon diluted in water to increase efficiency of capturing wood boring beetles (Allison et al., 2016). The collecting cups contained commercial RV antifreeze as preservative fluid (WinterProof Water System Antifreeze, 10–30% ethanol, 1–5% propylene glycol, Recochem Inc., Montreal, QC, Canada) with a drop or two of liquid dish detergent to reduce surface tension in the NB and NS sites. Each funnel trap was baited with four semiochemical lures known to increase captures of several species of bark and wood boring beetles in the families Cerambycidae and Curculionidae (subfamily Scolytinae), i.e., monochamol, ipsenol, alpha pinene and ethanol (Allison et al. 2003; Miller et al. 2016, Boone et al. 2019; Flaherty et al. 2019).
A new model of aerial spore collector was developed using technology from the 1960’s for its simplicity to fabricate and use by non-scientists. It consists of Grade 50 double layer cheesecloth (Uline, Pleasant Prairie, WI, USA) impregnated with silicone oil #378399 (Sigma) at 60 g per m2 mounted in 35 mm projector slides. The slides are held in place by an alligator clip inside a 25.4 cm diam. airport wind indicator (Airport Windsock Corporation, Lake City, MN, USA). The slide was positioned 2 m above ground.
Preservative fluids from each trap collection (approximatively 250 ml) from Ontario were filtered on 25 mm diam, 2.7 μm pore size Glass Microfiber GF-D (Whatman, Buckinghamshire, UK) using vacuum, whereas Nova Scotia and New Brunswick fluids were filtered on 90 mm diam., Whatman #1 qualitative filter papers 11 μm pore size (Whatman, Maidstone, UK). Filters and cheesecloth were sampled using a 6 mm paper punch. One 6 mm disk per filter and cheesecloth slide was ground using a Christison M3 Mixermill with a tungsten bead, twice for 2 min at 30 hertz with extraction buffer. Extraction was done with the Plant DNeasy mini kit (Qiagen Inc., Valencia, CA, USA) according to manufacturer instructions. One microliter of the eluate was used as genomic DNA (gDNA) template for Polymerase Chain Reaction (PCR).
In order to obtain a comprehensive data set of all fungal DNA present in our samples, a High Throughput Sequencing (HTS) method based on the Illumina Miseq sequencing system was used. DNA amplification, primer constructs, purification and sequencing was done as described in Bérubé et al. (2018). The primers ITS1F and ITS7G were used to amplify the ITS regions of the ribosomal DNA fragment (ITS1- 5.8S) for metabarcoding of fungi present on filters and slides.
Each of the 90 samples was tagged with differing indices, PCR amplified separately and then tagged amplicons were pooled in equimolar amounts of 4 ng DNA per sample. Final quantification, primer dimer removal and amplicon quality check were done with an Agilent 2100 BioAnalyzer (Agilent Technologies, Santa Clara, CA, USA). Pooled DNA samples were sent to the Next- Generation Sequencing Platform, Genomics Centre, CHU de Québec- Université Laval Research Centre, Quebec City, QC, Canada, which performed paired-end 300 bp sequencing using MiSeq Reagent Kit v3 (600- cycles) through an Illumina MiSeq system.
A bioinformatics treatment of HTS DNA sequences was executed to create Operational Taxonomic Units (OTUs), our proxies for fungal species (Huse et al. 2010; Kunin et al. 2010). Sequence analysis was done as described in Bérubé et al. (2018). In order to minimize the loss of rare and targeted fungal species, standard bioinformatic tools like QIIME2 and DADA2 were not used. Instead a custom bioinformatic pipeline that minimizes losses of singletons and rare OTUs was used (Gagné et al. 2020). Sequence sets were organized into clusters with USEARCH 64 bit v8.0.1623 with a sequence similarity threshold of 97% to agglomerate reads and form the OTUs, the most abundant sequence types serving as cluster seeds. As no single similarity threshold will accurately reflect the species level throughout the fungal kingdom, a 3% dissimilarity cut-off was selected as a compromise in order to avoid overestimating fungal diversity versus masking rare OTUs and putative new emerging fungal pathogens (Nilsson et al. 2008; Schoch et al. 2012; Huse et al. 2010). Representative sequences, which are the most frequent sequence in each OTU, were extracted and then screened against Genbank nr/nt database using BLAST to identify rare, and potentially new emerging invasive fungal species not found in curated databases. Output Excel files were then organized alphabetically by Latin names and parsed for plant pathogens of interest and those on the quarantine species list of Canada and other industrialized countries.
Cross-talk is a phenomenon that occurs when a DNA read is assigned to an incorrect sample (Edgar 2018). When dealing with experiments attempting to identify emerging pests, incorrect assignment of DNA reads can lead to troublesome conclusions about its presence and distribution. To alleviate this problem, positive controls of targeted pests were not used in this trial in order to avoid cross-talk contaminations. One unused index was also sent as sequencing blanks during the sequencing output processing to quantify cross-talk.
Statistical Analysis
The primary output Excel file was compressed according to Genbank accession number to fuse redundant OTUs created by our bioinformatic method. OTUs that were ≥ 97% similar were pooled. Data for OTUs with less than 10 reads in a weekly sample were converted to zeros. We then generated three groupings of OTUs for analysis: 1) forest pathogens, 2) plant pathogens (which included forest pathogens as well as non-forest plant pathogens); and 3) all fungal OTUs (including plant pathogens). DNA read counts in the weekly samples were summed over the entire 4-week sampling period to yield a balanced data set with one cumulative count of each OTU for each collection method, site type, site and replicate.
Because the pore diameter of filter papers used to filter trap fluids was larger in New Brunswick and Nova Scotia (11μm) than in Ontario (2.7μm), some fungal species (i.e., with spores < 11 μm diameter) present in trap fluids in New Brunswick and Nova Scotia may have been missed. If that were true we would expect the relative performance of trap fluids vs. aerial spore collectors to be greater in Ontario vs. the other two sites. To test this, we ran separate 2 x 2 contingency table analyses on the proportions of fungal species detected by trap fluids vs. aerial spore collectors at each of the three sites, as well as on data pooled from all sites, and then tested for differences in performance among sites using the Chi square heterogeneity test (Zar 1999, pp. 500–502). We ran separate analyses on each of our three groups of fungi, i.e., all fungi, plant pathogens, and forest pathogens.
We then ran generalized linear mixed-effect models (PROC GLIMMIX) in SAS 9.4 for Windows (v. 6.2.9200, ©2002–2012, SAS Institute Inc., Cary, NC, USA) to test the effects of collection method (aerial spore collectors vs. insect trapping solution), site type (forest vs. urban), collection method x site type interaction, and site (Sault Ste. Marie, Fredericton, Halifax) on species richness and abundance of forest pathogens, plant pathogens, and all fungal species, and abundance of the more common individual forest pathogens. Collection method and site type were considered fixed effects and sites were considered random effects. We ran the models with Gaussian (on both raw and log (y+1) transformed data), Poisson, and negative binomial distributions and report results from the model with best fit (negative binomial in all cases) as determined by lowest value of the corrected Akaike Information Criterion (AICc). Least-square means were compared using the Tukey-Kramer method with experiment-wise error controlled at α = 0.05, but means and standard errors are reported on raw (species richness) or log(y+1) transformed data (abundance). Of the total of 101 forest pathogens, only 25 of the most common were analyzed by generalized linear mixed models to test for effects of collection method and site type on their individual abundance; high numbers of zero counts in the remaining OTUs made them inappropriate for analysis by generalized linear models.