Persistence and accumulation of environmental DNA from an endangered dragonfly

Detection of environmental DNA (eDNA) has become a commonly used surveillance method for threatened or invasive vertebrates in both aquatic and terrestrial environments. However, most studies in this field favor vertebrate target species. Environmental DNA protocols can be especially useful for endangered invertebrates such as the Hine’s emerald dragonfly (Somatochlora hineana) where conservation efforts have been greatly hindered by training, time, overall costs, and environmental impacts associated with conducting surveys in the calcareous fens occupied by this species. An essential step in developing such a protocol is to evaluate the dynamics of eDNA concentration under controlled conditions. We used the quantitative polymerase chain reaction (qPCR) to examine seasonal shifts in the persistence and net-accumulation of eDNA from captive S. hineana larvae in experimental mesocosms at temperatures corresponding with their overwintering (5.0 °C) and active (16.0 °C) seasons. Environmental DNA persisted longer at 5.0 °C but accumulated more readily at 16.0 °C. Differences in the accumulation and persistence of eDNA reflect differences in the longevity of eDNA at different temperatures and seasonal differences in larval S. hineana behavior. This study highlights the importance of considering how seasonal changes in temperature influence not only the speed of eDNA degradation but also the target species’ eDNA shedding rates.

The potential advantages of using eDNA rather than traditional surveying methods include the reduction of field labor hours 10 , reduced impact to sensitive habitats 7 , and a lower threshold of detection 11,12 . Additionally, eDNA has proven to be an effective tool when traditional methods require timely/costly surveying efforts 6 and for detecting cryptic invasive species 10 .
Although there is always some risk of damaging the habitat when studying a system, environmental DNA sampling (i.e., water, soil, ice) is much less invasive and has far less potential for harming native and endangered species than many traditional surveying methods 7 . For example, electrofishing can cause damage in the form of removing/killing fish from the sample site 13 . Traditional sampling methods for larval populations of S. hineana include benthic sampling (monitoring populations in stream beds) and burrow-pumping (a novel technique used to locate larvae within crayfish burrows) 2 . These techniques can disrupt flow patterns within shallow streams, collapse burrows, and harm/kill sampled individuals.
While there has been some speculation that eDNA sampling may have high false-positive rates due to ancient DNA contamination from extirpated populations, studies show that eDNA typically becomes undetectable in water within 1-44 days after source removal 10,[14][15][16][17][18][19][20][21] and approximately 144 days in soil 22 . This suggests that eDNA surveys are contemporaneous and can be used to inform conservation efforts.
Environmental DNA degradation is likely more complex in a field setting, and the persistence (defined here as the length of time eDNA remains detectable within a habitat or mesocosm) and net-accumulation (defined here as the difference between the amount of eDNA produced and the amount of eDNA degraded over time) are likely to vary depending on numerous factors that alter source/sink dynamics 3 . Spatiotemporal dynamics are especially important in affecting the persistence and accumulation of eDNA in the field and need to be accounted for when developing eDNA methodologies 23 . Concentrations of eDNA may fluctuate spatially and/or temporally as a result of fluctuations in biomass 18,24,25 , transport through a flowing system 17,26-28 , age structuring of target populations 7,16 , feeding activity 29 , life-history events 5 , seasonal habitat preference 13,30 , water temperature 24,31-33 , hydrology 13,27 , inhibition 13,27 , and microbial activity 34 . Some studies show that water pH affects eDNA degradation rates 19 , while others do not 35 . Similarly, some studies show that UV light exposure affects eDNA degradation rates 17 , while others show no such effect 36 .
In this study, we focused on the effects that seasonal shifts in temperature have on the persistence and netaccumulation of larval S. hineana eDNA. Since temperature drives the production of eDNA through metabolic processes 31 and directly alters the rate of microbial degradation of eDNA 32 , it may be the most important variable driving seasonal shifts in eDNA detection.
Somatochlora hineana larval molting activity varies with seasonal changes, the net-accumulation of S. hineana eDNA within a habitat. Adult S. hineana females lay eggs within streams and streamlets during their flight period (July-early August). Eggs typically mature over winter. In the following year, hatching of pro-larva from eggs occurs between April and June. All S. hineana larvae go through approximately 12 larval instars (F-11 to F-0). The first 6 larval instars (F-11 through F-6) occur rapidly within the first year, and the final 6 (F-5 through F-0) occur more slowly over a period of 2-4 years 1 . Since S. hineana larvae take several years to fully mature, they survive overwintering in shallow, partially frozen streams within Cambarus (= Lacunicambarus) diogenes crayfish burrows. While S. hineana larvae overwinter within burrows, they rarely consume food or molt, thus reducing the amount of eDNA shed 2 .
The net-accumulation of larval S. hineana eDNA was likely to increase with increasing temperatures 2,31,37 , while the persistence of larval S. hineana eDNA was likely to decrease with increasing temperatures 32 . Therefore, we assessed the seasonal shift in persistence and net-accumulation of larval S. hineana eDNA in temperaturecontrolled mesocosms that reflect the larval overwintering period (5.0 °C) and the larval active period (16.0 °C). This study provided preliminary information regarding the seasonal shift in eDNA production for larval S. hineana. Understanding the seasonal dynamics of larval S. hineana eDNA is vital for efficient detection of this rare aquatic species using eDNA protocols. Our mesocosm results have informed subsequent field sampling of S. hineana eDNA.

Results
Quality control. No extraction blank or no-template-control tested positive, indicating no contamination.
All positive and spiked positive controls tested positive, indicating no inhibition or reagent failure. Average standard curve r 2 values, which measure replicate reproductivity, were r 2 = 0.96 and r 2 = 0.98 for CYTb/FAM and COX3/HEX, respectively. Average amplification efficiency, determined by the slope of the standard curve, was 97.76% and 98.61% for CYTb/FAM and COX3/HEX, respectively. The limit of detection (LOD) for CYTb and COX3 with four sample replicates was 2 copies/µ L and 20 copies/µ L respectively (unpublished data).
Net accumulation of S. hineana eDNA. At 16.0 °C, containers holding S. hineana larvae in the large size class (approx. 0.33 g) accumulated significantly more eDNA than those with larvae in the medium size class (approx. 0.06 g) (linear mixed effects, n = 64, df = 375, p < 0.01) and in the small size class (approx. 0.01 g) (linear mixed effects, n = 64, df = 375, p < 0.001). Containers holding small and medium size class larvae accumulated comparable amounts of eDNA (linear mixed effects, n = 64, df = 375, p = 0.20). At 5.0 °C, containers holding larvae in the medium size class accumulated significantly more eDNA than those containers holding larvae in the large size class (linear mixed effects, n = 64, df = 375, p = 0.04). There was no difference in the accumulation among large and small (linear mixed effects, n = 64, df = 375, p = 0.45) or small and medium (linear mixed effects, n = 64, df = 375, p = 0.19) larval size classes.

Discussion
This study of S. hineana eDNA persistence and net-accumulation provides an example of one of the first applications of using emerging eDNA detection technology for the conservation of an endangered dragonfly (Odonata). We studied seasonal shifts in eDNA concentrations under controlled conditions at environmentally relevant  Although organisms like fish readily shed mucous and/or epithelial cells during periods of inactivity, organisms with an exoskeleton are more conservative in eDNA shedding rates, making them more difficult to detect 12 . The primary source of eDNA from living S. hineana is from molting and the production of fecal matter 37 . Our results indicate that overwintering larvae rarely shed eDNA. Additionally, eDNA persists longer at overwintering temperatures but has greater net-accumulation at temperatures when larvae are active. Therefore, eDNA sampling for larval S. hineana populations is likely more efficient in the spring/summer than in the fall/winter.
Results from our persistence experiment are consistent with other studies indicating that eDNA does not persist in a system beyond several weeks 10,14,16,20 , and that eDNA degrades more quickly at higher temperatures 24 . These results are also consistent with other eDNA studies indicating that eDNA degrades exponentially over time 4,17,18 . Our results indicated that eDNA signals are contemporaneous, reflecting recent eDNA production.
When water temperatures are near 5.0 °C, S. hineana eDNA signals can be detected 2 weeks after shedding. However, our exponential decay curve (Fig. 3b) suggested that eDNA concentrations neared the limit of detection for CYTb within 10 days (LOD: 2 copies/µ L) and within 5 days for COX3 (LOD: 20 copies/µL). This suggests that larval S. hineana eDNA is unlikely to persist at a detectable level in the environment after 2 weeks, even at overwintering temperatures. Therefore, assuming this pattern applied to field-collected eDNA samples, larval S. hineana eDNA results are contemporaneous.
Previous studies indicated that eDNA shedding rates can vary among older/younger and smaller/larger individuals. Some studies suggest that smaller individuals produce more eDNA than larger individuals due to rapid rates of development and molting 12 , while other studies suggest that eDNA production increases with biomass 18,24 . Our results showed that variation in eDNA shedding rates were season-or temperature dependent. At 16.0 °C, large larvae produced more eDNA than medium or small larvae, suggesting that eDNA production increases with biomass. However, at 5.0 °C medium larvae produced more eDNA than large larvae, indicating that medium sized individuals were more metabolically active than larger individuals while overwintering.
Consistent with studies indicating that eDNA shedding rates increase with higher temperatures 24,33 , S. hineana eDNA accumulated more readily over time in our 16.0 °C mesocosms than in our 5.0 °C mesocosms. Just as behavioral changes in other invertebrate species affect eDNA shedding rates 16 , the decreased rate of eDNA production by S. hineana larvae is likely associated with the reduction of activity among S. hineana larvae later in the season at overwintering temperatures. Preliminary field sample analysis suggests that our qualitative predictions of seasonal eDNA concentrations deduced from these mesocosm experiments are borne out in S. hineana habitat 38 .
These results suggest that it is more important to consider the effects that seasonal changes, including temperature, have on behavior (e.g., molting and overwintering) rather than the effects of temperature on eDNA persistence (e.g., microbial decay) when planning efficient eDNA sampling periods. Determining the most efficient time to detect target DNA is especially important for organisms with an exoskeleton, which generally produce small quantities of eDNA 7,8,12 . Given this, we expect that field detection of larval S. hineana eDNA will be less efficient when water temperatures are low enough to induce overwintering behaviors.
However, additional life history characteristics such as emergence have also been shown to impact seasonal eDNA concentrations 5 . Therefore, we predict that summer (after hatching and prior to emergence) would be the  To test this, we developed field experiments to document temporal variation in eDNA concentration over several months at sites with known S. hineana occupancy and detected eDNA concentrations comparable to those reported in this study 38 . For field samples, we increased sample volumes to reduce the risk of false-negative results when sampling from a highly dilute environment 36 . Since one study showed that qPCR methods were ineffective at detecting eDNA from endangered aquatic insects due to inhibition 9 , we continued to monitor for inhibition in field samples using "spiked" controls containing a mixture of sample and positive control template 38 .
During efficient sampling periods (e.g., summer months), eDNA can be a useful tool in the identification of S. hineana habitat. Environmental DNA results can be used to prioritize sites for traditional surveying methods, significantly increasing the efficiency of habitat detection and assessment. Similar eDNA projects may also be beneficial in habitat identification of many other threatened, endangered or invasive aquatic invertebrates.

Methods
Larvae used for this study had been reared from eggs over the course of 1-3 years from S. hineana females obtained at sites in Wisconsin and Illinois as part of a captive rearing program. All larvae and eggs of S. hineana were handled according to Federal Fish and Wildlife Native Endangered Species Permit (TE805269-15) issued to D. Soluk. Once larvae reached their 6th instar (F-5, F-6), they were individually held in 120 mL capacity specimen cups that were filled with approximately 80 mL of water. Larvae were kept in oxygenated water that was collected from a shallow well in Door County, WI. This water was similar to the groundwater that feeds larval habitat areas but had never come into contact with S. hineana DNA prior to addition to the specimen cups.
In the fall, the larvae were kept at 16.0 °C and were regularly fed a diet of amphipods and larval chironomids. Larvae were given regular water changes every 2-6 weeks. At experimental day 0 for the 16.0 °C trial, water was removed from specimen cups, combined, placed into beakers for the persistence experiment, and replaced with fresh water. Following the experiments at 16.0 °C, the water surrounding the larvae was adjusted over a period of 2-3 weeks down to 5 °C leading up to S. hineana's overwintering period. Since we have observed that S. hineana larvae do not consume prey while overwintering at 5 °C, larvae at this temperature were not fed. At experimental day 0 for the 5.0 °C trial, water was removed from specimen cups, combined, placed into beakers for the persistence experiment, and replaced with fresh water. Experiments were conducted in this manner two years in a row. Sixty milliliter water samples were taken with a syringe and immediately filtered through 0.45 µ m 25 mm Whatman Nitro-Cellulose Filters for future eDNA extraction. Specimen cups and beakers were gently nutated prior to sample collection to homogenize eDNA concentration throughout the container.
Net accumulation sample collection. To measure the net-accumulation of S. hineana eDNA at 16.0 °C and 5.0 °C, we filtered 60 mL water samples from larval specimen cups 5 and 10 days post water change. Six larvae were selected on each collection day to provide samples. Two larvae from each of three size classes were chosen: large (instar class F-0 to F-1, approximately 0.33 g live mass), medium (instar class F-2 to F-3, approximately 0.06 g live mass), and small (instar class F-4 to F-6, approximately 0.01 g live mass). Different larvae were chosen at days 5 and 10 post water change. Each larval specimen cup was only sampled once during this experiment.
Persistence sample collection. To measure persistence of S. hineana eDNA at 16.0 °C and 5.0 °C, water and debris removed from all specimen cups during the water change at day 0 of the accumulation experiment were combined and placed in one of two beakers. The two beakers were considered replicates and stored in identical conditions. At days 1, 5, 10, and 15 post larval removal, 60 mL water samples were taken from each beaker which contained approximately 600 mL of water. Filtered water was not returned to the beaker after sampling.
Sample extraction. We extracted sample filters using the QIAGEN DNeasy Blood and Tissue Kit commonly used for eDNA studies 17,25 . Approximately ½ of each filter was extracted, and the other ½ of each filter was stored in 75% ethanol at room temperature as a voucher. For every set of extractions, one extraction blank using only reagents and no sample was also processed to ensure that the reagents used were not contaminated. The ½ of the filter intended for extraction was further cut into thirds to increase lysis buffer exposure throughout the filter. These pieces were placed in open microcentrifuge tubes and allowed to dry in a fumigation hood until all of the ethanol had evaporated from the sample (approximately 6-12 h). Two alterations were made to the QIAGEN Spin Column-Animal Tissue Protocol: (1) QIAGEN Mini-Prep Spin Columns were used after sample lysis and before the precipitation of DNA with Buffer AL and 100% ethanol; (2) Samples were eluted using 100 μL of elution buffer twice for an end volume of 200 μL. qPCR. Environmental DNA primers, probes, and g-block specific to S. hineana were developed at the US Geological Survey's Upper Midwest Environmental Science Center based on the dragonfly's full mitochondrial genome 39 . Primer/probe sets specific to S. hineana mitochondrial DNA sequences Cytochrome c oxidase subunit 3 (COX3: 139 base pairs [bp]) and Cytochrome b (CYTb: 125 bp) ( Table 3) were multiplexed using qPCR 39 . HEX-labeled probes were used in conjunction with COX3 primers, and FAM-labeled probes were used in conjunction with CYTb primers. The limit of detection for these primers was approximately 20 copies/µ L (unpublished data).
The primers and probes used to detect S. hineana for this study were cross-validated against several species of Somatochlora (S. minor, S. elongata, S. franklini, S. incurvata, S. kennedyi, S. walshii, S. williamsoni, S. tenebrosa, S. linearis, and S. ensigera) as well as several aeshnid and libellulid species (unpublished data). Furthermore, water samples from three locations outside the known range of S. hineana, which support numerous dragonfly species www.nature.com/scientificreports/ that are sympatric with S. hineana, were tested with qPCR for the presence of S. hineana primer, probe, and g-block eDNA sequences. All these samples were negative for the presence of the S. hineana mtDNA sequences used in this study (unpublished data). Each sample was run in quadruplicate with both primer/probe sequences for a total of eight possible amplifications per sample. Each sample replicate consisted of a 20 μL reaction containing 10 μL of IDT's 2 × Master Mix with hot-start DNA polymerase, a 0.5 μm concentration of each of the four primers, and 0.25 μm concentration of each probe. During the qPCR run protocol, samples were initially held at 95 °C for three minutes and then cycled 45 times through denaturation (95 °C for 15 s), annealing (1.6 °C decrease per second), and extension (55 °C for 60 s).
G-block DNA, a single stranded DNA molecule (526 bp) containing COX-3 and CYT-b sequences specific to S. hineana and ND 5 and CR2 sequences specific to C. diogenes 39 , was purchased through IDT's custom DNA Oligos tool. The g-block was used to create standard curves for eDNA quantification by providing equimolar concentrations of all targets, allowing for unbiased amplification. One microliter of g-block DNA was run in duplicate for each plate at dilutions of 10 1 to 10 6 copies/µ L in 10 1 µ L increments to produce a standard curve on each plate. Sample quantities were determined by comparison to this standard curve. The critical threshold value (Ct-value) was set at 0.1 cq for both primer/probe sets. This value determines the level of fluorescence at which the sample is quantified.
Quality control. To prevent contamination, samples were handled using aseptic techniques. Separate rooms were used for sample preparation, extraction, and amplification. Before use, all surfaces were cleaned with a 25% bleach solution and pipettes were cleaned using DNA Away. All qPCR plates were prepared in a laminar flow hood, which was sterilized using a 20 min UV exposure immediately before and after plate preparation. To prevent sample degradation, filters were stored at room temperature out of direct sunlight in 75% laboratory-grade ethanol prior to extraction and DNA extracts were stored at -20 °C.
To identify the presence of compounds that would inhibit the amplification of sample DNA, we ran three "spiked" positive controls for each sample. Spiked positive controls consisted of 1 μL of 10 3 copies/μL g-block DNA mixed with 1 μL of sample template. If spiked samples had not amplified, this would have indicated the presence of inhibiting compounds. Four positive controls were run on each plate using 1 μL of previously extracted S. hineana DNA as sample template. One extraction blank was run for each extraction to monitor possible extraction reagent contamination. Ten no-template controls were run on each plate to monitor possible qPCR reagent contamination. Limit of Detection was calculated following standard eDNA qPCR assays 41 . Sample analysis. COX3 and CYTb sequence amplifications were multiplexed to increase likelihood of detection. Figures 1 and 2 were produced using the package ggplot2 40 . Figure 3 was produced using qplot from the R base package 42 . To analyze eDNA net-accumulation, we calculated the difference of least squared-means using linear mixed effects models (R 42 lme4 package 43 ) comparing eDNA quantities among both sampling day and larval size class including interactions between COX3/Cytb amplification as a source of random error. To analyze eDNA persistence, we calculated the difference of least squared-means using linear mixed effects models (R 42 lme4 package 43 ) comparing the natural log of eDNA sample quantity (+ 1 to include samples quantified at 0.00 copies/μL) among sampling dates including interactions between COX3/Cytb amplification as a source of random error. Exponential decay rate was analyzed with a non-linear least squares model (NLS) 4 . Specifically, we implemented an NLS model in R 42 using the SSasymp function 44 .

Data availability
The datasets generated during the study are available from the corresponding author on reasonable request.