Activity, energy use and species identity affect eDNA shedding in freshwater fish

The quantitative measurement of eDNA form field-collected water samples is gaining importance for the monitoring of fish communities and populations. The interpretation of these signal strengths depends, among other factors, on the amount of target eDNA shed into the water. However, shedding rates are presumably associated with species-specific traits such as physiology and behavior. Although such differences between juvenile and adult fish have been previously detected, the general impact of movement and energy use in a resting state on eDNA release into the surrounding water remains hardly addressed. In an aquarium experiment, we compared eDNA shedding between seven fish species occurring in European freshwaters. The investigated salmonids, cyprinids and sculpin exhibit distinct adaptions to microhabitats, diets, and either solitary or schooling behavior. The fish were housed in aquaria with constant water flow and their activity was measured by snapshots taken every 30 s. Water samples for eDNA analysis were taken every 3 h and energy use was determined in an intermittent flow respirometer. After controlling for the effect of fish mass, our results demonstrate a positive correlation between target eDNA quantities as measured with digital PCR, fish activity and energy use, as well as species-specific differences. For cyprinids, the model based on data from individual fish was only partly transferable to groups, which exhibited lower activity and higher energy use. Our findings highlight the importance of fish physiology and behavior for the comparative interpretation of taxon-specific eDNA quantities. Species traits should therefore be incorporated into eDNA-based monitoring and conservation efforts.


Abstract 24
The quantitative measurement of eDNA form field-collected water samples is gaining 25 importance for the monitoring of fish communities and populations. The interpretation of 26 these signal strengths depends, among other factors, on the amount of target eDNA shed 27 into the water. However, shedding rates are presumably associated with species-specific 28 traits such as physiology and behavior. Although such differences between juvenile and adult 29 fish have been previously detected, the general impact of movement and energy use in a 30 resting state on eDNA release into the surrounding water remains hardly addressed. 31 In an aquarium experiment, we compared eDNA shedding between seven fish 32 species occurring in European freshwaters. The investigated salmonids, cyprinids and 33 sculpin exhibit distinct adaptions to microhabitats, diets, and either solitary or schooling 34 behavior. The fish were housed in aquaria with constant water flow and their activity was 35 measured by snapshots taken every 30 s. Water samples for eDNA analysis were taken 36 every 3 h and energy use was determined in an intermittent flow respirometer. After 37 controlling for the effect of fish mass, our results demonstrate a positive correlation between 38 target eDNA quantities as measured with digital PCR, fish activity and energy use, as well as 39 species-specific differences. For cyprinids, the model based on data from individual fish was 40 only partly transferable to groups, which exhibited lower activity and higher energy use. 41

Introduction 46
The sensitivity, non-invasiveness and cost-efficiency of environmental DNA (eDNA) based 47 methods has been proven for diverse habitats and species making them powerful new tools 48 for conservation biology and biodiversity assessments ( article 1, point 5a), all fish were reared according to regular agriculture (aquaculture) 133 practice, including provision of appropriate tank size, sufficient rate of waterflow, natural 134 photoperiod, ad libitum food supply, and temperatures within the species' thermal tolerance 135 range. This ensured that no pain, suffering, distress or lasting harm was inflicted on the 136 animals, confirmed by the fact that mortality rates were low and equal between rearing 137 groups. Based on the legislative provisions above, no ethics approval and no IACUC protocol 138 was required for the experiments performed. In particular the respirometry experiments were 139 discussed with the legislative authorities (Austrian Federal Ministry of Education, Science 140 and Research and University of Veterinary Medicine, Vienna) and the conclusion was that 141 the assessment of basic metabolism under these conditions (small fish sizes in relatively 142 large chambers) does not incur pain, suffering or distress to the fish and no formal animal 143 experimentation protocol was required. 144 Five aquaria (60 l) and corresponding plastic lids were used in the experiment, each 145 of which was thoroughly cleaned with sodium hypochlorite (5 %) and then rinsed with tap 146 water (fish-DNA free) prior to each experimental run (i.e. changing the fish under 147 investigation). The flow-through rate (aquaria were tap-water fed) was set to 5.45 l/min to 148 mimic natural conditions and keep eDNA concentrations in the fish tanks constant based on 149 previous test runs (data not shown). Water temperature was kept stable at 15 °C for inflowing 150 water automatically stabilizing temperature in the aquaria at this level. Each tank was further 151 equipped with an air-stone to ensure water mixing. At the start of each experimental run, a 152 water sample (negative control) was taken from one of the aquaria and processed as 153 described below. Then, five fish individuals of each species were selected aiming at similar 154 sizes across all fish species. Each fish was placed individually in an aquarium using DNA-155 free fishnets (Fig. 1). For P. phoxinus and S. cephalus, the experiment was carried out twice: 156 once with individual fish, and once with groups of three fish per aquarium. The day before the 157 experiment and for its duration, the respective fish were not fed to avoid contamination by 158 fish feed and minimize effects of defecation. Each run started with one day of familiarization 159 in the aquaria. 160 161 Water sampling, filtration and pH 162 All equipment used for this process was cleaned with sodium hypochlorite (5 %) and rinsed 163 with tap water prior to each use; DNA-free gloves were always worn. On the second day, 2 l 164 water samples were taken every three hours from 9 o'clock to midnight at the back end of 165 each aquarium (opposite to the inflow) using flexible tubes and 2 l wide neck bottles (Fig. 1). 166 The water samples were immediately filtered in an adjacent laboratory using glass microfiber 167 filters (1.2 µm pore width, 47 mm diameter, Whatman GF/C) and one negative control 168 consisting of 2 l MilliQ-water was included per sampling event. Thereafter, the filters were 169 individually placed in 2 ml reaction tubes and stored at -28 °C until further processing in a 170 special diagnostic molecular laboratory at the Department of Zoology, University of Innsbruck 171 (Austria). After each sampling, pH was measured in three arbitrarily selected fish tanks using 172 a Hach HQ40 device. 173 174 Activity measurement 175 During the familiarization time (day 1) and between water samplings, fish swimming activity 176 was quantified using a custom-made activity monitoring system consisting of one high-177 definition USB camera (Ziggi HD Plus, IPEVO.COM, Great Britain) per aquarium. The 178 cameras were placed at the front of each tank and the focus was set towards the back end 179 ( Fig. 1). To enable recordings during the night, aquaria were lighted throughout the two 180 recording days. Additionally, white polystyrene plates were used to cover the bottom and the 181 sides of each fish tank to exclude influences from neighboring aquaria and standardize 182 reflections. The signals from the cameras were acquired with a frame rate of 2 fpm ("frames 183 Thereafter, the remaining two fish individuals were placed in two measurement chambers for 210 one day of familiarization followed by one day of measurements (day five and six; Fig. 1). 211 The third chamber was left empty, but measured as well, to evaluate potential 212 microorganism-induced oxygen decrease. After the respirometer measurement day, the 213 mass [g] and total length [mm] of each fish was determined before placing them together in a 214 fish tank. For respirometer measurements of fish groups, the three individuals previously 215 sharing an aquarium, were put together in a respirometer chamber. well in the first plate. Using custom "binding" steps of the robotic platform, the DNA contained 236 in the first plate was transferred to the second one, next a binding step was carried out in the 237 second plate before transferring and releasing the entire collected DNA into the third plate, 238 which was then used for the Biosprint 96 tissue extraction protocol. After extraction, each 239 eluate was transferred to a 1.5 µl reaction tube for subsequent PCR. 240 All used primers (Table 1) (Table 1) prior 256 to determining target copy numbers per µl for each DNA extract. Per primer pair, a positive 257 (DNA extract from target species) and a negative control (molecular grade water) were 258 included in dPCR, all of which resulted positive and negative, respectively. 259 260

Statistical analysis 261
All calculations and visualizations were carried out in R Version 4.0.2 (R Core Team, 2020) 262 using the packages "ggplot2" (Wickham, 2016), "ggpubr" (Kassambara, 2019), "nlme" 263 (Pinheiro et al., 2020), "AICcmodavg" (Mazerolle, 2020), "rsq" (Zhang, 2020) and "sjPlot" 264 (Lüdecke, 2020). As pH was not measured in all aquaria after each water sampling, missing 265 values were estimated by averaging measurements taken at the respective fish tank before 266 and after the skipped time step. If measurements at the first or last water sampling were 267 missing, the values of the following or previous time step, respectively, were carried over.

1
The cleared activity dataset was visually inspected and summarized for each time 269 step: for example, data obtained during the preceding day were associated with the first 270 eDNA sampling event at 9 AM and measurements between 9 AM and 12 noon were 271 considered relevant for the second water sampling at 12 noon. Mean activity was calculated Concerning the fish-eDNA copy numbers obtained from dPCR, 21 filtered water 292 samples did not lead to an amplification. They were removed from the dataset, as other fish 293 individuals of comparable size and other samplings reliably produced positive results and 294 hence, errors in sample processing might have occurred. One group of P. phoxinus had to 295 be excluded from further analyses, as two of three individuals were accidentally chosen 296 S. cephalus. To determine whether the pH measurements, mean activity and eDNA copy 297 numbers were significantly influenced by sampling (i.e. time of the day), a one-way repeated 298 measurements ANOVA with rank transformation was calculated for each variable using a 299 combination of fish species and aquarium as random factor. A significant trend could not be 300 detected (Table 2). Despite efforts to standardize the mass of the chosen fish individuals 301 within and between species, fish mass was identified as confounding variable (SI 3). Hence, 302 eDNA copies, mean activity, and energy use were all normalized by the mass of the 303 respective fish individual prior to all further analyses. 304 Generalized Linear Models (GLM) for a Gamma-distributed dependent variable (i.e. 305 eDNA copies; positive continuous variable) and a log-link function were set up to investigate 306 the effects of fish species, energy use, mean activity, and pH (Faraway, 2016). Data 307 obtained from fish groups were excluded from the comparison of model performance. Due to 308 the small number of tested fish individuals and species examined, no models with random 309 factors were considered, but sampling and fish individuals were included in several models to 310 show their ineffectiveness in explaining the obtained data (Table 3). The variable "fish 311 species" was entered via dummy coding into the models using "C. gobio" as base category. 312 Corresponding with the focus of this study to investigate the effect of species identity, fish 313 physiology and activity on eDNA shedding, a set of six candidate models was chosen (Table  314 3). AICc, Δ AICc, and AICc weights (ω) were used to evaluate the strength of the six models 315 for describing the data including Nagelkerke, Cragg & Uhler's pseudo-R² values (Burnham 316 and Anderson, 2002). Simulated, scaled residuals were calculated based on the best-317 performing candidate model, (package: DHARMa (Florian Hartig, 2020); function: 318 "simulateResiduals"; n = 1000); the best performing model passed the consecutive check for 319 outliers and overdispersion. 320 To test the differences between single and grouped fish in the different stages of the 321 experiment, a data subset containing only values obtained from single and grouped 322 P. phoxinus and S. cephalus was analyzed. Target eDNA copies, energy use, and mean 323 activity (all normalized by fish mass) of the four distinct fish categories were tested for 324 normality and homogeneity of variance with Shapiro-Wilk and Bartlett tests. Then, 325 differences between groups were examined via Kruskal-Wallis tests followed by Wilcoxon 326 rank sum tests with Benjamini-Hochberg-corrected p-values. In a final step, target eDNA 327 copies for groups of P. phoxinus and S. cephalus were predicted using the model previously 328 established for single fish. Pairwise Wilcoxon tests were used to verify whether there was a 329 significant difference between predicted and measured target eDNA copy numbers for both 330 species separately and combined. 331 332 333 334

335
The mean mass of individually housed fish was 3.06 g ± 1.56 g (SD) and C. gobio individuals 336 had the highest mass (5 g ± 2.1 g (SD)). Water samples from P. phoxinus and T. thymallus 337 aquaria had the highest eDNA copy numbers per µl extract and gram fish mass (31.13 ± 338 53.23 (SD) and 47.68 ± 41.13 (SD), respectively; Fig. 2). The normalized mean activity was 339 highest for S. fontinalis (1.08 ± 0.33 (SD)) and lowest for C. gobio (0.34 ± 0.10 (SD); Fig. 2). 340 The energy use per gram fish mass was highest for O. mykiss (1.81 J/h ± 0.91 J/h (SD)), 341 while S. fontinalis and S. trutta aquaria had the lowest pH. 342 The Δ AICc-based comparison of model weight (single fish only) resulted in model #3 343 outperforming five other candidate models (Table 3 and Table 4). Therein, mean activity, 344 energy use and fish species were contained as explanatory variables (pseudo-R² NCU = 0.60; 345 dispersion parameter = 1.02). Increased activity had a significantly positive effect on eDNA 346 copy numbers (p < 0.01) and P. phoxinus, S. cephalus and T. thymallus displayed 347 significantly higher copy numbers compared to C. gobio (base group), after controlling for the 348 effect of fish mass. The relationship between energy use and copy numbers was also found 349 to be positive, but not significant (p = 0.05; Table 5 and Fig. 3). 350 For single and grouped individuals of P. phoxinus and S. cephalus, significant 351 differences were detected between the four groups regarding target eDNA copies 352 (Chi² = 10.59; p < 0.05), mean activity (Chi² = 80.95; p < 0.001) and energy use 353 (Chi² = 36.77; p < 0.001): mean activity was significantly higher when fish were kept solitary 354 compared to having them in groups for both species (p < 0.001). Contrastingly, energy use 355 was significantly higher for grouped individuals in both species (p < 0.01). Target eDNA 356 copies were significantly higher for grouped P. phoxinus (42.61 ± 48.04 (SD)) compared to 357 single and grouped S. cephalus and characterized by few outliers with particularly high eDNA 358 concentration ( Fig. 2 and Fig. 4) 359 To test the suitability of model #3 for describing eDNA shedding also for grouped fish, higher than for C. gobio and P. phoxinus. This is in concordance with general differences in 398 resting metabolic rates between these ecological guilds (Roberts, 1975 Based on the ontogenetic decrease in metabolic rate (Winberg, 1960), this could be an 433 alternative explanation for the low energy use of these two species. For studies investigating 434 eDNA shedding directly from live animals, biomass will always be an influential and 435 potentially confounding variable and should thus be considered carefully already during 436 experimental design. Depending on the actual main source of eDNA (feces, mucus…) it is 437 furthermore questionable, if fish mass is the best index variable to describe study animals. 438 Fish length should also be considered as it is a good approximator for activity measurements 439 via videotaping, an estimator of fish surface, and at least for eels a good estimator of eDNA  Tables and Figures   667 668 Cottus gobio (used as base category for dummy coding) and the respective fish species. 694 695 Figure 1: The setup of the aquarium experiment carried out with seven fish species: five 696 individual fish were put in fish tanks for water sampling (eDNA) and activity recordings (day 1 697 and 2) followed by respirometer measurements (three individuals on days 3 and 4; two 698 individuals plus empty control chamber on days 5 and 6). For Phoxinus phoxinus and 699 Squalius cephalus the experiment was repeated using groups of three individuals. 700 701 Figure 2: Key parameters obtained during the experiment for single fish. Boxplots display 702 target eDNA copies per µl extract, energy use [J/h], mean activity, and pH per fish species. 703 Fish species are abbreviated: "Cot gob": Cottus gobio; "Onc myk": Oncorhynchus mykiss; 704 "Pho pho": Phoxinus phoxinus; "Sal fon": Salvelinus fontinalis; "Sal tru": Salmo trutta; "Squ 705 cep": Squalius cephalus; "Thy thy": Thymallus thymallus. The variables target eDNA copies, 706 mean activity and energy use were normalized by fish mass to control for the effect of this 707 confounding variable. 708 709 Figure 3: Graphic representation of the GLM model coefficients (model #3) best describing 710 the obtained target eDNA copy numbers. Significance codes of denoted fish species indicate 711 differences in comparison to the base category Cottus gobio, whiskers denote the 95%-CI. 712 713 Squalius cephalus. Different lower case letters above boxplots code for significant 716 differences (p < 0.05) between categories, which are abbreviated as: "Pho pho": Phoxinus 717 phoxinus (single fish); "Pho pho g": Phoxinus phoxinus grouped fish; "Squ cep": Squalius 718 cephalus (single fish); "Squ cep g": Squalius cephalus grouped fish. 719 720 Figure 5: For groups of Phoxinus phoxinus (Pho pho g) and Squalius cephalus (Squ cep g) 721 measured and predicted copy numbers are plotted: left: against each other; middle: predicted 722 copy numbers are compared between species; right: comparison of measured copy numbers 723 between the two species. The measured copy numbers were log-transformed to enable a 724 direct comparison with the values predicted by the Gamma GLM with log-link function. For 725 S. cephalus a significant difference between measured and predicted copy numbers was 726 detected (t = 2.37; p = 0.02). 727