Evaluation of the Dietary Composition of Eurasian Perch (Perca Fluviatilis) in an Interconnected River-lake-gulf Aquatic System as a Supplementary Tool for the Interpretation of Measured Mercury (Hg) Concentrations (One-Year Study).

Eurasian Perch (Perca uviatilis) is one of ecologically signicant sh species in the Baltic Sea and has been recognized as a suitable organism to measure concentrations of hazardous substances that characterize levels of local pollution (e.g. heavy metals or persistent organic pollutants). However, the ability of the species to inhabit a wide range of feeding grounds raises concerns about the adequacy of monitoring data in relation to the representativeness of measured levels of hazardous substances at specic locations. Accounting of the migratory characteristics of this species can shed light on the origin of the analyzed specimens and thus trace the pollution uptake chain. Perch samples and potential perch prey were collected at three remote stations in a fully interlinked system river– lake– coastal waters of the Gulf of Riga. Mercury (Hg) concentration and stable isotope ratios ( 13 C/ 12 C and 15 N/ 14 N) were measured in each sampled item. The perch data were divided into three subgroups associated with specic feeding grounds and one mixed group. A Bayesian mixing model was implemented to quantify the feeding preferences of each group, and based on the results, inuence of each food source on Hg uptake by perch was modelled by means of Gaussian GAM model. = of migratory species between coastal and pelagic ecosystems based on changes in dietary preferences during the migration [21,29], which is possible due to clear isotopic differences between 13 C depleted freshwater and 13 C enriched marine food webs. Change in nitrogen isotope ratios 15 N/ 14 N has been used to distinguish trophic levels in freshwater and marine environments [20, 25].


Background
The global mercury (Hg) cycle is dominated by anthropogenic and natural emissions of gaseous substances of Hg to the atmosphere [19]. Simulatously, the wide use of mercury and mercury containing products [4] has resulted in more than 3000 localized mercury contaminated sites worldwide [15]. Furthermore, the improperly disposed industrial, household and medical products as well as pesticides used in the past have created legacy pollution sources.
Although the sources are not classi able as mercury contaminated sites, they are impacting environmental quality of geographically localized water basins they are discharging into.
In order to address local pollution sources and so to improve environmental quality, various programs, like river basin management plans in European Union, are being developed and implemented. To assess effectiveness of implemented measures, pollution levels and their trends are usually analysed in frame of environmental monitoring programs. Consequently, environmental research employs a variety of scienti c methods [8,23] in quantitative and qualitative analyses of the mercury in aquatic systems, including selection of a matrix for analysis such as water, sediments and biological material. Eurasian Perch (Perca uviatilis) is one of the ecologically signi cant sh species proposed by HELCOM as a biological matrix for environmental studies in the Baltic Sea, along with Baltic herring (Clupea harengus membras), cod (Gadus morhua) and eelpout (Zoarces viviparus) [7]. Perca uviatilis is omnivorous in the rst years of life, although the adults mostly follow a piscivorous diet [11]. The species occupy high trophic position; therefore, signi cant levels of the hazardous substances are commonly found in their tissues. The perch is widespread in freshwater and brackish water ecosystems, but usually are not considered to be an anadromous sh. It has been found as a suitable organism to measure concentrations of hazardous substances characterizing levels of local pollution. At the same time, it has been put forward by Järv [12] that the home-range migration (average 20 km, maximum observed 180 km) is a common behavioural feature for perch. The salt tolerance of perch and relatively low water salinity of the Baltic Sea and the Gulf of Riga allows this species to move from inland lakes and rivers to coastal waters. Generally, it has been assumed that once the feeding grounds in the coastal waters have been reached the specimens become reasonably stationary, consequently they can be used as a representative biological organism to characterize level of pollution in the speci c area. However, the high mobility of perch, the ability of the species to inhabit a wide range of feeding grounds [12,11], in addition to relatively high variability of measured concentrations raises concerns about adequacy of monitoring data regarding to the representability of measured levels of hazardous substances at speci c locations. These concerns are most pronounced in cases where different water bodies form interlinked network fully in range of perch migration distance, like the system river-lake-marine coastal waters.
Consideration of migratory characteristics of this species can shed light on the origin of the analyzed samples and thus trace the chain of pollution uptake, thus lling knowledge gaps on pollution distribution in interconnected aquatic systems.
Stable isotopes of nitrogen (N) and carbon (C) in soft tissues of sh are commonly used to study food webs and migration in aquatic ecosystems [27,17,10]. Carbon stable isotope ratios 13 C/ 12 C in marine and fresh water systems have been used to determine the movements of migratory species between coastal and pelagic ecosystems based on changes in dietary preferences during the migration [21,29], which is possible due to clear isotopic differences between 13 C depleted freshwater and 13 C enriched marine food webs. Change in nitrogen isotope ratios 15 N/ 14 N has been used to distinguish trophic levels in freshwater and marine environments [20,25].
By employing 13 C/ 12 C and 15 N/ 14 N stable isotope analysis in combination with perch stomach content analysis, the aim of this study was to determine how fundamentally different food bases affect the uptake of mercury from the food chains, as well as whether the perch, caught at a particular location, are representative for this location, or rather for the entire interconnected coastal-freshwater aquatic system.

Study site description
For this study the fully interlinked system, river Daugava -lake Ķīšezers -coastal area of the Gulf of Riga (Baltic Sea) was chosen ( Figure 1). The river Daugava sampling site (Station 1) was selected on the last section of the river between Riga Hydroelectric Power Plant (HEPP) and the estuary Daugavgriva, circa 5 km downstream to the HEPP and well upstream of the channel connecting lake Ķīšezers and river Daugava. The site can be characterized by rocky sediment type and rapid stream velocity. For the sampling Station 2 a lagoon type lake -Ķīšezers was selected. The lake is connected to river Daugava by a natural channel. It is rich with aquatic vegetation and represents a stagnant water pool. The sampling Station 3 was located in the Gulf of Riga, near the mouth of Daugava. The location represents brackish water coastal ecosystem with signi cant amount of detritus originating from adjacent rivers.
Sampling and pre-treatment The sampling campaigns took place in April and August 2017. Perch (Perca uviatilis) and other sh species were collected by means of scienti c nets: gill nets "Nordic" and nylon gill nets. Benthic organisms were sampled by Wildco Petite Ponar Grab, while zooplankton and cray sh by bag seine net and two-ring drop net, respectively. For suspended particular matter (SPM) surface water samples (up to 5 L) were taken from each sampling site by precleaned plastic bottle. Collected surface water samples were vacuum ltered for at least 30 minutes on precombusted (at 450°C for 2 h) 24 mm diameter Whatman GF/F lters to collect su cient amounts of material.
Length and weight of the whole sh was determined immediately after sampling by measuring board (accuracy to ±0.1 cm) and technical scale KERN FCE3K1N (accuracy to ±1 g). Thereafter, dorsal muscles were extracted. The dorsal muscles were placed into plastic container and frozen at temperature -18 °C. The samples of muscle tissues, zooplankton and benthic organisms were dried in vacuum freeze dryer until sample weight loss stopped and then homogenized by knife mill or agate pestle. Plastic containers with dried tissue samples were stored in desiccator in dark at room temperature (+ 20 °C) until further analyses.

Analytical methods
The concentration of mercury (Hg) in dorsal muscles of sh and other organisms was determined in laboratory at Latvian Institute of Aquatic Ecology (Daugavpils University) according to US EPA method 7473 "Mercury in Solids and Solutions by Thermal Decomposition, Amalgamation, and Atomic Absorption Spectrophotometry" [30] using "Teledyne Leeman labs" direct Hg analyser "Hydra IIc" (Mason, Ohio, USA). Calibration curves were generated using  (Table 1). Our results were in good agreement with the certi ed value given for the reference materials as well as our laboratory takes part regularly and successfully in the Quality Assurance of Information for Marine Environmental Monitoring in Europe (QUASIMEME) studies for quality assurance. Method blank (50 µL of deionized water) and random sample duplicates were also run during each batch. Method blank was less than 25% of the lowest detected Hg content in sample (0.20 ng) and was considered acceptable. Relative percent difference between sample duplicate analyses was 1.5 ± 2% (n = 22).

Data analyses and statistical assessment
Ward's minimum variance Clustering analysis was performed for identi cation of perch sub-groups with agglomeration of objects based on variables δ 13 C, δ 15 N and sampling location. The aim of the analysis was to split spatially sub-groups of perch with consideration of sampling location and geographical markers provided by the signals of stable isotope ratios. Optimal number of clusters was selected according to the Silhouette Widths method [3].
Bayesian mixing model SIAR was chosen for quanti cation of common diet in the computed sub-groups of perch, based on δ 13 C and δ 15 N ratios [24]. Trophic discrimination factor was approximated for each station separately, based on δ 13 C and δ 15 N ratios and trophic levels found in the literature (Table S1, Additional le 1) of every organism sampled. The estimated trophic discrimination factors were 3.37±1.27 (δ 15 N) and 0.36±1.00 (δ 13 C) for Station 1, Biomagni cation factor (BMF) was calculated only from species included into station-speci c diet of perch and perch itself. The food web BMF was computed from parameter b or slope of the following equation [22,5]: where BMF = 10 b Analysis of covariance (ANCOVA) was implemented to compare obtained biomagni cation regression curves. During the analysis interactions between assigned groups (from cluster analysis) and δ 15 N were evaluated to understand either the focus on speci c diet shows signi cant difference in slopes of Hg biomagni cation. Two regression models were compared: M1 -LOG concentration of mercury estimation from independent variable δ 15 N and independent factor Group; M2 -LOG concentration of mercury estimation from interrelated variable δ 15 N and factor Group. Group (the dataset was divided onto sub-groups, based on results of the clustering analysis mentioned above) as independent factor was considered for identi cation of differences between intercepts of the biomagni cation regressions. For the analysis selected value of signi cance level α was 0.05.
Smoothing function of Generalized Additive Models (GAM) was used to cover slightly non-linear relationship of LOGtransformed mercury concentration in perch dorsal muscles and length of specimens, thus allowing more sensitive evaluation of effect of dietary preferences. Data exploration protocol recommended by Zuur et al. [34] was applied before modelling process. The obtained models were validated according to the guide suggested by Zuur & Ieno [35], including check of homogeneity, independence, in uential observations, normality and t of estimated values. Akaike Information Criterion (AIC) [33] was applied to compare the obtained GAMs and determine the best t for the data, thus identify feeding sources and other concomitant factors impacting mercury accumulation in consumer tissues.
Due to collinearity of some variables, such as Crustacea and Neogobius melanostomus (correlation coe cient -0.8), Crustacea and Neomysis integer (correlation coe cient -0.7), Gymnocephalus cernua and N. integer (correlation coe cient -0.7), G.cernua and N.melanostomus (correlation coe cient -0.7), N.melanostomus and N.integer (correlation coe cient -0.7) three different models (A, B and C) were performed. Each of the models includes combination of non-collinear variables, and the three models together contain all the food items selected as sources for SIAR model mentioned above. Ammodytes tobianus was excluded from the models because of high covariance with C. harengus membras (correlation coe cient 1.0), thus further it may be considered, that the species have similar effect on mercury uptake. The following three models were selected: The Model C, with the most negative slope coe cient demonstrated by signi cant food item C. harengus membras, was selected as an example for visualisation of modelling results. Wilcoxon rank sum exact test were implemented to test differences between distribution and in the rank sums comparing mercury concentrations in perch tissues estimated from the Gaussian GAM model. The model was simulated for the scenarios with the maximum and minimum contribution ratio of C. harengus membras and continuously ranged from the maximum to minimum consumption (contribution) ratios of the other food items. The range limits at speci c sampling stations were same as computed by the SIAR model mentioned above. The visualisation example can be found in Additional le 2.
Data exploration, artworks, and statistical analyses were performed using R software for Windows, release 4.0.3.

Results
Mercury (Hg) concentrations and stable isotope analysis Hg concentrations measured in the dorsal muscles of perch varied notably in all three stations (Table 2)

Stomach content analysis
The analysis of stomach content showed that dietary preferences of perch signi cantly differ between fresh water and brackish water habitats (Figure 2). At sampling Stations 1 and 2, the crustaceans (found in 56% and 42% of the analyzed stomachs, respectively) were the predominant prey. Juvenile perch (22% at Station 1 and 25% at Station 2) and Chironomidae larva (11% at Station 1 and 21% at Station 2) were second favorite prey organisms while O. limosus and G. cernua were found mainly only in the digestive tract of perch from Station 1. At the same time, N. integer was the most preferred prey in Station 3 (found in 78% of stomachs). N. integer was also found in 25% of perch stomachs from freshwater Station 2. The N. melanostomus was the second most common prey in Station 3, where it was found in 29% of perch stomachs. The A. tobianus and C. harengus membras were represented only in 12% and 6% of stomachs from Station 3.

Cluster analysis
Scatterplot of the calculated stable isotope ratios δ 13 C and δ 15 N demonstrated clear evidence that perch specimens migrate between the sampling stations ( Figure 3). Substantial proportion of specimens sampled in Stations 1 and 2 had isotopic signals consistent with feeding in Station 3 ( Figure 3A). Consequently, we divided the dataset into four subgroups, according to the three characteristics: sampling place, stable isotope ratios δ 13 C, and stable isotope ratios δ 15 N related to a trophic position of organism ( Figure 3B and 3C).
The division was done as a cluster analysis based on the linear model criterion of least squares. Three of the subgroups were clearly representing respective sampling stations, while the fourth subgroup was well positioned as the mixed group with overlapping isotope ratio signals, which cannot be associated to any of the three sampling stations.

Exploration of the identi ed groups
The data was re-examined comparing Hg concentrations and distribution of individual's length among the new groups designated via cluster analysis. Group 1 exhibited the highest Hg concentrations ( Figure 4A) while lowest mean concentrations of Hg was found in group 3. Opposite to concentration levels, the highest mean length of perch was found in group 3 while the lowest one in group 2. The groups 1 and 4 exhibited the middle values ( Figure 4B).
Although the calculated bioaccumulation slopes were quite similar among the groups (coe cient values from 0.015 to 0.029), the intercepts differed noticeably (coe cient values from 1.4 for group 3 to 2.1 for group 1), thus indicating high variation of background Hg concentrations ( Figure 4C). The biomagni cation curve of the mixed group (group 4) was signi cantly different from the others by a steeper slope (

Discussion
The combination of stomach content analysis, as a sort of "snap-shot" of the recently consumed prey, with metabolically active tissues (such as muscles) that provide dietary and source information for up to several weeks [9] were instrumental in sorting out to which geographically distinct sampling area each perch specimen should be assigned. Since perch in the Gulf of Riga (Station 3 area) do not have suitable spawning and nursing grounds the specimen group assigned to that area has obviously migrated to the Gulf of Riga from freshwater similarly to that observed elsewhere by Järv [12]. This agrees with behavioural features of perch, like seasonal patterns in their distribution and movement between habitats [28]. The distinct stomach content and isotopic signal characteristic for this group suggests that once migrated to the coastal waters the perch specimens stay there whether as stable kinrelated groups as suggested by Gerlach et al. [6] and Semeniuk et al. [28] or as separate individuals. The approach applied in this study enabled us also to identify recent arrivals, e.g., specimens that have been feeding and accumulating Hg in another distinct area than that they were caught in.
As we successfully demonstrated, the perch specimens in freshwater ecosystems (river and lake stations) have substantially higher Hg concentrations. So, with some degree of certainty we can speculate that observed inter-annual differences, from 30 μg kg -1 ww in 2019 to 103 μg kg -1 ww in 2015, of Hg values obtained within national monitoring program (LIAE database) in coastal waters represented by Station 3, can mostly be explained by different proportion between recent arrivals from adjacent freshwater basins and specimens that have been feeding in area for more extended time period. Furthermore, the seasonal factor produced by all three GAM models, e.g., higher mercury levels were associated with spring sampling, can be clearly related to recent migration from inland waters to the coastal.
Although the concentrations of Hg in specimens representing freshwater ecosystem are substantially higher than in specimens representing marine coastal waters, the subtraction of values measured in recent arrivals from calculation of mean concentration resulted in slight increase of mean concentration in the coastal group. Most likely observed phenomenon is related to signi cant upward change of median size of perch, and not to the pollution level itself.
Therefore, it can be argued, that comparison of concentration means alone is poor pollution assessment approach.
At the same time, the Hg bioaccumulation curves in relation to the individual's length gave more detailed information about the speci c uptake tendencies. The results indicate that functional processes responsible for mercury accumulation (for example fish biometrics), Hg bioavailability and chemical composition of Hg substances [32] are quite similar, independently on the origin of specimen or local feeding base. So, the geographical differences in Hg concentration were mainly observed because background concentrations of Hg are substantially higher in the inland water bodies than in the Gulf of Riga. This conclusion is supported by notably higher Hg levels in suspended matter, used as a proxy of phytoplankton, measured in river and lake stations than in the Gulf of Riga. And, as stated by Kehrig [14], Hg enters food web at phytoplankton level and is transferred to higher organisms via trophic transfer.
The general structure of perch diet was quite similar among the studied areas, e.g., mostly several types of crustaceans, Chironomidae larvae and small sh. However, C. harengus membras presented only in the gulf station exhibited noticeable mercury reduction properties, which explains the substantial differences in the levels of mercury measured in the station-associated groups indicated by the clustering analysis. Moreover, according to the study, the trophic position of prey alone (in our case, δ 15 N) cannot be associated with the intensity of Hg uptake by consumer.
For example, Chironomidae larvae (δ 15 N 8.3÷13.4) and Crustacea (δ 15 N 7.0÷12.9) within the comparable maximum consumption ratio exhibited the opposite effects on the estimated mercury concentration in perch tissues, and N. integer (δ 15 N 11.4) despite the twofold maximum consumption ratio showed a neutral impact. Similarly, higher pollution rates cannot be associated with the trophic position of prey within the same feeding ground, which was well demonstrated by Chironomidae larvae and G. cernua (δ 15 N 15.7÷18.5), where the prey with lower values of nitrogen isotope ratios had stronger correlation with high Hg concentrations estimated from the model. Therefore, the suggestion by Le Croizier et al. [16] to utilize precise determination of the food sources for better facilitation of tracing of metal accumulation requires information on background concentrations at the site is important in our study as well.
The limitation of this study is that we present complete picture only from a single year perspective. We can of course speculate that the site-speci c food items de ned in this study will in uence perch Hg levels at an equal level also during following years. However, the well-known opportunistic feeding behaviour of perch [28] suggests that they will inevitably switch to other taxa if availability of previously consumed taxa becomes limited, or if appears more pro table source of energy, similarly as round goby (Neogobius melanostomus, invasive in the Baltic Sea) became a highly preferred prey for perch in recent years [1,26]. Another weak point to be considered is that the isotopic signal changes faster than the level of accumulated Hg [3,31]. So, the specimens, that at the onset of feeding period have spent su cient time in one area to equilibrate Hg concentration with the level characteristic for that area and then Page 17/25 migrates to another area and have time to change isotopic signals before they are caught, might not have su cient time to adjust also Hg levels. This could be improved by more regular sampling, which would give more precise information about in uence of perch mobility on measured mercury concentrations. Also, comparison with another distant Gulf of Riga station insigni cantly affected by the large freshwater ecosystems, could be useful adjunction for the further studies.

Conclusion
In general, the study clearly showed that the high mobility of perch along associated aquatic systems has notable effect on the measured mercury levels, which could be an issue for consideration during pollution monitoring events.
Jones et al. [13] suggests that, to avoid misinterpretation of spatial and temporal trends, sh biometrics modelling is of high signi cance when designing any monitoring program focused on seafood safety. In its turn the current study showed that trophic position and isotopic signatures can also serve as important supplementary tools for more accurate data interpretation.