In total, more than 15,000 data sets were analyzed (Table 1). The majority of the FS data refer to Hg and HCB, which have been monitored for a long time now and are included, e.g., in the FIS database. FS monitoring of PFOS started only in 2005. The ESB data go back to the mid-1990s making use of retrospective trend analysis of archived samples.
ESB sampling is highly standardized and focuses entirely on bream (A. brama). The publicly available FS data of the FIS database and the additionally provided data of the federal states refer to about 35 different fish species. The most commonly sampled non-predatory fish were bream (A. brama), chub (Squalius cephalus), roach (Rutilus rutlius), and eel (Anguilla anguilla), while pikeperch (Sander lucioperca), pike (Esox lucius), and perch (Perca fluviatilis) were the most frequently sampled predatory fish. The fish species distribution reflects the monitoring objective of the different federal states as well as the abundance of the species in the respective river systems.
Based on the trophic levels given in FishBase, about 50 % of the FS data for Hg refer to species that occupy the trophic levels 2.0–3.2 (HCB: 48 %, PFOS: 35 %), about 31 % to fish of trophic levels 3.3–3.9 (HCB: 33 %, PFOS: 44 %), and about 19 % are predatory fish of TL ≥ 4.0 (HCB: 19 %, PFOS: 24 %). The trophic levels of the most commonly monitored fish species are summarized in Table S1 (Online Resource).
For five ESB sites, the actual trophic levels of bream were calculated based on stable isotope analysis of nitrogen. According to these data, bream occupied trophic levels of 2.8–3.5 (mean 3.1, Online Resource, Table S2) which is in good agreement with the TL value of 3.1 given in FishBase and used in the normalization procedure.
Weight, length, age, sex, dry mass, and lipid content are included in all ESB data sets. Less biometrical data were available for the FS data: Depending on substance, 86–99 % of the data sets included fish weights and 52–97 % fish length. Age data was available only in 6–23 % cases and data on sex in 0–48 %. Lipid content was included in 21 % of the HCB data sets. No dry mass data was available for FS samples.
Analysis of the trend data was performed for overlapping sampling periods of the ESB and FS monitoring programs (Table 2). An additional analysis included all available data (i.e., also those years for which only ESB or FS data were available). The respective statistical parameters and the ESB data sorted by sampling sites including trends are compiled in the Online Resource (Tables S3–S14).
HCB was analyzed in ESB samples from the years 1993–2013 (Online Resource, Table S3).
The data of the FS monitoring of HCB go back to the 1980s. Lipid contents, however, were available only for 21 % (n = 1393) of the data sets and refer to the years 2000–2010. Correlation analysis for these data revealed a strong relationship between HCB and lipid contents (Pearson’s correlation coefficient, p < 0.0001).
In the years 2000–2010, HCB levels in ESB samples range between <0.2 and 77 μg/kg w/w (corresponding to 1.4–416 μg/kg w/w when adjusted to a standard fish of 5 % lipid content and TL 4). Highest levels were detected at the ESB sites in the Elbe and Mulde while levels in bream from the Danube were generally low.
In the FS samples, reported HCB concentrations ranged between <0.2 and 1331 μg/kg w/w and between <0.1 and 591 μg/kg w/w when normalized to 5 % lipid content and TL 4 (Table 2, details in Online Resource, Table S4).
Normalization resulted in higher HCB concentrations in the ESB samples but lower values in the FS samples. Since HCB accumulates in fatty tissue, the lipid content of the fish is a crucial factor. Of all ESB samples, more than 50 % had lipid contents below the fat standard of 5 % recommended in the EU guidance document. For these samples, the normalization resulted in higher HCB concentrations. In contrast, 68 % of the fish analyzed by the FS had lipid contents higher than 5 % (most of them were eel), and normalization resulted in lower HCB levels. Furthermore, the majority (77 %) of the FS fish belonged to higher trophic levels than bream, so that the adjustment to TL 4 had stronger effects on the ESB data.
Based on normalized data, the EQSBiota for HCB of 10 μg/kg w/w was exceeded at 10 of the 17 ESB sampling sites in 2013 (or 2012 at the sampling site Weil/Rhine; Online Resource, Table S3). Regarding the FS data and the years 2009 and 2010, six of 26 FS sites exceeded the EQSBiota (in the lower Rhine “Düsseldorf to Meerbusch,” “Rees to Grietherort,” “Aalschokker at Grietherort,” and “Emmerich,” as well as the sites “Elbe Abstiegskanal” and “Haiming” in the river Salzach).
Figure 1 summarizes the results of the trend analysis for HCB at different levels of complexity.
For the FS data, trend analysis was performed for the whole data set (including all species) as well as for eel as representative of a single fish species. Eel was chosen because most of the HCB data sets that included lipid contents referred to this species (e.g., for the period 2000–2009, 20–100 % depending on year; Online Resource, Tables S4, S5, S6) while considerably less lipid data were available for bream and fish of trophic levels TL 2.0–3.2.
HCB decreased significantly in bream sampled by the ESB between 2000 and 2013 (linear trend, p < 0.01) when all available data from this monitoring program were included in the analysis (Fig. 1, level 1). In contrast, no significant trends were detected for fish sampled by the FS in German freshwaters between 2000 and 2009/2010 (linear trends, all species: p = 0.06; eel: p = 0.08).
In view of the available FS data for HCB, further trend analysis at the next steps (levels 2 and 3) focused on the river Danube. This, however, had the drawback that ESB data were available only since 2004.
When considering the ESB data from all Danube sampling sites together at level 2 (Fig. 1), a significant decreasing trend was detected for the period 2004–2013 (linear trend, p = 0.01). Significantly decreasing HCB concentrations were also detected in fish sampled by the FS in the Danube between 2000 and 2009 (linear trends, all species: p < 0.01, eel: p = 0.01). When trend analysis included only the common sampling years 2004–2009, neither ESB nor FS data showed significant trends.
At level 3, trend analysis had to rely on relatively small sample numbers: for the common sampling period 2004–2009, six annual pool samples were available from the ESB site Kelheim (km 2404) and 24 individual fish (all species), respectively, 14 individual eel from the close-by FS sites between Danube km 2434 and 2400.
Nevertheless, slightly decreasing HCB concentrations were noticeable for both ESB and FS time series, but trends were not significant. When extending the trend analysis for the ESB data to the year 2013 (Fig. 1, level 3), the trend became more pronounced but was still not significant (linear trend, p = 0.06). In the case of the FS data, inclusion of earlier years (2000–2009) resulted in significant decreasing trends (linear trends, p = 0.01 (all species, n = 41) and p = 0.02 (eel, n = 23)).
Taken together, the findings indicate that normalization to 5 % lipid content and TL 4 did not overcome differences between individual fish or species at any of the three levels. Reducing complexity by focusing on one species only did not result in more homogenous data.
The most likely reason for this is the high variability within the eel data. HCB levels differed considerably between individual eels sampled in one year at the same or close-by sites. For example, for HCB and eel sampled between Danube km 2434 and 2400, standard deviations of the normalized data ranged between 41 and 128 %, depending on the year. This can be explained, on the one hand, by the opportunistic feeding strategy of eel, which includes fish, amphipods, decapod crustaceans, and terrestrial species (Froese and Pauly 2016; Jacoby and Gollock 2014). On the other hand, the duration of exposure may have varied (which could not be analyzed due to lack of fish age data).
Both factors complicate the interpretation of trends based on eel data and question the use of eel in chemical monitoring under the WFD.
Between 1993 and 2013, reported Hg concentrations ranged between 21 and 881 μg/kg w/w in the ESB samples (Table 2, Online Resource, Table S7). The concentration range was wider in fish sampled by the FS (i.e., reported concentrations of <10–9080 μg/kg w/w, Online Resource, Table S8). This is mainly because the FS data refer to individual fish and a large variety of different fish species and sizes whereas the ESB data refer to pool samples of bream only. In 2013, the EQSBiota for Hg of 20 μg/kg w/w was exceeded at all ESB and FS sites.
Normalization to 26 % dry mass and TL 4 led to higher Hg concentrations in both data sets because the majority of the sampled fish (ESB: 100 %, FS: 81 %) belong to trophic levels below 4.0. Adjustment to TL 4—especially when using a high TMF of 4.3 as done here—therefore results in higher Hg levels.
Trend analysis revealed significant decreasing trends (linear trends, p < 0.01) for Hg in all ESB samples and levels of complexity (i.e., samples from all ESB sites in German freshwaters (level 1), samples from all ESB sites in the Elbe (level 2), and samples from the ESB site Prossen at Elbe km 13 (level 3)). Analysis of the FS data was performed for the whole data set (including all species) and for bream (Online Resource, Tables S8, S9, S10). Decreasing Hg trends were only detected when focusing on bream but not when data of all fish species were included in the trend analysis (Fig. 2). Similarity between the relatively homogeneous ESB data and the FS data obviously increased with a reduction in sample variability.
The difference between species is also evident when comparing the relative standard deviation (in %) of the normalized data sets. With respect to the Elbe sampling site Prossen and the years 1996–2013, the relative standard deviation of calculated mean Hg concentrations in fish samples including all fish species ranged between 49 and 128 % (mean 81 %) depending on year, whereas it was only 0.5–51 % (mean 22 %) when only bream were considered.
Perfluorooctane sulfonic acid
PFOS was analyzed in ESB samples of the years 1995–2010 (Theobald et al. 2011) and 2013–2014. Adjustment to TL 4 using a TMF of 3.6 resulted in higher PFOS levels because bream occupy a trophic level below 4.0. Reported concentrations during the entire study period ranged between 0.3 and 91 μg/kg w/w, which correspond to normalized concentrations of 1.4–320 μg/kg w/w (Online Resource, Table S11). Lowest values were detected in Lake Belau and highest in the Rhine and at the lower Elbe site Blankenese. If compliance was based on normalized concentrations, only fish from Lake Belau met the EQSBiota of 9.1 μg/kg during the entire study period.
When considering only the period 2005–2010 (the time span for which also FS data were available), PFOS in ESB bream was in the range of 0.6–70 μg/kg, respectively, 2.3–234 μg/kg when normalized to 26 % dry mass and TL 4 (Table 2).
PFOS data from the FS monitoring were available only for rivers in Bavaria (n = 16, Van de Graaf et al. 2008) and North Rhine-Westphalia (n = 1252) covering the years 2005 and 2006–2010, respectively.
In contrast to the HCB and Hg data, which originate mostly from regular surveillance monitoring programs of the federal states, PFOS was analyzed within various programs addressing different questions. This included, for instance, the operational and investigative monitoring in highly contaminated waters. Accordingly, the PFOS data vary widely.
Similar to the ESB data, adjustment to TL 4 resulted in higher PFOS levels for 77 % of the fish. Reported concentrations ranged between <0.2 and 2749 μg/kg w/w, corresponding to normalized concentrations of <0.8–4852 μg/kg w/w (Table 2, and Online Resource, Table S12). Highest reported PFOS levels above 500 μg/kg were detected in a small creek in North Rhine-Westphalia, which is highly contaminated by a known point source, and in ponds in North Rhine-Westphalia with unknown water supply. In contrast, relatively low PFOS concentrations were found in fish from the large streams Danube and Rhine (North Rhine-Westphalian section), i.e., 4.9–9.9 μg/kg (normalized: 5.6–11.4 μg/kg) in the Danube in 2005, and 2.6–72 μg/kg (normalized: 3.0–236 μg/kg) in the Rhine in 2006–2010. These values are lower or in the same range as the ESB data for Danube (2005, reported 14.5–33 μg/kg; normalized 54–106 μg/kg) and Rhine (2006–2010, reported 0.6–70 μg/kg; normalized: 2.3–235 μg/kg). Based on the normalized data, only two sampling sites of the FS monitoring programs met the EQSBiota of 9.1 μg/kg w/w in 2010 (i.e., Urft /“Urfttalsperre” and Große Aue/“an der Landesgrenze”).
Trend analysis indicates a decrease in PFOS in the ESB samples (Fig. 3). Trends, however, were not significant when considering only the period 2005–2010. When extending trend analysis to the years 1995–2014, a significant decreasing trend (linear trend, p < 0.01) was detected for the combined data of all ESB freshwater sites (level 1).
Regarding the FS data from 2005 to 2010, trend analysis was performed for data of all fish species and for fish of TL 2.0–3.2 (i.e., trophic levels similar to bream) because not enough bream data were available for direct species comparison (Online Resource, Tables S12, S13, S14). The PFOS data referring to fish from ponds with unknown water supply and from the small highly contaminated creek were not included in trend analysis to avoid a bias. No significant trends were detected in the final data set. In contrast to the ESB data, PFOS seems to increase in fish sampled by the FS when data of all German sites are analyzed together (level 1). However, as mentioned above, these results are influenced by the heterogeneity of the sampling programs.
In view of the available FS data, further comparison between the ESB and the FS data focused on the Rhine. The FS data, however, refer to the lower Rhine (km 640–870) only, while the ESB data also include sampling sites in the upper and middle Rhine. Trend analysis including all Rhine sampling sites (Fig. 3, level 2) indicates that mean PFOS levels in the ESB samples have decreased between 1995 and 2014, but trends were not significant. PFOS in FS samples comprising all fish species were lower and remained more or less constant between 2006 and 2010. These lower concentrations are, at least in part, related to the normalization to TL 4, because 66 % of the fish collected by the FS belong to higher trophic levels than bream, which means that normalization had less effect on the PFOS levels. Similarly, no significant trend was observed when considering only fish of TL 2.0–3.2.
When focusing trend analysis on only one sampling site (level 3), decreasing PFOS levels were observed in the ESB samples from Bimmen (Rhine km 865) (Fig. 3, level 3). The trend, however, was only significant when all sampling years (1995–2014) were included in trend analysis (linear trend, p < 0.01) while no significance was detected for the period 2006–2010. No temporal trend was observed for fish sampled by the FS in the Rhine section km 780–860 (Fig. 3, all species). The picture here is quite similar to level 2 because the data overlap to more than 50 %. No trend analysis was possible for fish of TL 2.0–3.2 sampled between Rhine km 780 and 870 because the data cover only 3 years (Online Resource, Table S14).
Intra-annual variability of PFOS levels in fish of different species sampled between Rhine km 780 and 870 was relatively high (Online Resource, Table S14). The standard deviations of the normalized data ranged between 31 and 102 % (mean 73 %) depending on year. When considering only fish of TL 2.0–3.2, relative standard deviations ranged between 36 and 86 % (mean 68 %) and were thus only slightly lower. Since the database for PFOS is relatively small, it can only be speculated about the reasons for the observed variability. One possible reason is that normalization—at least when based on default values as done here—cannot overcome species-specific differences in PFOS uptake, excretion, metabolism, and/or accumulation. Another reason may be that PFOS levels differed between the sampling sites located between Rhine km 780 and km 870, which were evaluated together in order to obtain enough data for trend analysis.
Discussion of normalization procedure
The present study integrates fish monitoring data of different monitoring programs in Germany. We compared relatively homogeneous ESB data referring to annual pooled samples of bream muscle with muscle samples of individual fish of different species analyzed by the federal states. In order to compare these data in a meaningful way, it was necessary to standardize the reported concentrations to reduce biases caused, e.g., by different accumulation behavior between fish of different species and size, and overcome the effect of biomagnification in the food web (EC 2014).
The new EU Guidance Document No. 32 (EC 2014) recommends normalization to a predatory fish belonging to trophic level 4.0 with 5 % lipid and 26 % dry mass. Trophic level, lipid content, and dry mass are not only species-specific but vary between individuals and ecosystems. The EC guidance document, therefore, recommends basing the normalization on measured site- and fish-specific data. If measured values are not available, the guidance document suggests to use default values, for instance from FishBase (Froese and Pauly 2016).
This pragmatic approach is applicable to all priority substances, in contrast to, e.g., the procedure used by Åkerblom et al. (2014) who converted measured Hg concentrations in fish from Swedish lakes to a standard pike of 1 kg fresh weight. This Hg-specific approach based on Meili et al. (2004) relies on an empirically supported transfer function and default values derived from a database for Nordic fish (Munthe et al. 2004). Similar to the EU procedure, the measured Hg data are adjusted to a common trophic level.
In the present study, normalization followed the recommendations of the Guidance Document No. 32 (EC 2014). Since no measured values were available for trophic levels and, in the case of FS data, for dry mass, normalization had to resort to default values from FishBase (EC 2014; Froese and Pauly 2016).
The results for Hg show that similarity between the relatively homogeneous ESB data and the FS data increases when sample diversity is reduced (i.e., FS samples including all fish species compared to FS samples of bream only). These findings indicate that the adjustment to a common trophic level was not successful. A likely reason for this is the use of default TL values, which do not adequately reflect the actual trophic level of the respective fish.
For HCB and PFOS, the databases were relatively small and allowed no sound conclusion regarding the usefulness of normalization for data reporting under the WFD.
In order to evaluate the effectiveness of the normalization procedure more closely, we analyzed the data of selected sampling sites where fish of different species had been sampled in the same year. It was assumed that the basic exposure at the sites was similar for all species and normalization of the fish data would thus reduce the variability, leading to a decline in relative standard deviation. Table 3 summarizes the data.
The results are very heterogeneous and demonstrate that the normalization is not generally effective in reducing the variability between species.
Normalization of the HCB data used measured lipid levels and default values for TLs. The approach was limited to three sampling sites where enough fish of different species had been sampled in the same year and where HCB levels were above the LOQ. Normalization to 5 % lipid content and TL 4 reduced data heterogeneity in all cases by 31–57 %. However, in two cases, lipid normalization alone had stronger effects than combined lipid and TL normalization.
Hg and PFOS data were normalized using default values for dry mass and TL. For both substances, normalization to 26 % dry mass alone had very little effects. Combined normalization of the Hg data to 26 % dry mass and TL 4 resulted in reduced data variability by 6–43 % in six of nine cases while data heterogeneity increases in three cases. Results for PFOS are quite similar with reduced data variability by 3–38 % in six of eight cases while combined normalization strongly enhanced variability in two cases.
These results indicate that the normalization procedure proposed in the WFD Guidance Document No. 32 on biota monitoring (EC 2014) may be feasible for rather simple lipophilic compounds like HCB. However, it might oversimplify the real situation for substances like Hg and PFOS that behave in a more complicated manner (i.e., binding to sulfhydryl groups of proteins or to proteins in general).
Moreover, the findings question the relevance of normalizing chemical monitoring data based on default values. The fact that strong positive effects (i.e., reduction of variability) were obtained when normalization was based on measured values (i.e., lipid content in the case of HCB) underlines the importance of including the measurement of dry mass and lipid content in fish monitoring programs. Similarly, site-specific trophic levels of fish (determined, e.g., from stable isotope ratios against reference organisms like mussels) are required for the adjustment to a common trophic level. Furthermore, the derivation of TMFs should be standardized, and more TMF data from riverine systems are needed. This is especially important in the case of Hg where a strong dependency of trophic magnification on physical and chemical parameters like pH, DOC, and productivity is reported (Lavoie et al. 2013; Clayden et al. 2013). The generic application of the same TMF for different waters may therefore lead to erroneous results.
Taken together, a reconsideration of the recommended normalization approach may be necessary which refrains from using default values. Furthermore, preconditions for normalization should be defined. Lipid normalization, for instance, should only be applied on substances for which a relationship between lipid content and contaminant level is given (Hebert and Keenleyside 1995).