Interactions with freshwater bio ﬁ lms cause rapid removal of common herbicides through degradation – evidence from microcosm studies †

We investigated the role of periphyton bio ﬁ lms for the fate of three common herbicides, i.e. bentazone, metazachlor and metribuzin, at low, environmental levels and 100 times higher, during a 16 days laboratory experiment. We found that herbicide water concentrations were stable during the ﬁ rst 8 days, whereas substantial declines (>78%) occurred between days 8 – 16 for all three herbicides. These rapid declines were explained only to a small extent (<8% of the total herbicide loss) by bio ﬁ lm sorption. As herbicide concentrations in light and dark treatments without bio ﬁ lms were similar, and the applied lightregimendid not coverthe UV-spectrum, herbicide photolysis was ruled out as a possible explanation for the observed declines. Furthermore, based on the compounds' characteristics, also volatili-zation was judged negligible. Therefore, we conjecture that the observed declines in herbicides were due to biodegradation and subsequentevasion of 14 CO 2 thatwasdrivenbyenzymatic actionfrom heterotrophic microbes. We reason that heterotrophic microbes used herbicide molecules as labile organic C-sources during C-limitation. Future studies should identify the microbial communities and genes involved in biodegradation in order to understand better the role of bio ﬁ lms for the self-puri ﬁ cation of surface waters.


Introduction
Freshwater periphyton biolms are complex assemblages of algae, bacteria, fungi, protozoans, and meiofauna embedded in a matrix of extracellular polymeric substances, EPS. 1 Biolms are sites of high biological activity and play important roles in primary production, 1 as a basal food resource for higher trophic levels, 2 and for carbon and nutrient cycling in freshwater ecosystems. 3Biolm EPS consists of microbial polysaccharides, lipids, proteins, nucleic acids, and heteropolymers that are essential for biolm integrity and stability. 4This EPS, as well as the microbial cells in biolms, represent a large surface area with efficient sorption sites for both heavy metals 5,6 and organic contaminants. 7,8For example, Tien and Chen 9 found that copper, nickel, chromium and lead were enriched by a factor ranging 1.6 Â 10 À5 to 7.15 Â 10 À5 L kg À1 in biolms, while Rooney et al. 8 reported bioconcentration factors (BCFs) between 12 and 6864 for 20 organic pesticides in biolms.As the detection frequency of current-use pesticides in biolms was 4fold larger than in sediments, while also better reecting ecological risks (e.g. for invertebrate communities), Mahler et al. 10 suggested that pesticide monitoring should also involve biolm as a complement to sediment.These studies emphasize the key role of biolms for the fate of pesticides in aquatic ecosystems.
Modern pesticides are used in crop protection worldwide, with almost 6 million tons of active ingredients applied in 2017, of which 7% (by weight) were herbicides. 11Through leaching or spray dri, pesticides commonly enter surface waters where they pose a risk to non-target aquatic organisms. 12For example, pesticides negatively affect the abundance of invertebrates, 13 as well as the density, antioxidant defence and photosynthetic efficiency of diatoms, 14 although the latter may reect shortterm inhibition. 15Herbicides are commonly more watersoluble than insecticides and fungicides and thus comprise a large fraction of the pesticides frequently detected in surface water monitoring. 16,17Herbicides in surface waters are submitted to both abiotic (e.g., photolysis, hydrolysis) and biotic (microbial) degradation, the latter usually being quantitatively more important.Microbial degradation of pesticides includes the enzymatic transformation by heterotrophic microbes via conjugation and complexation, resulting in compounds that have either increased or decreased persistence. 18Heterotrophic microbes degrade pesticides as these are relatively labile molecules that constitute a source of organic carbon, nutrients, and other elements necessary for growth 19 .Knowledge of herbicide behaviour and fate in aquatic ecosystems is important, especially as their use is expected to increase with climate change 20 and a growing human population. 21n this laboratory study, we assess the role of periphyton biolms for the fate of common herbicides in inland surface waters.We hypothesized that the herbicide water concentrations would be slowly declining during our 16 d experiment, driven by sorption to biolms.We also hypothesized that the high biological activity in biolms could contribute to accelerating the degradation of sorbed herbicides.We envision that our results provide an insight in the sorption characteristics of herbicides to biolms and their behaviour and fate in freshwater ecosystems.

Herbicide selection
Our study addressed the fate of metazachlor, metribuzin and bentazone, three herbicides that are commonly found in water from agricultural streams. 16,17These compounds are commonly applied in cultures of rapeseed, potatoes and carrots to prevent weed growth.Metazachlor is an inhibitor of cell division, while bentazone and metribuzin are photosynthesis inhibitors.Beside their common occurrence in inland waters, these herbicides were selected based on their similar and low log K ow values and their availability as 14 C-labeled standards.Applied exposure concentrations were similar to those found in European inland waters, 16,22 as well as 100-fold higher (referred to as 'low', and 'high' concentrations, respectively, Table 1).

Experiment description
An inoculum of epilithic biolm was collected from mesotrophic Lake Erken (59 50 0 15.6 00 N 18 38 0 06.1 00 E) by brushing the upper sides of a number of submersed cobbles collected in the littoral zone and transported back to the laboratory in a cooling box.This inoculum was used to grow biolms on unglazed ceramic tiles (3 Â 3 cm) covered with L16V medium 25 at 12 AE 0.2 C and 16 h : 8 h light : dark cycle with a light intensity of 924 AE 144 lux.The L16Vmedium is a broad synthetic medium that has been used for culturing multiple species of algae belonging to major taxonomic groups (e.g.Ahlgren et al. 1990 26 ).Aer 9 weeks of growth, biolms were well-established, and microscopic analysis showed that green algae made up 90% of the total biovolume, while diatoms and cyanobacteria accounted for 8% and 1%, respectively (detailed in 15 ).These biolms were used for exposure to pesticides as described below.
At the start of the experiment (t ¼ 0) four biolm-covered tiles each were placed in experimental units (i.e. 1 L glass beakers, n ¼ 4) containing 0.25 L of autoclaved L16V medium with additions of 14 C-labeled herbicides (IZOTOP Institute of Isotopes Co. Ltd., Budapest, Hungary).Blanks (n ¼ 4) were set up similarly, but did not receive herbicide additions.Experimental units were placed randomly on a table and provided with continuous, gentle aeration using glass Pasteur pipettes and aquaria pumps.Aer 0.5, 2, 4, 8 hours and then aer 1, 2, 4, 8, 12, and 16 days, 500 mL water samples were collected with an automated pipette from each of the experimental units and transferred to 20 mL scintillation vials.Biolm samples were collected from by removing single tiles from the same experimental units aer 1, 8 and 16 days.Biolms were detached from tiles with rubber cell scrapers, transferred to 20 mL scintillation vials, suspended in 2 mL of tissue solubilizer (Soluene 350, PerkinElmer), placed in an oven at 60 C for 4 h and allowed to cool to room temperature.
All samples then received 10 mL of scintillation cocktail (Ultima Gold®, PerkinElmer) and were kept in the darkness at room temperature for 24 h to obtain stable scintillation readings.Scintillation counts were made for at least 5 minutes or 10 000 CPM using a Beckman LS 6000TA-liquid scintillation counter.Quench corrections were done using internal standards ratios (Perkin Elmer).Disintegration rates for samples were corrected for the background values obtained for corresponding blanks.The QA/QC was performed using Internal Standard Kit, 14 C-W for aqueous samples and results are expressed as disintegrations per minute (DPM) and corrected for evaporation.Evaporation was quantied by weighing the experimental units before and aer each sampling.Sorption (in)to biolm is dened here as the ratio between the herbicide concentration in the biolm and that measured in water at the start (expressed as %).We thus consider the whole biolm, and do not distinguish among the different sorption/uptake mechanisms (e.g., diffusion, sorption to EPS or cellular uptake, etc).Herbicide loss from the experimental units was calculated as the difference between the DPM measured in water at the start and the DPM at the end (biolm + water).Blanks (no herbicides added) were used to quantify background To facilitate comparisons of our results with other studies, concentrations are also presented as mg L À1 and mg kg À1 , back calculated from DPM based on the specic activity of each compound (Table 1), using eqn (1) and (2).
where C w and C b are the herbicide concentrations in water (mg L À1 ) and in biolm (mg kg À1 ), respectively.DPM w is the upscaled disintegrations-per-minute from scintillation counting for 250 mL of water (the volume of the test units), corrected for blanks, evaporation, previous sampling and external standard recovery; Â4 is the multiplication factor for the conversion of the volume in our experimental units (250 mL) to litres.DPM b is the upscaled disintegrations-per-minute from scintillation counting for the whole biolm from one tile, corrected for external standard recovery; 1.67 Â 10 À8 is the conversion factor from DPM to MBq; SA is the specic activity of the herbicide in MBq mg À1 ; W b is the biolm wet weight in kg.
In an additional set up, treatments with autoclaved L16V medium lacking biolms were deployed in dark and light (629 AE 34 lux) conditions, respectively, to quantify the photolysis of herbicides.For this, water samples were analysed for pesticide concentrations at the start (day 0) and at the end (day 16) of the experiment by LC-MS according to the standard methods US EPA 535 27 and US EPA 1694. 28LC-MS was used in this experiment instead of scintillation counting as all 14 C-labelled herbicides had been used in the sorption experiment.The nutrient concentrations in the blanks were determined at start and on day 8, by the standard methods SS-EN ISO 6878:2005 mod., Bran Luebbe, Method No G-175-96 for AAIII (Total-P) and SS EN 12260:2004 (Total-N).

Data analysis
Pearson's correlations were used to assess relationships between herbicides' concentrations in water and in biolms, as well as between log K ow and mean biolm sorption.Repeated measures ANOVAs were run to test for effects of herbicide start concentrations and time on log-transformed herbicide sorption to biolms, and Tukey HSD-tests were used for post-hoc pairwise comparisons.Both tests were also used for investigating differences in biolm sorption between compounds.Normality of residuals was assessed from normal quantile plots.

Results
Herbicide water concentrations in all treatments decreased over time, but without a corresponding concentration increase in the biolms.Herbicide water concentrations were rather constant during the rst 8 days of the experiment, with a CV of 2 to 22%, but rapidly dropped by 78-98% of the initial concentration aer 12 days (Fig. 1).The most dramatic decline was observed in the bentazone-high treatment, where water concentrations dropped from 150.5 to 5.4 mg L À1 between days 8 and 12.By the end of the experiment (16 d), water concentrations of all herbicides had dropped by more than 94% of their initial concentrations.
Biolm sorption was generally low, on average less than 0.16 AE 0.18% of initially added concentrations.The highest average biolm sorption was 0.51%, reached aer 1 day in the metribuzin-low treatment.Biolm sorption of bentazone-low and metribuzin decreased linearly over time, whereas for bentazone-high and metazachlor there was an increase in sorption from day 1 to day 8, followed by a decrease between days 8 and 16 (Table S2 †).The initial herbicide concentration had a signicant effect on the biolm sorption of metazachlor and metribuzin, but not of bentazone (Table 2).Most notably, metribuzin sorption was 5 times higher at the low concentration than at the high concentration (Table S2 †).Moreover, in the low-concentration treatments, metribuzin sorption to bio-lms was on average 17-times higher than that of bentazone and 4-times higher than that of metazachlor (p < 0.0001 for both), despite the much lower initial concentrations (3-fold and 7-fold lower, respectively).In the high-concentration treatments, metazachlor sorption exceeded that of bentazone and metribuzin (p < 0.0001 and p ¼ 0.035, respectively), whereas metribuzin sorption was higher than that of bentazone (p < 0.0001).Although these differences were signicant, sorption to biolms was generally very low and accounted for only a small fraction of the total herbicide losses from the water.
The hydrophobicity of the herbicides (log K ow ) decreased their average sorption to biolms, albeit not signicantly (r ¼ 0.99, p ¼ 0.0591).The largest fraction (i.e., 8% of total loss) was recorded in metazachlor-high on day 8 of the experiment.By the end of the experiment (day 16), biolm sorption accounted for at most 0.01, 0.08 and 0.09% of the total loss of bentazone, metazachlor and metribuzin, respectively.Water concentrations of bentazone and metribuzin were strongly and positively correlated to those in biolms, especially in the highconcentration treatments (i.e., r ¼ 0.84 and 0.87, p ¼ 0.0006 and 0.0002, respectively), despite low sorption, whereas no correlation was found for metazachlor, when looking at the different concentration levels separately.
Fig. 1 Herbicide concentrations (mean AE SE; log scale) in water (circles and blue lines, mg L À1 ) and biofilms (triangles and green lines, mg kg À1 ww) during the experimental period in treatments with low and high concentrations of bentazone (upper panels), metazachlor (middle panels), and metribuzin (lower panels).SE ranged 0.004-0.09and 0.16-8.89for biofilms from low-and high-concentration treatments, respectively, and between 0.02-0.68 and 0.07-55.89for water from low-and high-concentration treatments, respectively.

Discussion
Our study shows a more than 78% decrease in herbicide water concentrations in all treatments aer 8 days.These rapid declines were explained only to a small extent (<8% of the total herbicide loss) by sorption and accumulation in biolms.Instead, more than 94% of the added compounds were lost from the experimental vessels by day 16.We conjecture that this was due to mineralization and subsequent evasion of 14 CO 2 , driven by enzymatic action from heterotrophic microbes on herbicide molecules when labile organic C-sources became limiting for their growth (cf. 29).Modern pesticides are relatively small organic molecules that can be readily used by heterotrophic microbes (e.g. 30,31).Our 14 C-labeled herbicides had one or multiple 14 C-atoms well integrated in their molecules (Table 1), leaving mineralization as the single option for the observed large loss of label from our experimental units.Likewise, and for similar time frames as applied in our study, Bohuss et al. 32 concluded that biodegradation was the main removal pathway and that sorption to biolms explained less than 0.6% of the total loss of the herbicides atrazine and acetochlor, whereas Lawrence et al. 33 showed similar sorption for triazine herbicides in river biolms.As sorption is a prerequisite for biodegradation, it is likely that even the low sorption observed in our biolms was sufficient to induce a rapid microbial degradation of the herbicides and ultimately their elimination from the microcosms as CO 2 .
The observed rapid degradation/mineralization of herbicides in the biolms implies a rapid turnover of sorbed herbicides and little accumulation.Also other studies 34,35 concluded that microbial degradation, rather than sorption, was the primary fate of herbicides (carbamates and diazinon) in 10-14 d experiments with river biolms.Possibly, the architecture of biolms may have changed with herbicide exposure, 36 thus altering the availability of sorption sites and allowing rapid degradation of herbicides in the biolms.Interestingly, the rapid mineralization of herbicide molecules in our study occurred when only two of four tiles remained in our experimental units, implying an increase in degradation rates per surface area of biolm, further stressing the adaptation of biolm microbiota towards a high efficiency in herbicide degradation/mineralization.If similar degradation/ mineralization rates occur under eld conditions in summer, then monitoring programs may seriously underestimate the run-off and/or leakage of pesticides from agricultural soils.
Beside C also herbicide-N and -S were likely readily taken up by microbes, as these are key elements in the synthesis of proteins and specic amino acids 37 and frequently more limiting than C.However, although N makes up 11.6-26.1% (by weight) of the three herbicides tested, while S makes up 14.9% in metribuzin and 40.0% in bentazone, the contributions of herbicide-N and -S were a negligible share of the total-N and -S in the algal growth medium, i.e. less than 0.008% and 0.030%, respectively, in the high-exposure treatments.These low numbers, however, should be seen as underestimates of their relative importance, as N and S originating from herbicides occur in organic molecules and will have a higher bioavailability than the nitrate and sulphate molecules in the medium.Comparisons of herbicide-N and -S with organic molecules in the periphyton biolms (e.g.originating from algal exudates/ decay and microfaunal excretions) would give a more justied estimate of the relative role of herbicide-associated N and S for the metabolism of heterotrophic microbes.Unfortunately, such data were not available from our study.
The water N : P ratio (by weight) on day 8 was 16 : 1, showing conditions for algal growth that are close to optimal, i.e. close to the Redeld ratio, 38 whereas the continuous aeration of the test vessels guaranteed a constant supply of atmospheric CO 2 .Earlier studies have also shown that recovery from herbicideinduced photosynthetic inhibition can be fast, 15 i.e. through rapid, adaptive evolution (cf. 39).Moreover, algae have short generation times, which also facilitates fast recovery.It is possible, however, that biolm-associated bacteria were limited by low-molecular organic C-sources and used added herbicides as a C-source.While most studies address herbicide effects on algae (e.g., inhibition of photosynthesis), much less is known about their effects on and interactions with heterotrophic bacteria in aquatic systems. 40erbicide loss through volatilization and photolysis is likely negligible in our study.First, the selected herbicides are clas-sied as 'non-volatile' according to Henry's law constants (Table 1), and it is thus unlikely that they partitioned from the water into the air phase.In line with this, the known metabolites of the investigated herbicides (Table S1 †), are also generally less lipophilic than their parent compounds (except for metazachlor oxalic acid), and hence their sorption is not expected to be higher.Second, photolysis was likely not quantitatively important, because the lights in our experiments did not cover the UV spectrum.Despite differences in light intensity (629 vs. 924 lux) between our main experiment and additional run to test for abiotic degradation, the lack of UV-range wavelengths in both light sources, and thus the lack of energy necessary to break the chemical bonds within the herbicide molecules 41 should have prevented direct photolysis.The latter is further supported by our observation that herbicide concentrations in light and dark treatments without biolms were similar (Fig S1 †) and did not show a decrease in herbicide concentration aer 16 days (Fig S2 †).Also, observed herbicide declines were between 3 and 18 times faster than expected from their documented hydrolysis half-life only (Table 1), thus stressing the role of biolms in this process.This further supports our conclusion that the observed rapid declines in herbicide concentrations most likely were due to microbial degradation in the biolms and subsequent evasion of 14 CO 2 .Potential growth inhibition from herbicide action can lead to a decline in the excretion rates of low-molecular compounds by algae and limit the growth of microbial heterotrophs in the biolms.The equivalent ratios between the EC 50 for algae and the high exposure concentrations for metribuzin and bentazone were 2.3Â and 0.014Â, implying that negative effects likely were negligible.The EC 50 values are based on tests with single planktonic species, where the compounds' bioavailability likely is much higher than in the complex biolms in our experiment.Also should the multispecies assemblages of our biolms likely have a higher resilience than single-species populations of plankton algae in standardized tests, further alleviating herbicide effects.
Spiking concentration affected sorption of metribuzin and metazachlor to biolms in our study (Table 2).In particular, the fact that metribuzin sorbed to biolms to a larger extent (i.e., 5fold more) in low than in high concentrations is an important nding, as it illustrates a high uptake at low environmental levels, which can affect phototrophic biolm community structure 42 and pose a risk for transfer to higher trophic levels. 43or bentazone, sorption was similar for the low and high concentration treatments (Table S2 †), suggesting saturation at the lowest concentration due to saturation of algal kinetic uptake rates.

Conclusions and outlook
Our study highlights the importance of biolms for selfdepuration of aquatic ecosystems, and suggests that biodegradation is the main degradation pathway for herbicides, mass balance-wise (sensu 44 ).Our ndings illustrate a rapid removal of herbicides from the water phase, with more than 94% of the amount eliminated aer 16 days.As biolm sorption only explained a small fraction of herbicide loss from the water, and as photolysis and volatilization of herbicides were judged negligible, we conclude that biodegradation was the main pathway of herbicide loss from our experimental units.This implies that modern herbicides likely are short-lived in surface waters during the growing season, where microbes compete for low-molecular carbon-substrata (including herbicides).Such degradation, including complete mineralization, contributes to the valuable ecosystem service of self-purication of surface water that biolm microbes provide and contribute to a systematic underestimation of pesticide run-off/leakage from agricultural soils based on water concentrations.

Table 1
Herbicides' physicochemical properties and nominal 'low' and 'high' concentrations used for biofilm exposure.Data from Pesticide Properties Database 23 include DT 50 (half-life of the compound in water), water solubility (at 20 C), Henry's constant (at 25 ) and EC 50 (median effect concentration for growth inhibition of planktonic algae); data from the US EPA 24 are empirical log K ow (octanol-water partitioning coefficient)

Table 2
ANOVA-effects of herbicide concentration (high and low) and exposure time on biofilm sorption; n.s. is not significant