Using recirculating ﬂ umes and a response surface model to investigate the role of hyporheic exchange and bacterial diversity on micropollutant half-lives †

Enhancing the understanding of the fate of wastewater-derived organic micropollutants in rivers is crucial to improve risk assessment, regulatory decision making and river management. Hyporheic exchange and sediment bacterial diversity are two factors gaining increasing importance as drivers for micropollutant degradation, but are complex to study in ﬁ eld experiments and usually ignored in laboratory tests aimed to estimate environmental half-lives. Flume mesocosms are useful to investigate micropollutant degradation processes, bridging the gap between the ﬁ eld and batch experiments. However, few studies have used ﬂ umes in this context. We present a novel experimental setup using 20 recirculating ﬂ umes and a response surface model to study the in ﬂ uence of hyporheic exchange and sediment bacterial diversity on half-lives of the anti-epileptic drug carbamazepine (CBZ) and the arti ﬁ cial sweetener acesulfame (ACS). The e ﬀ ect of bedform-induced hyporheic exchange was tested by three treatment levels di ﬀ ering in number of bedforms (0, 3 and 6). Three levels of sediment bacterial diversity were obtained by diluting sediment from the River Erpe in Berlin, Germany, with sand (1 : 10, 1 : 1000 and 1 : 100 000). Our results show that ACS half-lives were signi ﬁ cantly in ﬂ uenced by sediment dilution and number of bedforms. Half-lives of CBZ were higher than ACS, by organic micropollutants is an emerging problem. Global-scale risk assessment and modeling, as well as regulatory decision making rely on in-depth knowledge of major factors that in  uence the fate of micropollutants in rivers.Sediment bacterial diversity and hyporheic exchange are parameters that are o  en neglected when testing persistence of substances. We present a novel experimental setup consisting of 20 recirculating  umes and a response surface model to test the in  uence of these two parameters on micropollutant half-lives. Our setup proved useful to study the fate of micropollutants in rivers. Our results reveal that the underrated factors, hyporheic exchange and bacterial diversity signi  cantly a ﬀ ect micropollutant half-lives in rivers. experimental design with replicates. This study guides a way forward for more elaborate experimental setups that address how environmental processes a ﬀ ect the fate of micropollutants in rivers.


Introduction
Growing consumption of pharmaceuticals and personal care products renders contamination of freshwaters by wastewaterderived trace organic micropollutants an increasing problem for aquatic ecosystems and drinking water quality. 1 Understanding the processes that determine the degradation of micropollutants in the aquatic environment is crucial to (1) better predict aquatic risks on a global scale, 2 and (2) enhance river management at a local scale to promote favorable conditions for in-stream reduction of micropollutant concentrations. 3 Identication of key drivers of the attenuation of micropollutants in the environment has been attempted over the last few years in both eld [4][5][6][7] and lab studies. [8][9][10][11][12] While the complex interplay of spatially heterogeneous processes and transient boundary conditions in eld experiments hinders repeatability and oen reduces the validity of general conclusions about mechanisms, the repeatable results obtained in lab experiments oen lack diagnostic power for processes in real systems. 13 Bottle incubations carried out according to OECD guidelines 308 and 309, 14,15 for instance, are commonly used to test aerobic and anaerobic biodegradability of micropollutants in water-sediment systems. Compound specic half-lives resulting from these tests are used in current chemical legislation and for risk and exposure modeling. 2,16 However, these stagnant tests do not cover processes specic for river systems. They disregard dynamic hydrological processes, such as hyporheic exchange uxes and the characteristics of the site-specic bacterial communities. [17][18][19] To this end, systematic tools that are more suitable for simulating conditions in rivers are necessary to study the fate of micropollutants.
Flumes are experimental mesocosms that can bridge the aforementioned gap between controllability of lab experiments and the authenticity of eld experiments. Flumes have been proven valuable for investigating drivers of hyporheic exchange under controlled conditions 20,21 and are thus suitable to simulate hydraulic conditions in rivers. Flume designs can vary in complexity, from simple recirculating systems to those with advanced ow-control functions which can be used to simulate gaining and losing conditions in the hyporheic zone. 22 While more advanced designs are useful for testing complex hypotheses, they are clearly more challenging to replicate due to the need for additional infrastructure. In contrast, simple ume designs offer the opportunity to run experiments in parallel which enables testing of a wide range of scenarios and application of statistical models. Despite the apparent advantages of ume studies, to the best of our knowledge, there have been only a few studies investigating the fate of micropollutants in umes. When Li et al. 23 and Kunkel and Radke 24 used a recirculating ume to test degradation of micropollutants, only one ume was available, which is why they repeated the experiment to test the difference between at streambed and streambed with bedforms 23 and two different ow velocities, 24 respectively. However, no replicates and no statistical design were implemented.
Biodegradation is a key process for micropollutant transformation in rivers and is controlled by a variety of physical and biological factors. The hyporheic zone in particular represents a hot-spot of microbial activity. 25,26 It has been described as a key compartment for nutrient turnover and other biochemical reactions in rivers. 27,28 Schaper et al. 29 found that reach-scale removal of non-persistent polar organic micropollutants in a South Australian river was controlled by attenuation in the hyporheic zone during the wet season. Hence, enhancing the potential for hyporheic exchange will likely promote micropollutant transformation. While hyporheic ow can occur on many scales, 27 small scale exchange and short ow paths are expected to be the most effective in turnover processes. 30,31 Small scale hyporheic exchange is determined by sediment characteristics, mainly hydraulic conductivity and sediment morphology. 32 Bedforms such as dunes can induce hyporheic exchange causing a so called "pumping effect". High pressure on the upstream side of a dune forces the surface water into the sediment and low pressure on the downstream slope of the dune causes exltration of hyporheic water to the surface water. 20 While the extent of mass transport into the hyporheic zone and residence time distributions control the potential for turnover, the biochemical conditions control the quality of turnover. The composition of the hyporheic bacterial community is expected to play a major role in biochemical processes, but there has been little research on their impact on micropollutant degradation in rivers. Recent studies suggest that bacterial diversity in wastewater treatment plants is associated with biodegradation of certain micropollutants. 33,34 It was proposed that this positive correlation is observed when the biodegradation mechanism of a certain micropollutant can be performed by only a few bacteria taxa. There is scarce information on the compound-specic catabolic and co-metabolic pathways that are required for the biodegradation of the variety of micropollutants, and whether these are general or rare functions in bacterial communities. Stadler et al. 33 used a "dilution-to-extinction" approach to manipulate bacterial communities from activated sludge into different diversity levels and by this means identied taxa potentially driving the biotransformation of individual micropollutants. It was also shown that the change in taxonomic richness was associated with a change in functional richness across the treatment levels. In the absence of detailed knowledge on specic bacteria activity and functional characteristics, taxonomic diversity is a rst step to understand the link between bacterial community composition and biodegradation of micropollutants.
In this study we present a novel experimental setup to test the inuence of hyporheic ow and bacterial diversity on degradation half-lives of micropollutants. The experiment was based on a central composite face factorial design and used 20 ume mesocosms run in parallel to simulate different river conditions. The hyporheic ow was manipulated with the presence and number of bedforms. The bacterial diversity was controlled with a dilution-to-extinction approach. 33 The umes were inoculated with sediment from the River Erpe in Berlin, Germany, which receives high loads of treated wastewater. 6 As the sediment of this river is continuously exposed to micropollutants, its bacterial community has likely developed the capacity to degrade synthetic organic chemicals. A response surface model was used to assess the effects of the two variables (i.e., sediment dilution and number of bedforms) on micropollutant dissipation half-lives (DT50s). The DT50s of the arti-cial sweetener acesulfame (ACS) and the anti-epileptic drug carbamazepine (CBZ) are discussed to evaluate the performance of the setup. These chemicals were chosen as model compounds due to their widespread occurrence in freshwaters, high average concentrations in River Erpe (mg L À1 -range) 6 and contrasting behavior in the environment. Although both compounds were previously reported as relatively persistent, 35,36 recent research found increasing degradability of ACS within the last decade, likely caused by the adaptation of microbial communities in treatment plants. 37 In addition, a study conducted in the hyporheic zone of River Erpe showed that along a vertical ow path of 40 cm, ACS was removed by 78 AE 1% while CBZ was not removed signicantly. 30 Therefore, we expect generally lower half-lives and a higher inuence of the tested variables for ACS than for CBZ. The study aims at: providing a new experimental method to obtain a more accurate understanding of dependencies of micropollutant half-lives on river-specic biological and physical conditions and discussing the performance of the specic experimental setup.

Experimental design
A response surface model (RSM) and a central composite face design were employed to evaluate the effects of bacterial diversity manipulated by sediment dilution (S) and bedforminduced hyporheic exchange (B) on the attenuation of ACS and CBZ in ume mesocosms. The DT50 of ACS and CBZ in the ume system was used as a dependent variable. The two independent variables (S and B) were set at three levels (high, medium and low) as follows (Table 1): (1) Bacterial diversity in sediment (S) was achieved by diluting river sediment with commercial sand in different proportions, based on the dilution-to-extinction method. In this method, the less abundant species are "removed" by stepwise dilutions, and this loss in species richness results in lower diversity. 33 The low dilution level (S1) had a sediment dilution of 1 : 10 and was expected to have the highest level of bacterial diversity, the medium level (S3) was diluted in a 1 : 10 3 ratio, and the high dilution level (S6) corresponding to the lowest expected bacterial diversity, was set to a 1 : 10 6 dilution.
Bacterial diversity in S1, S3 and S6 levels was investigated through Illumina sequences of the 16S rRNA taxonomic gene (details in Chapter 2.7).
(2) Hyporheic exchange (B) was regulated by forming triangular-shaped stationary bedforms in the ume sediment. Bedforms cause a so-called "pumping" effect, inducing advective ow through the porous streambed by pressure differences between the upstream and downstream side of the bedform. 20 Hence, we anticipated, that higher amount of bedforms would lead to higher total exchange ux, i.e. the volume of water exchanged between sediment and surface water per day. 21 Consequently, the increasing solute transfer to the hyporheic zone leads to higher contact of solutes to bacterial communities and thus higher potential for micropollutant degradation in general. We aimed at creating three contrasting levels of hyporheic exchange by minimum, medium and maximum number of bedforms feasible within the present setting. Minimum exchange was expected for at sediment (B0), followed by a medium level (3 bedforms, B3) and a high level (6 bedforms, B6). The amount of sediment was identical in all umes and the shape of the individual bedforms was the same in the B3 and B6 umes. The triangular shape was determined by practicality within the setting aiming at uniform shapes across umes and inducing hyporheic exchange rather than mimicking shapes commonly found in the eld. Differences in hyporheic exchange between levels B0, B3 and B6 were investigated through a salt tracer dilution test (performed at the end of the experiment) from which exchange ux, exchange volumes and average residence times were calculated (details in Chapter 2.8).
The response surface model explores non-linear effects of the bacterial diversity (S) and hyporheic exchange (B) on the dissipation half-lives (DT50s) by tting the responses to a quadratic equation (eqn (1)): The central composite face design used here is a factorial design consisting of 20 umes (Fig. 1a): eight umes with the factorial variable combinations, eight umes with axial combinations and four replicates of the center-point experiments to validate the response surface model. Central composite designs are commonly used for response surface models because they are easy to expand (e.g. to include more variables) and exible in terms of choosing the values of each variable at the axial and center-points.

Preparation of the sediment-mixtures
Sediment from the River Erpe in Berlin, Germany, was collected 1 month prior to ume-setup. The sediment was homogenized and cooled at 4 C immediately aer sampling and until use. A mixture of commercial sands (Wickes, Watford, United Kingdom) was washed with tap-water to remove the nest fractions. Aerwards the sand was oven dried at 120 C for 24 h to reduce potential microbial contamination and stored in acidrinsed plastic barrels. The three sediment mixtures of the sediment dilution variable comprised of different ratios of sand and Erpe sediment to yield 20 L sediment for each ume ( Table  2). The components of the sediment-mixtures were merged and homogenised prior to application in separate acid-rinsed containers to avoid microbial cross-contamination. The three sediment dilutions S1, S3 and S6, have similar grain size distribution, total carbon (TC), hydraulic conductivity (K f ) and porosity, as they were to at least 90% comprised of the same sand type (Table 2). Thus, the physical hydraulic characteristics across the sediment dilution levels were comparable, which was important to diminish their inuence on hyporheic exchange. Any observed differences in hyporheic exchange in umes with the same number of bedforms but different sediment dilution level should be attributable to the bacterial inocula. The inuence of sediment dilution on hyporheic exchange is further discussed in Chapter 3.2.

Flume setup and pre-incubation period
The umes were oval (2.0 m Â 0.4 m), recirculating channels made of glass-reinforced plastic. The channels had an inner dimension of ca. 15 cm depth and 15 cm width (Fig. 1b). Prior to setup all 20 umes were thoroughly cleaned with water, rinsed with 10% HCl and covered with plastic lm until lling. The umes were positioned inside a white tent (without oor) to shield them from weather and direct solar radiation and ensure homogenous distribution of diffuse solar radiation. However, the umes were open to air-borne contamination from surrounding vegetation and microbes and exposed to day-night cycles of temperature. Each ume was leveled and equipped with an aquarium pump (NWA 1.6 adj 2.6 W, Newa Wave Industria, Loreggia, Italy) (see Fig. S1 † for pictures of the setup procedure). Before injection of micropollutants and the start of the attenuation phase (day 0), a pre-incubation was performed to allow regrowth of the bacterial communities in diluted sediments to similar abundance (day À12 to À1; Fig. 1c). At day À12, the sediment mixtures were distributed in the umes as shown for B0 in Fig. 1b and covered with 60 L (S1) or 58 L (S3 and S6) of deionised water (purchased from ReAgent Chemicals, Cheshire, England). The pumps were turned on simultaneously (surface water ow velocity initially ca. 8 cm s À1 ). Aerwards, nutrient mix N1 (see Table S1 †) was injected into the owing water to promote the growth of the bacterial communities. Glucose was added during the pre-incubation period as carbon and energy source. On day À3, bedforms were formed by hand using custom-built wooden plates to ensure standardized bedforms of about 8 cm (bottom to crest) Â 12 cm (start to crest) (Fig. S1e and f †). The plates were inserted in the at sediment at equal distance to the position of the crest and pushed together until the desired bedform shape was achieved. The at areas were leveled to about 3.5 cm sediment depth. In B6-umes, 3 bedforms were formed on both straight sides of the ume. In B3-umes, 3 bedforms were formed on the straight side opposite of the pump, while the side of the pump remained at (see Fig. 1b). In the B0-umes no bedforms were formed. During bedform formation, 10 cm porewater samplers (Standard Rhizons, Rhizosphere Research Products B.V., Wageningen, The Netherlands) were installed at 1.5 cm height in the center of the second and third bedforms (in ow direction) of B3 and B6 umes and at equal position in the at sediment of the B0 umes (Fig. 1b). Three of the S1 umes had additional porewater samplers in their rst bedforms.

Injection of micropollutants and attenuation phase
Aer 12 days of pre-incubation and right before the addition of micropollutants, surface water and porewater samples were collected from each ume. Thereaer, the attenuation phase was started by the addition of 3 mL of methanol (Fisher Scientic UK, analytical grade) containing 31 compounds (see Table S2 †) at about 200 mg L À1 to yield an initial concentration of 10 mg L À1 of each compound in each ume (day 0). The experiment ran for 78 days and samples were taken 10 times within this period (see Fig. 1c). Each set of samples was comprised of both surface water and porewater (approx. 10 mL per ume). Concentrations of micropollutants in the porewater are not discussed in the present work, but will be covered in a follow-up study. The samples were split into two aliquots and stored at À20 C prior to micropollutant analysis at Stockholm University (SU) and the Swiss Federal Institute of Aquatic Science and Technology (Eawag), respectively. For analysis of nutrients (NO 3 À , NO 2 À , NH 4 + , PO 4 3À and total dissolved nitrogen) and dissolved organic carbon (DOC) the same set of porewater and surface water samples were collected on days 0, 21, 42 and 78 and surface water was sampled on the rst day aer the start of pre-incubation (day À12). The umes were relled with 3 to 5 L deionised water one time during the preincubation period and 5 times during the attenuation phase to account for reduction of water volume due to evaporation. Two measurements of nutrient levels during the attenuation phase revealed that dissolved nitrogen was consumed completely in the umes of the S1 treatment. Therefore, nutrient mixtures N2 and N3 were added at days 10 and 46, respectively, to all umes (see Table S1 †). To ensure complete mixing, samples were never taken earlier than 4 days aer addition of deionised water or nutrients. Flow velocities were estimated three times by recording the travel time of a oater within a set distance of the straight sides of the umes. Two additional umes of the S3/B3 treatment were set up (not included in the RSM), to which no micropollutants were added. They served as a control for re-mobilisation of micropollutants from the Erpe sediment to the surface water and as a control for the reaction of the bacterial community to the micropollutants.
No re-mobilisation of ACS or CBZ was observed, as concentrations remained below limits of quantication (LOQ) throughout the experiment. Table 2 Mixtures and sediment properties of the three sediment dilution levels (S1, S3, S6), the undiluted Erpe sediment and the commercial sand a S1 S3 S6

Boundary conditions and abiotic parameters
Inside-air and water temperature (107 probes, Campbell Scientic, Logan, UT, USA), as well as inside-solar radiation (PAR Quantum, Sky Instruments, UK) were measured in 5 min intervals during the pre-incubation period and during the rst 5 weeks of the attenuation phase. Temperature and solar radiation for the remaining time of the experiment were derived from comparison of inside-values to air temperature and shortwave irradiance measured by a weather station in about 400 m distance to the experimental site ( Fig. S2 S2 and S3b †). pH in the surface water was measured two times. From day À4 to day 45, the average pH in the umes rose from 8.1 (AE0.1) to 8.5 (AE0.3) (Fig. S4 †). While there were no signicant differences between treatments at day À4 and day 45, the sediment dilution treatment had signicantly inuenced pH (ANOVA; p < 0.05). Treatment S1 (8.2 AE 0.3) had a signicantly lower pH than S3 (8.7 AE 0.1) and S6 (8.6 AE 0.2; Tukey post hoc test, p < 0.05). Dissolved oxygen (Pro 20 DO Instrument, YSI Incorporated, Yellow Springs, OH, USA) in the surface water was measured 4 times during the attenuation phase. Average O 2 saturation in all umes ranged from 101 to 110% between days 28 and 86 (Fig. S5 †). Average O 2 saturation at days 28, 36, 44 and 86 was signicantly inuenced by the sediment dilution treatment (ANOVA; p < 0.05). S1 treatments showed signicantly lower O 2 saturation (103% AE 0.8) than S3 (105% AE 1.7) and S6 (106% AE 1.9; Tukey post hoc test, p < 0.05).
The method for analysis of nutrients (NO 3 À , NO 2 À , NH 4 + , PO 4 3À and total dissolved nitrogen) and dissolved organic carbon (DOC) is detailed in the ESI. † There was little variation in surface water nutrient dynamics between the bedform treatments, but nutrient concentrations were highly impacted by the level of sediment dilution (Fig. S6 †). This is why at day 0 (injection of micropollutants), nutrient concentrations differed between sediment dilution levels. Generally the depletion of nitrogen and DOC during pre-incubation (day À12 to day 0) was higher and faster in treatments with lower dilution. Accordingly, aer addition of nutrient solution N2 at day 10, removal of NH 4 + was especially high in the lowest dilution (S1) (see ESI † for detailed discussion on nutrient dynamics).

Chemical analyses
Swiss Federal Institute of Aquatic Science and Technology (Eawag). Frozen water samples (À20 C) were equilibrated to room temperature, which was followed by a centrifugation step (2 mL sample, 3020g for 30 min at 20 C, Megafuge 1.0R, Heraeus) and the transfer of a supernatant aliquot (1 mL) into a HPLC sample vial. Isotope-labeled internal standards (IS) were directly added to the vial (20 ng mL À1 ). The calibration series was prepared in NANOpure™ water (0, 1, 10, 100, 500, 1000, 2500, 5000 and 10 000 ng L À1 ). Calibration standards and water samples were then injected (50 mL) into a reversed-phase C18 liquid chromatography column (Atlantis T3, 3 Â 150 mm, 3 mm, Waters, USA). The samples per ume were measured in four batches and in an inverse time order (e.g. day 56 rst and day 0 last), with blanks before and aer each ume. A new calibration series was prepared and separate LOQs were calculated for every batch. Water and methanol, both acidied with 0.1% formic acid, were used as eluents for the chromatographic gradient (0% to 95% methanol in 18.5 min, 95% methanol for 10 min, 95% to 0% methanol in 4 min). The analytical column was coupled to a high-resolution tandem mass spectrometer (QExactive or QExactive Plus, Thermo Scientic, USA) by an electrospray ionization interface. Mass spectra were acquired in full-scan mode (polarity switching) at a mass resolution of 140 000 (FWHM at m/z 200) with subsequent data-dependent MS2 (Top5, mass resolution 17 500). For the quantication of target analytes, chromatographic peaks were automatically detected (5 ppm mass tolerance) and integrated (minimum three data points) using the ICIS algorithm of TraceFinder (version 4.1 EFS, Thermo Scientic, USA). Peak integrations were reviewed manually. To each target analyte (CBZ, ACS), a matching IS (ACS-D4, CBZ-D8) was assigned (internal standard method). Linear 1/x-weighted calibration curves were generated by tting the analyte concentration (x) against the STD-to-IS peak area response ratio (y) without forcing the t through zero. Nanopure water was generated with a lab water purication system (D11911, Barnstead/Thermo Scientic, USA), methanol was of LC/MS grade (Optima™, Fisher Scien-tic, Switzerland), and formic acid of analytical grade ($98%, Merck, Germany). Overall, 28% of the measured samples were quality control samples (blanks, calibration standards, spike recovery samples). The LOQs of ACS were between 22 and 100 ng L À1 in porewater and 21 and 100 ng L À1 in surface water. For CBZ, LOQs were between 10 and 100 ng L À1 , both in porewater and surface water. The spike recovery (accuracy) was determined in porewater and surface water of ume 2 (S6/B3) aer 28 days, i.e. approximately in the middle of the ume experiment. Spike recoveries of the two analytes were in an acceptable range, ACS was recovered at 112% and 104%, CBZ at 85% and 72%, both at high precision (RSD < 3% among duplicates) in porewater and surface water, respectively. The spike recovery test does not account for the freezing of samples but signicant losses of analytes during freezing is unlikely. All samples were treated in the same way, and as the relative attenuation compared to the initial concentration is the most important parameter for this study, freezing will not have a major impact on the results.
Stockholm University (SU). Samples were analyzed using a small volume direct injection-ultra high performance liquid chromatography method coupled to tandem mass spectrometry (UHPLC-MS/MS) following a standard protocol established previously. 38 Briey, samples were stored at À20 degrees, defrosted at room temperature and thoroughly vortexed before processing. A sample volume of 800 mL was then combined with 195 mL methanol and the isotope-labeled internal standard mix in 5 mL methanol and aer a further vortexing step ltered (Filtropur S 0.45 mm, PES membrane, Sarstedt AG&Co, Nuembrecht, Germany) into microvials (2 mL; Thermo Scientic, Dreieich, Germany). The injection volume was 20 mL. A blank sample and a quality control standard was injected every 15-20 samples. The precision determined with the quality control standard was 1.5 and 8% RSD for ACS and CBZ, respectively. In accordance with the method applied at Eawag, internal standards for ACS and CBZ were ACS-D4 and CBZ-D8. The LOQ of ACS and CBZ was 88 and 25 ng L À1 , respectively. Spike recoveries were in the range of 95 to 108% for ACS and 86 to 112% for CBZ. Detailed information on instrument settings, quantication and further QA/QC parameters can be found in the method section of Posselt et al. 38

Bacterial diversity in sediment dilution treatments
Sediment samples (approximately 10 g per sample) were collected from each ume on day 21 of the attenuation phase and immediately frozen at À80 C. DNA was subsequently extracted from a 0.5 g sub-sample according to the rapid protocol for the extraction of total nucleic acids from environmental samples. 39 DNA concentration was then determined with Quant-iT® PicoGreen DNA assay kit (Invitrogen, Germany). Total bacterial community was quantied based on the 16S rRNA gene using quantitative real-time PCR according to published protocols. Results from quantitative real-time PCR were used as an estimate for bacteria biomass 40,41 to evaluate the effectivity of the pre-incubation period.
Illumina Miseq amplicon sequencing targeting the 16S rRNA gene using the bacteria specic primer pair 341F and 806R was performed by LGC Genomics GmbH (Berlin, Germany) followed by post processing of the raw data as previously published. 42 Taxonomy was assigned using the Ribosomal Database Project (RDP) classier. For comparative diversity index analyses, uneven sequencing depth among the samples was adjusted by rarefying each sample to an even sequencing depth. The sequence data were submitted to NCBI's sequence reads archive (http://www.ncbi.nlm.nih.gov/sra/) under accession no. PRJNA531245.
Fisher-alpha diversity index was calculated at a genus taxonomic level for all the samples and ANOVA was used to compare the diversity between bedform levels (B0, B3, B6) and sediment dilution levels (S1, S3, S6). Calculations were performed in the R soware 43 using the vegan 44 and phyloseq 45 packages.

Salt tracer dilution test
A salt tracer dilution test was conducted at the nal point of the experiment, to calculate hyporheic exchange metrics (exchange ux [L d À1 ], exchange volume [L] and residence time [d]) in each ume. 46 Aer termination of the attenuation phase 50 mg NaCl were added to each ume. Flumes 11-20 were equipped with loggers (CTD-Diver, van Essen Instruments, Del, the Netherlands) for electrical conductivity (EC), while in umes 1-10 EC was monitored using a hand-hold EC-meter. The recessions of the electrical conductivity resulting from the dilution of NaCl were measured for 163 hours. The electrical conductivity of the surface water will be diluted as a consequence of hyporheic exchange, as surface water with high salt concentration will gradually mix with porewater of lower salt concentration. To account for deviations between devices, all loggers and the manual EC-meter were calibrated relative to one of the loggers, using ve NaCl solutions of different concentrations. The EC values were then corrected by the calibration curves of the single loggers. Details on the calculation of hyporheic exchange metrics can be found in the ESI. †

Dissipation half-lives and response surface model
The concentrations of ACS and CBZ (C x in eqn (2a)) were rst divided by the respective measured initial concentration at day 1 (C 0 in eqn (2a)) to scale time trends from 0 to 1. Concentrations below LOQ in both data sets were excluded. Normalized concentrations measured in Eawag and SU have a good agreement with an average RSD of 7.9% for CBZ and 6.8% for ACS (Fig. 2). Eawag and SU data sets were therefore pooled and averaged for all the individual timepoints. Measurements that were available in only one data set (e.g. concentrations at days 2 and 78 were only available in the SU data set) were used directly in the analysis (Fig. 2).
The DT50s for CBZ and ACS were calculated assuming rst order kinetics by tting the timepoint-averaged measured concentrations to an exponential function (eqn (2a) and (2b)). If no dissipation was observed in the initial timepoints, this period was considered as lag-phase and was excluded from the DT50 calculation.
The goodness of t of the rst-order dissipation assumption was assessed with a one-tailed t-test of the kinetic constants (k dis ) to be signicantly different from zero. Only DT50s obtained from kinetic constants (k dis ) signicantly different from zero (p # 0.05), n ¼ 20 for ACS and n ¼ 18 for CBZ, were tted to a quadratic response surface model (RSM; eqn (1)), using the rsm package (Lenth RV, 2009) in the R soware. 43 The coded levels (À1, 0 and 1) were used for the sediment dilution (S) and bedform (B) variables ( Table 1). The use of the coded variables to t the RSM is adequate for our specic goal to understand the relative size and effect of the variables, as in the present study we do not aim to predict the DT50s or optimize the attenuation of micropollutants. The model coefficients b x (eqn (1)) were rst calculated using ordinary least squares, then tested to be signicantly different from zero (two-tailed t-test), and nally an analysis of variance (ANOVA) was used to evaluate the signicance of the rst (b 1 and b 2 ), second order (b 4 and b 5 ) and interaction (b 3 ) terms. The adjusted-R 2 , F-test and model lack-of-t were calculated to assess goodness of t and adequacy of the regression.

Effect of sediment dilution and bedforms on bacterial diversity
The sediment dilution caused a signicant decrease in bacterial diversity (based on the Fisher's alpha index) observed in the ume sediment at day 21 (ANOVA; p # 0.01, Fig. 3b), whereas no signicant effects on diversity were observed for the bedform variable or the interaction term (Fig. 3a). Hence, the sediment dilution used in our study based on the "dilution-to-extinction" method 33 effectively induced different bacterial diversity levels in the ume sediments.
Although the combination of medium sediment dilution and medium bedform number (S3 : B3) had slightly higher average bacterial diversity than the other medium dilution (S3) samples (S3 : B0 and S3 : B6), this difference was not signicant in a Tukey post hoc analysis. The overall difference in Fisher's alpha bacterial diversity was less pronounced between medium and high sediment dilution levels (S3 and S6) than between low and medium dilutions (S1 and S3, Fig. 3b). Since Fisher's alpha index gives more weight to the total number of species, and less to the number of individuals in each species, 47 the medium and high dilutions (S3 and S6) were similar in taxonomic richness, but the bacterial community composition could still be considerably different. The dispersion of the Fisher's alpha index differed between the sediment dilution levels: the low dilution (S1) had the widest range in values (RSD 18%), followed by the medium dilution (S3; RSD 15%) and nally the high dilution (S6; RSD 13%). Due to the extinction effect, the communities in the low dilution umes (S1) evolved from the original heterogeneity of river sediment causing a wider spread of diversities, while the communities in the medium and high dilution umes (S3 and S6) evolved from inocula with similar pre-selected bacterial communities with lower taxonomic richness.
Due to the dilution-to-extinction method, the number of bacterial cells at the start of pre-incubation (day À12) and corresponding number of copies of the 16S rRNA gene, was theoretically lower in the medium and high sediment dilutions (S3 and S6) compared to the low dilution (S1). This was not the case anymore at day 21. Sediment treatments at day 21, obtained an average of 4.7 Â 10 5 copies of the 16S rRNA gene per gram of sediment dry weight aer real-time PCR (Fig. S8 †) and were not signicantly different between the sediment dilution levels (ANOVA; p > 0.05). Further, there was no signicant effect of the bedform variable on the copy numbers of 16S rRNA gene (ANOVA; p > 0.05). Thus, substantial bacterial growth in the high dilutions (S3 and S6) was necessary during the preincubation phase to reach a similar number of copies to the low dilution (S1) at day 21, suggesting that the pre-incubation was successful. This reveals that the main effect of the sediment dilution was likely caused by differences in bacterial diversity, and not differences in biomass. Still, it should be pointed out that the copy numbers of 16S rRNA genes do not necessarily correlate to bacterial biomass or number of bacteria. 48 First, because the number of copies of the 16S rRNA gene in the chromosome is taxon specic (from 1 to more than 10). Second, because the 16S rRNA gene copy number for a specic taxon can differ between growth phases. Our results for bacterial diversity and 16S rRNA gene copies show the status of bacterial communities in the umes at day 21, which is representative of the period where most micropollutant attenuation occurred, yet it is possible that the communities evolved over the duration of the test and in response to environmental conditions such as sunlight, temperature and addition of nutrients.

Effect of sediment dilution and bedforms on hyporheic exchange
We aimed to quantify differences in surface water À porewater exchange ux [L d À1 ] and exchange volume [L], as well as average residence time [d] between bedform levels by the salttracer dilution test 46 aer termination of the attenuation phase (ESI †).
The test showed that the sediment dilution-variable signicantly inuenced the exchange ux (ANOVA; p < 0.05), while the bedform-variable had no signicant effect at this late point in time. Neither the bedform, nor the sediment dilution variable inuenced the exchange volume or the residence time signicantly (Fig. S7 †). In contrast to what we found, we expected that an increasing number of bedforms would induce higher hyporheic exchange due to advective pumping 20 and this way increase contact of micropollutants to hotspots of favorable turnover conditions. Previous studies have found differences in hyporheic exchange in umes with at beds and in umes with bedforms, containing the same sediment. 21 We attribute the lack of signicant effect of bedforms on exchange parameters to the fact that the salt-tracer dilution test was conducted at the nal point of the experiment (days 78-84), a time by which the umes had changed from their initial setup conditions ( Table 3). Formation of biolms and algae and settling of ne particulate matter, more pronounced at the end of the test, likely inuenced the permeability of the sediment by clogging and affected hyporheic exchange. Bedform heights had dropped by 19% compared to the beginning of the setup due to gradual erosion, potentially reducing exchange ux in B3 and B6 treatments (Table 3). Small ripples had formed in the umes without bedforms caused by turbulence in the ume curves, which likely increased the exchange ow in the B0 Fig. 3 Fisher's alpha diversity index at day 21 plotted by (a) bedform and (b) sediment dilution levels. The bacterial diversity was significantly different between the sediment dilutions (p # 0.01), whereas the bedform variable did not significantly affect the diversity.
treatments over time. The average ow velocity had decreased by 19% from initial bedform setup (day À3) to day 82 due to decreasing pump performance (Table 3). Differences in ow velocity might also have contributed to the effect of the sediment dilution treatment on the exchange ux at the time of the salt tracer test. By then, S1 treatments had the lowest average ow velocity (6.1 cm s À1 ), followed by S3 treatments (6.8 cm s À1 ) and S6 treatments (7.9 cm s À1 ), although these differences were not signicant (ANOVA; p > 0.05). The lower ow velocities in S1 umes might have been caused by clogging of the pumps due to oating algae, which appeared in the second half of the attenuation phase in S1 umes (Fig. S10 †).
While the salt tracer dilution test showed that hyporheic exchange did not differ between bedform treatments at the nal point of the experiment, it can not be ruled out that the bedforms caused differences in hyporheic exchange as anticipated in the beginning of the experiment and most relevant time for degradation (Fig. 2a). Hydrodynamic modelling could be used to calculate bedform-induced hyporheic exchange under initial setup conditions in the different bedform treatments. 49,50 However, this was outside the scope of the present study. A follow-up study will cover this endeavor.

Response surface model to evaluate dissipation of acesulfame and carbamazepine
We observed dissipation of both ACS and CBZ in the umes. ACS was dissipated by 83-99% in S6 umes, by 97-100% in S3 umes and 100% in all S1 umes, at day 78. A lag-phase period of around 20 days in the attenuation of ACS can be clearly identied in umes with S6 sediment (highest dilution level) (Fig. 2a). The lag phase can be interpreted as the time it takes for the bacterial community to adapt to transform ACS, as the bacteria with that function may originally have been less abundant in sediment S6. The DT50s of ACS were between 2.1 and 36 days (Fig. 4a and b), and all umes had a kinetic constant signicantly different from zero. In contrast, the concentrations of CBZ in ume surface water had decreased by 37-61% in S1 treatments at day 78, and a smaller decrease was observed in umes with S3 treatment (16-36% removal) and S6 levels (14-30% removal). The DT50s of CBZ ranged between 65 and 838 days (Fig. 4d and e). The concentrations of CBZ over time in two of the umes (treatments S6 : B0 and S3 : B3) could not be tted to 1 st order kinetics because no discernable attenuation was observed (Fig. 2b). DT50s of those umes were not included in further analysis nor in Fig. 4d and e. Of the 18 DT50s obtained for CBZ, 6 had kinetic constants that were not signicantly different from zero (p > 0.05) and 70% of the DT50s calculated for CBZ exceeded the 120 day persistence criteria for sediment in the REACH regulation. 51 ACS and CBZ were ubiquitously found in a eld study in River Erpe, the river from which the sediment was collected for the umes, with surface water concentrations in the mg L À1range and both compounds were more persistent relative to other micropollutants. 6 While DT50s of ACS observed in the River Erpe were in the same order of magnitude (4-30 days) as the DT50s in our ume setup, the in situ DT50s of CBZ were about one order of magnitude lower (4-13 days) compared to in our umes. Also, Writer et al. 52  Three in situ studies that tested ACS and CBZ found poor to no degradation of both compounds. 5,29,54 The variability in literature values of DT50s for both ACS and CBZ reect the complex interactions between dissipation processes and environmental conditions that in turn may inuence degradation rates.
The RSM of ACS explained roughly 90% of the variance of observed DT50s (adjusted-R 2 of 90.2%) and the overall model was signicant, meaning that the coefficients together tted the DT50s better than just the mean (F-statistic: 35.79 on 5 and 14 degrees of freedom, p # 0.05, see Table 4 and Fig. 4c). The ANOVA showed that both rst, second order and interaction terms were signicant for the RSM. However, the lack of t of the model was also signicant, which means that the difference between the model predictions and the average of measurements for each variable combination (S1 : B0, S3 : B0, S6 : B0, etc.) was large compared to the pure error that was expected due to chance. The difference between predicted vs. observed DT50s was higher in S3 and S6 levels, which also have the highest variations in DT50s (Fig. 4b). Therefore, variation in the DT50s was not adequately explained by the variables S and B in the ACS-RSM. The CBZ-RSM was found signicant and had an adjusted-R 2 of 44.4% (Table 4 and Fig. 4f). Only the linear term of the model was signicant for the t and thus the quadratic terms were not signicantly improving the model performance.  The overall interpretation of the CBZ-RSM model is that only the sediment dilution variable (S) was a good explanatory variable of the DT50s of CBZ in our ume set-up.
The main removal process of ACS and CBZ in this experiment was expected to be biodegradation. Although CBZ was oen found persistent and thus not biodegradable in sediment-water systems, 8,13,23,55 in experiments comprising longer incubation times, slow biodegradation has been observed previously. 9,56-58 Biodegradation of ACS in WWTPs was only recently reported and has evolved either by independent evolution or global spreading of organisms or genes responsible for the biotransformation pathway. 37 Bacteria in sediment of River Erpe have been chronically exposed to ACS via the effluent of the local WWTP and could have developed the capability to degrade ACS similar to the WWTP communities. Neither photolysis nor hydrolysis has been relevant for ACS and CBZ attenuation in previous studies. 8,59 Sorption is expected to inuence the dissipation from water for compounds with high log K OW as they sorb to hydrophobic fractions of organic matter, or for positively charged compounds due to electrostatic interactions with negatively charged binding sites of clay minerals, biolm or organic matter. 52,60,61 Low sorption potential of ACS is expected, as ACS is negatively charged at ambient pH of around 8 and has a log K OW of À0.6. Accordingly, previous studies showed low sorption 62 and a low retardation coefficient (max. 1.2) 30 in sediments. All ume sediments had low organic carbon content (TC < 0.01%) and low content of ne particles (silt + clay < 1%), which generally rendered them poor sorbents. CBZ is neutral, has a log K OW of 2.77, and has been shown to sorb to organic matter 63 and to have relatively high retardation coefficient in sediment (3.6). 30 However, a test following OECD guideline 106 64 to examine potential sorption of CBZ to the ume sediments showed no sorption of CBZ to S1, S3 or S6 sediments (see ESI for details, Fig. S11 and S12 †). This can be explained mainly by the low carbon content of the ume sediments. Therefore, sorption was not a relevant attenuation process of CBZ or ACS in our experimental setup.

Impact of sediment dilution treatments on dissipation half-lives of acesulfame and carbamazepine
The level of sediment dilution, and hence the bacterial diversity, had a signicant inuence on the dissipation of ACS and CBZ in our ume experimental setup. ACS had signicant linear and quadratic coefficients (Table 4) of the sediment dilution variable (S). Both coefficients were negative and indicate that faster degradation was observed in umes with the lowest sediment dilution (S1), i.e. umes with the highest bacterial diversity (Fig. 4b). The CBZ model had the linear term of the sediment dilution (S) signicant and negative, which conveys that the sediment dilution had an effect on the DT50s and the lower sediment dilution (S1) generated shorter DT50s (Table 4, Fig. 4e).
Manipulating the bacterial community without changing chemical and physical background conditions was challenging. The difference in pH ($0.5) and O 2 ($3%) in the low dilution (S1) compared to the medium and high levels (S3 and S6) was signicant, but likely too low to inuence the degradation of micropollutants considerably. In contrast, the differences in nutrient concentrations and nutrient depletion over time were high between the different sediment dilution treatments (Fig. S6 †). The low sediment dilution level (S1) had the highest depletion of nutrients and lower long-term availability of nutrients for bacteria. Hence, DT50s decreased in the lower sediment dilution although nutrient concentrations were lower. Thus, the deciencies in nutrient concentrations could have counteracted the positive effect of the high bacterial diversity in S1 which implies that the effect of the sediment dilution on Table 4 Response surface model (RSM) coefficients fitted to a quadratic equation and p-value, and the ANOVA of each model component (first, second order, and interaction terms) for the sediment dilution (S) and bedform (B) variables. The coefficients b 1 and b 2 correspond to the linear effects of the S and B variables respectively, b 3 to the interaction term SB and b 4 and b 5 to the quadratic effects S 2 and B 2 (see eqn (1) DT50s of ACS and CBZ might have been even higher at equal nutrient availabilities. Therefore, despite some differences in chemical conditions, the signicant inuence of sediment dilution levels on degradation of ACS and CBZ can be attributed to the differences in bacterial diversity. A positive association between biodegradation rates and biodiversity has been reported for bacteria in WWTPs. 33,34 The faster dissipation of ACS and CBZ in sediment with the highest diversity (S1) is consistent with a mechanism referred to as "sampling effect". Briey, if the biodegradation of a micropollutant is a rare function performed by few bacteria species, which is likely for slowly degrading compounds, the bacteria capable of degrading these compounds are more likely found in sediment with the highest bacterial diversity. In addition, the "complementarity effect", meaning that higher diversity leads to higher functionality and more cooperating specialist strains, might increase complexity and efficiency in transformation pathways of particular micropollutants. 65 The "sampling" and "complementarity" effects can be applied also to describe the increasing variability of DT50s of ACS in the higher sediment dilutions (S3 and S6): key species necessary for the efficient degradation of ACS are less likely to be present or randomly found in bacteria communities with low diversity. The variation in bacterial communities in the sediment dilutions thus might be responsible for the lack of t of the RSM of ACS.

Impact of bedform treatments on dissipation half-lives of ACS and CBZ
The presence of bedforms increased degradation of ACS as the linear coefficient for the bedform variable (B) was signicant and negative for ACS ( Fig. 4a and Table 4). This indicates that despite no measurable differences in exchange at the end of the experiment, at times most relevant for ACS degradation, the bedform treatment induced different levels of hyporheic exchange as hypothesized. The redox-sensitivity of ACS might have contributed to the nding of this effect in contrast to CBZ. 66,67 In addition to more intense transport, hyporheic exchange results in altered redox zonation along the ow paths, primarily by transport of dissolved oxygen into the sediment. 68,69 ACS was previously reported to degrade effectively under oxic and denitrifying conditions 66 and in the hyporheic zone. 30 Therefore, higher presence of bedforms results in better degradation of ACS by increasing the transport of ACS into the hyporheic zone and by providing qualitatively favorable redox conditions for ACS degradation. The coefficients in the RSM showed that the effect of the bedform variable (B) on the DT50s was smaller than the effect of the sediment dilution (S) and the interaction between the two variables (S and B), was signicant and positive for ACS (Table 4, Fig. 4a). Therefore, the combination of high sediment dilution (low bacterial diversity) and absence of bedforms (low hyporheic exchange), had a negative effect on the attenuation of ACS and resulted in longer DT50s (Fig. 4c). The signicant interaction between sediment dilution and bedform also implies that the effect of bedform treatment was higher at higher sediment dilution (Fig. 4c), indicating that the strong impact of high diversity on the dissipation rate of ACS in the low dilution treatment might have covered the effect of the bedform variable.
For CBZ, no effect of bedforms on dissipation rates could be shown, because the coefficients in the RSM of the bedform variable (B) and the interaction coefficient were not signicant (p > 0.05, Table 4, Fig. 4d). In two of the umes, one from level B0 and one from B3, it was not possible to calculate DT50s, and they were excluded from the RSM calculations. The absence of these two potentially long DT50s could have additionally masked an effect of the bedform variable. Given the generally slow degradation of CBZ, the effect of the bedform levels might have been too low to be detectable in the present setup.

Conclusion
We showed that this elaborate ume setup can be used to study the degradation of micropollutants under more natural conditions than batch experiments without having to sacrice the repeatability and control over test conditions. Our experimental design highlights the importance of hyporheic exchange, as higher number of bedforms can signicantly increase the attenuation of particular micropollutants, as shown for ACS in the present case. Furthermore, our results reveal that bacterial diversity of river sediment has a paramount effect on the degradation of micropollutants. This particularly shines new light on the interpretation of DT50s in literature, which have been obtained from various experimental setups in which the composition of the bacterial communities have usually been ignored.
In the future, the setup we presented can be applied to other parameters that inuence degradability, for instance sediment properties, nutrient concentrations or plant cover. For future studies, we suggest a set of preliminary studies or modelling to test the suitability of appropriate treatment levels for all variables tested. For instance, higher effects of and differences in hyporheic exchange can be ensured by preliminary hydrodynamic modeling of bedform morphologies. A second important improvement would be the continuous monitoring of chemical parameters to maintain comparable nutrient and chemical conditions. We used coded variable levels to investigate the suitability of the experimental design, but future studies could enhance the use of this design to quantify the effect of variables on DT50s of micropollutants.
The use of a central composite face design to t a response surface model provides a robust statistical method to study the response of micropollutant attenuation to certain environmental drivers and highlight bacterial diversity as a disregarded factor in testing persistence of micropollutants. To the best of our knowledge, no other ume study includes a full experimental design with replicates. This study guides a way forward for more elaborate experimental setups that address how environmental processes affect the fate of micropollutants in rivers.