Bacterial Diversity Controls Transformation of Wastewater-Derived Organic Contaminants in River-Simulating Flumes

Hyporheic zones are the water-saturated flow-through subsurfaces of rivers which are characterized by the simultaneous occurrence of multiple physical, biological, and chemical processes. Two factors playing a role in the hyporheic attenuation of organic contaminants are sediment bedforms (a major driver of hyporheic exchange) and the composition of the sediment microbial community. How these factors act on the diverse range of organic contaminants encountered downstream from wastewater treatment plants is not well understood. To address this knowledge gap, we investigated dissipation half-lives (DT50s) of 31 substances (mainly pharmaceuticals) under different combinations of bacterial diversity and bedform-induced hyporheic flow using 20 recirculating flumes in a central composite face factorial design. By combining small-volume pore water sampling, targeted analysis, and suspect screening, along with quantitative real-time PCR and time-resolved amplicon Illumina MiSeq sequencing, we determined a comprehensive set of DT50s, associated bacterial communities, and microbial transformation products. The resulting DT50s of parent compounds ranged from 0.5 (fluoxetine) to 306 days (carbamazepine), with 20 substances responding significantly to bacterial diversity and four to both diversity and hyporheic flow. Bacterial taxa that were associated with biodegradation included Acidobacteria (groups 6, 17, and 22), Actinobacteria (Nocardioides and Illumatobacter), Bacteroidetes (Terrimonas and Flavobacterium) and diverse Proteobacteria (Pseudomonadaceae, Sphingomonadaceae, and Xanthomonadaceae). Notable were the formation of valsartan acid from irbesartan and valsartan, the persistence of N-desmethylvenlafaxine across all treatments, and the identification of biuret as a novel transformation product of metformin. Twelve additional target transformation products were identified, which were persistent in either pore or surface water of at least one treatment, indicating their environmental relevance.


SI.A Compound properties
Reference standards and isotope-labeled internal standards used for the chemical analysis at Stockholm University (SU) and the Swiss Federal Institute of Aquatic Science and Technology (Eawag) were purchased from CDN Isotopes (Canada), Dr. Ehrenstorfer (Germany), HPC Standards (Germany), LGC Standards (Switzerland), Molcan (Canada), MolPort (Latvia), Monsanto (Belgium), Novartis (Switzerland), Riedel-de-Häen (Germany), Sigma-Aldrich (Switzerland/Germany) or Toronto Research Chemicals (Canada) at purities ≥ 95% (analytical grade) (see Table SI.A-1 and Table SI.A-2 for details). Reference standards used to fortify watersediment test flumes were either purchased from Sigma-Aldrich (UK) or Toronto Research Chemicals (Canada) at purities ≥ 96% (details in ESM.A). NANOpure water (Eawag) was generated with a lab water purification system (D11911, Barnstead/Thermo Scientific, USA), LC/MS grade water at SU was generated with a Milli-Q water purification system (Merck KGaA, Germany) or purchased from VWR (Germany). Methanol (Optima™, Fisher Scientific, Switzerland or Merck KGaA, Germany) was of LC/MS grade, formic acid and acetic acid of analytical grade (≥ 98%, Merck KGaA, Germany and ≥ 99.7% Sigma-Aldrich, Germany).

SI.A.a Parent compounds
Major microspecies at pH 8.3 (average pH across flumes) were calculcated using the ChemAxon's Calculator (cxcalc) in batch mode.
S-9  Pore water samplers (10 cm length, 0.15 μm pores, RHIZON FLEX, Rhizosphere Research Products, Netherlands) were glued to plastic holders and thus positioned at equal heights (1.5 cm above the flume bottom) inside the sediment. In B3 and B6 flumes samplers were placed in the 2nd bedform and a 2nd sampler was placed in the the 3rd bedform as backup in case of clogging. In B0 flumes, pore water samplers were installed at the same position but inside the flat sediment. Flumes 4 and 8 were equipped with additional samplers in bedform 1. In bedform 1 of flume 18, three samplers were installed along a hypothetical hyporheic flowpath in 4 cm distances. Due to the small extraction volumes (15 mL per sampling event) relative to the pore water volume exchanged per day (ca. 10 L for 3 and 6 bedforms and about 0.5 L in flat sediment) [30] and the distribution of water extraction over the entire channel cross section we expected no substantial disturbance of flow paths or overall residence times by sampling. Photographs of pore water samplers and holders are provided in Fig. SI.B-2. S-14

SI.B.a Sorption
We generally assume that sorption and photolysis played a minor role compared to biodegradation for dissipation of test compounds in the present experimental setup. For anionic and neutral compounds, organic matter is the main potential sorbent 2 . However, the total carbon content of the three sediment mixtures was <0.08% and thus the dissipation of spiked contaminants through sorption to sediment organic matter was likely to be negligible.
This was confirmed by a sorption test of CBZ (neutral speciation at flume pH), IBU (anionic) and SMX (anionic) to the flume sediments (see figures below; sorption of CBZ has previously been addressed by Jaeger et Coll 1 . CBZ has one of the highest log DOW (2.77) among the investigated compounds (only BENP and IRB feature higher log DOW). Hence, most compounds are less likely to partition into organic matter than CBZ. Since the sorption test showed no considerable sorption of CBZ, this probably accounts for all other compounds with lower log DOW.
Likewise, the low sorption of IBU and SMX indicates, that there was generally a low sorption potential for anions.
Cationic compounds typically tend to sorb to negatively charged surfaces 3 . Sediment-water column tests previously showed highest retention of cations by sorption 4 . Analysis of the cation exchange capacity, however, confirmed that our flume sediments were poor sorbents for cations (median cation exchange capacity of 0.6 cmol kg -1 , max: 3.6). The reason for this was probably the low fraction of fine mineral material (particle size <0.063 mm was below 1%) in the sediment and the low carbon content 1 . Thus, sorption was generally of little importance for most compounds. Any observed differences in the dissipation behavior of BENP and IRB (both have higher log Dow than CBZ) across the different sediment dilution levels must be attributed to biodegradation due to very similar sediment properties of the sediment mixtures 1 .
Equilibrium concentrations of carbamazepine (CBZ), ibuprofen (IBU) and sulfamethoxazole (SMX) in the water phase (Cw) and in the sediment phase (Cs) in the sorption test. Freundlich isotherms were not fitted to data from sediment flumes (S1, S3 and S6) due to very low sorption and thus large data scatter in equilibrium concentrations.
Negligible sorption recorded for sediment taken from flumes with sediment dilutions S1, S3 and S6.

SI.B.b Photolysis
The tent shielding the flumes from weather consisted of a white, UV-protected polyethylene plastic sheet (170 g m -2 , Heavy Duty Tent, KMS direct, UK). Similar material was found to block the largest fraction of UVA and UVB radiation 5 . Also the photosynthetically active radiation (PAR, 400-700nm) inside the tent was reduced to 68.7 µmol m -2 s -1 , which is considerably low compared to 400-1400 µmol m -2 s -1 , the estimated hourly PAR average in the UK 6 . Additionally, UVA and especially the powerful UVB radiation are often only considered relevant for direct photolysis of organic compounds in aquatic systems, whereas sunlight in the visible spectrum hardly contributes to direct photolysis 7,8 . Indirect photolysis was also unlikely in the present setup, as the common photosensitizers nitrate and nitrite, as well as dissolved organic matter were only present in low concentrations 1 .
In a study comparing photolysis of IBU, METP and CBZ in different solutions and under radiation at differing wavelengths it was demonstrated, that all three compounds were hardly degraded (DT50s > 400 days) under UVA-visible light in pure water, which is similar to the exposure conditions in the tent 8 . Therefore, we generally consider photolysis a process of low importance for the dissipation of organic compounds in the present setup. S-16

SI.C.b Target analysis
For the quantification of target analytes (30 parents, 27 TPs), 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 Scientific, USA). Peak integrations were reviewed manually. To each target analyte, a matching IS was assigned (internal standard method). If the matching IS was not available, an IS with a similar retention time was selected. Linear 1/x-weighted calibration curves were generated by fitting the analyte concentration (x) against the STD-to-IS peak area response ratio (y) without forcing the fit through zero.

SI.C.c Suspect screening of biological transformation products
To screen flume SW for suspected biological TPs, raw data obtained by RP-LC-ESI-HRMS/MS was submitted to Compound Discoverer (version 2.1, Thermo Scientific). This allowed for the automated detection (minimum intensity: 10'000, 30% intensity tolerance) and grouping (ESI adducts, isotope peaks) of peak features (characterized by m/z, retention time and intensity) with subsequent compound assignment (elemental composition based on predefined maximum elemental counts within a mass tolerance of 5 ppm). Mass spectra acquired in positive and negative ESI polarity mode (contained within the same raw data due to ESI polarity switching) were extracted and processed in separate workflows. The underlying suspect list consisted of biological TPs that were either predicted from the molecular formulas of the parent compounds using an Excel spreadsheet (transformations: oxidation, reduction, cleavage, conjugation), or from the molecular structures of the parent compounds using the Eawag-Biocatalysis/Biodegradation Database Pathway Prediction System (Eawag-BBD PPS, http://eawag-bbd.ethz.ch/predict/, settings: relative reasoning, no immediate rules, aerobic and anaerobic transformations allowed, 3 generations). Output of the suspect screening were two lists of assigned compounds (one per ESI mode), from which compound time-series were extracted for every experimental S-17 treatment (1.3 M time-series in total, peak areas as average/median among replicate flumes). Time-series were prioritized by two approaches, i.e. (1) by only considering series with MS2 information (0.2 million), and (2) by only considering Eawag-BBD PPS predicted compounds independent of the MS2 information (0.6 million). In subsequent steps, target analytes and IS were excluded, and time-series only further evaluated if the compound matched the suspect list. Another criterion was the absence/minimal presence (maximum peak area in unspiked flumes / minimum peak area in time-series ≤ 20%) of the suspected TP in unfortified flumes. In a first run, 15 unspiked flume samples were considered for this comparison. Because one unspiked sample appeared crosscontaminated, the respective sample was excluded, and the comparison repeated with 14 unspiked samples (approach 1). This time, the peak area ratio threshold of 20% (see above) was only applied, if the suspected compound was present in 7 or more unspiked samples. The final (prioritized) time-series were assigned to 50 hierarchical clusters using the hclust function in R 9 , and clusters grouped (group 1: increase, 2: increase and decrease, 3: other trends). For parent compounds without tentatively identified suspect TP, literature was surveyed, and potential candidates were looked up in Compound Discoverer.

SI.C.d Compound Discoverer Settings
The Compound Discoverer 2.1 (Thermo Scientific, USA) workflow is presented in Fig. SI.C-3, the detailed settings in Table SI.C-7 below. S-18

SI.D Instrumental analysis at 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. Details with regards to sample preparation standards and reagents can therefore be found S-20  S-23

SI.G.a DNA extraction
DNA was extracted following the rapid method for the extraction of total nucleic acids from environmental samples 11 . Co-extracted RNA was removed using DNase-free RNase (Promega, Mannheim, Germany). DNA concentration was subsequently determined with Quant-iT® PicoGreen DNA assay kit following manufacturer's protocol (Invitrogen, Germany) and the Tecan Infinite® plate reader (Tecan, Switzerland).

SI.G.b Quantification of total 16S rRNA genes
The bacterial 16S rRNA gene copy numbers were quantified using real-time quantitative PCR (qPCR).

SI.G.c 16S rRNA gene sequencing
Sequencing of the bacterial 16S rRNA gene fragments was implemented using the Miseq Illumina® sequencing platform. An initial PCR amplification step was performed at LGC Genomics GmbH (Berlin, Germany). Briefly,

S-24
The Illumina sequence data were submitted to the NCBI sequence reads archive under the accession number PRJNA531245.

SI.G.d Statistical analyses
To check the efficacy of pre-incubation phase on bacteria regrowth following sediment dilution, the 16S rRNA gene copy numbers across treatments were compared using analysis of variance (ANOVA) and Tukey test.
Bacterial alpha diversity as a function of sediment dilution and bedform elements during attenuation phase was determined based on species richness, Shannon diversity and Shannon evenness indices using rarefied sequence data based on OTUs defined at 97% similarity applying post-hoc Tukey test in PAST v3.15 13  DESeq2 multifactorial design in R was employed to determine genera whose abundance changed significantly between amended and unamended controls with incubation time as a covariate. Genera with a Log2FoldChange > 0 were considered enriched while a Log2FoldChange < 0 indicated that the abundance of the genus decreased in the amended relative to unamended controls. Differential abundance was considered significant at (adj. p < 0.01) using the Benjamin-Hochberg correction.
S-25 SI.G.e Detailed results of the bacterial analysis   Bacterial community composition at the phylum level in relative abundance values (%) in the respective treatments sampled at days 0, 21 and 56. "Others" represents all phyla whose relative abundance was below 0.5%.

SI.K.a Pooling Eawag and SU concentration data and their use in the RSM
If a target analyte was covered by both methods, the chemical data was pooled: averaged concentrations were used in case of two data points and data below the limit of quantification (LOQ) set to 0. With regards to parent compound concentrations used for the response surface model (RSM) a slightly different method of pooling was applied due to generally higher concentrations: for each compound, the first concentration time point in a flume that reached levels below LOQ (the higher LOQ in case they were different at SU and Eawag) was substituted by the LOQ. Subsequent time points, which were mostly below LOQ, were excluded from further analysis.

SI.L Background concentrations of target analytes in blank flumes
In addition to the target analytes (Table SI. L-19), we analyzed two stable compounds -tramadol (TRA) and oxazepam (OX) that were not part of the spiked parent compounds -to test for release from Erpe sediment in all flumes and throughout the experiment. TRA has been suggested as a chemical indicator of water contamination with sewage 15 while OX was shown to be highly persistent in aqueous sediment 16 and both compounds were previously found in Erpe pore and surface water [2,14,16]. Both compounds were found at low concentrations in pore and surface water of many S1 flumes, supporting their suitability not only as wastewater tracers 15 , but also as potential indicators of sewage contaminated sediments, e.g. when assessing soil contamination of decommissioned sewage farms or after land application of raw or treated sewage sludge. Their levels remained low throughout the experiment (TRA: <252 ng L -1 , OX: <99 ng L -1 ). Of the spiked compounds, very few were detected in blank flumes, mostly at levels just above their respective LOQs and mostly in the first days of the experiment (day 1 to 14) in both surface and pore water (Table SI. L-19). Since all parent compounds studied in this experiment are expected to be less or similar persistent than TRA we expected no significant effect on their concentrations by remobilization from Erpe sediment. Overall, the low organic carbon content of the flume sediment mixture (TC < 0.01%) and the low content of fine particles (silt+clay<1%), made sorption and desorption unlikely 1 . S-41

SI.M.a Molar sum concentrations of PC and TPs
The molar concentration curves of all target TPs (mean cTP) in the different S treatments initially appeared ambiguous (see below): formation of TPs in S1 treatments started earlier, showed a steeper rise, and flattened earlier compared to S3 and S6 treatments. In comparison to S1 treatments, the onset of TP formation in S3 treatments was shifted by 3-4 days (or more) and a further 2-10 days in S6 treatments. Measured TP concentration maxima increased in the order S1<S3<S6 ranging from 7.1 × 10-8 to 8.0 × 10 -8 mol L -1 and thus differing by less than 15%. At day 56 (last full pore water analysis) the differences appeared negligible. The highest recorded cTP (8.0 × 10-8 mol L -1 ) corresponded to 6.2% of the spiked TrOCs (1.3 × 10 -6 mol L -1 ). A closer look revealed that TP.VAA was dominating the mole fractions, accounting for up to 80% of the total molar TP mass (S1 at day 56) and increased with increasing diversity. Omitting TP.VAA laid open more meaningful cTP curves (see below); we now saw a very clear difference between all three S levels. The maximum cTP in S1 (1.75 × 10 -8 mol L -1 ) was detected on day 1 and remained then almost constant throughout the experiment while cTP max. in S3 and cTP max. in S6 equaled 3.9 × 10 -8 and 5.7 × 10 -8 mol L-1 and were thus 125% and 225% higher, respectively. We therefore suspect that high bacterial diversity leads to a more complete degradation of parent compounds (excluding VAL and IRB). Highest molar mass ratios of TPs (cTP cPC-1) were 0.71, 0.35 and 0.29 including TP.VAA and 0.15, 0.12, 0.20 without TP.VAA for S1, S3, S6, respectively, and were reached at different sampling time points (see below). Only in S1 surface water samples collected on day 56 the total molar concentration of our target TPs surpassed the residual amount of spiked parent compounds and this was again mostly due to high TP.VAA concentrations.     24 Therfore, this TP could have been more stable in the mostly anaerobic pore water of our flume systems (confirmed by O2 profile measurements, data not shown), while it was further degraded in aerobic surface water. Alternatively, its formation rate in pore water could have exceeded its transport rate to surface water.

SI.M.g Biodiversity affects transformation of hydrochlorothiazide
We observed significant linear and quadratic impacts of bacterial diversity on dissipation of hydrochlorothiazide.
Most of the dissipated mass of hydrochlorothiazide was transformed to either 4-amino-6-chloro-1-3benzenedisulfonamide (TP.ABS) or chlorothiazide (TP.CTZ, Fig. 4) -ca. 48 % and 10 % on average after 30 days (average DT50), respectively. Both compounds were previously described as hydrochlorothiazide products and we found no additional suspect TPs of hydrochlorothiazide. Whereas transformation of hydrochlorothiazide was affected by the biodiversity the formation of TP.ABS was similar at all three S levels which supports results from previous studies describing it to form through abiotic hydrolysis or photodegradation. [25][26][27] In addition, we found similar TP.ABS concentrations in both surface and pore water at similar time points which suggests hydrolysis as major attenuation process. From days 42 to 56 we observed a concentration drop in surface-but not pore water, possibly attributable to photolysis (mean solar radiation over 78 days: 68.7 µmol m -2 s -1 PAR). 1 In contrast, more than two-fold higher average concentrations of the second product (chlorothiazide) were observed, along with an earlier onset of formation in higher-compared to lower diversity systems. Levels were also slightly higher in pore water which suggests biotransformation as the major process. Both products showed signs of further degradation in pore water while TP.ABS was stable in surface water of high diversity systems. Overall, this shows a complex interplay of biotic and abiotic processes in transformation. S-48

SI.N Concentration time trends
The following concentration time trends are alphabetically ordered by parent compound (PC) group. Within such a group, first, the PC time trends are shown in surface water (SW) and (PW), followed by the time trends of the target transformation products (TP) in SW and PW (if available), and suspected transformation products (sTP) in SW and PW (if available).

SI.O Details on tentatively identified suspect TPs
Only positive ESI mode.