Microbial retention and resistances in stormwater quality improvement devices treating road runoff

ABSTRACT Current knowledge about the microbial communities that occur in urban road runoff is scarce. Road runoff of trafficked roads can be heavily polluted and is treated by stormwater quality improvement devices (SQIDs). However, microbes may influence the treatment process of these devices or could lead to stress resistant opportunistic microbial strains. In this study, the microbial community in the influent, effluent and the filter materials used to remove dissolved heavy metals from two different SQIDs were analyzed to determine microbial load, retention, composition, and mobile resistance genes. Although the microbes were replaced by new taxa in the effluent, there was no major retention of microbial genera. Further, the bacterial abundance of the SQIDs effluent was relatively stable over time. The heavy metal content correlated with intl1 and with microbial genera. The filter media itself was enriched with Intl1 gene cassettes, carrying several heavy metal and multidrug resistance genes (e.g. czrA, czcA, silP, mexW and mexI), indicating that this is a hot spot for horizontal gene transfer. Overall, the results shed light on road runoff microbial communities, and pointed to distinct bacterial communities within the SQIDs, which subsequently influence the microbial community and the genes released with the treated water.


INTRODUCTION
Industrialization and technological advancement have put an increasing burden on the environment by releasing large quantities of hazardous contaminants inflicting serious damage on the ecosystem. Traffic area runoff is widely recognized as a major transport vector of pollutants released in the urban environment, as it collects precipitation and snowmelt-related discharges of mostly impervious surfaces (e.g. sidewalks, parking lots, feeder streets, major roads and highways). The majority of pollution caused by traffic area runoff originates from vehicle brake emissions, tire wear, lubricating oil and grease, pet waste and atmospheric deposition on the road surface (Ball, Jenks and Aubourg 1998, Legret and Pagotto 1999, Adachi and Tainosho 2004. The chemical quality of traffic area runoff has been analyzed and indicated the presence of different contaminants including heavy metals, polycyclic aromatic hydrocarbons, polychlorinated biphenyls (PCB), and other organics (Eriksson et al. 2005, Huber, Welker andHelmreich 2016). Heavy metals in traffic area runoff continue to create serious global health concerns, as they persist in suspended particulate matter and have the ability to travel long distances through water-air-soil systems with subsequent risks to human health (Markiewicz et al. 2017).
The awareness of stormwater runoff pollution and increasing concern about its impact on the environment, has led to the development of stormwater control measures (SCMs) for pollution control and contaminant retention from urban road runoff. One SCM to minimize the contaminant emissions to the environment is the usage of sustainable urban drainage systems (Dierkes, Lucke andHelmreich 2015, Lucke et al. 2017). These include decentralized technical systems, referred to as stormwater quality improvement devices (SQIDs) or manufactured treatment devices that treat stormwater runoff with a comparably low footprint and are particularly suitable in dense urban environments (Dierkes, Lucke and Helmreich 2015). SCMs have historically been constructed for pollutant and nutrient reduction from different environments Helmreich 2013, Huber, Welker andHelmreich 2016). Nonetheless, some studies on the effects of SCMs have also evaluated their function for bacteria removal; in particular, filterbased bioretention systems have shown fecal bacteria removal efficiencies between 50% and 70% with significant difference between inflow and outflow counts (Pennington, Kaplowitz andWitter 2003, Hathaway, Hunt andJadlocki 2009). Pathogenic bacteria, viruses and protozoa can be found in runoff (Pandey et al. 2014, Ahmed et al. 2019 and are transported to surface waters through sewer overflows, representing one of the major human health risks (Page et al. 2010, Ma et al. 2016.
While the microbial load of sewer overflows has gained considerable attention, microbes of traffic area runoff in general are scarcely investigated with only few exceptions. Due to the heavy pollution and harsh conditions, traffic areas and their runoff can be classified as an extreme environment. Early research looked mainly at microbial communities found in sediments of infiltration basins (Rotaru et al. 2012), on the effect of de-icing salts (Ostendorf, Rotaru andHinlein 2008, Ostendorf et al. 2009), the biofilm of road gutters (Hervé and Lopez 2020), or on the denitrification potential of road runoff receiving biofilters (Luo et al. 2020). More comprehensive analysis on the microbiology of road runoff are missing, and most of the microbes and their community functioning in this polluted environment remain unknown, although they are expected to be of relevance for the receiving water bodies (groundwater and surface waters) (Lee et al. 2020, Scharping andGarey 2020). Furthermore, understanding how antibiotic resistance genes (ARGs) are distributed along road runoff treatment processes is important because they pose a potential risk to public health. The location of ARGs on mobile genetic elements, such as integrons (e.g., intl1) makes the horizontal transfer of antibiotic resistance genes possible and easy to achieve among bacteria with same or diverse origins. Previous studies assessed the presence of Intl1 in the genome of bacterial isolates from wastewater and drinking water treatment plants (Allen et al. 2010, Lu et al. 2018 showing a high possibility of horizontal gene transfer of ARGs within these systems. Therefore, an investigation of these microbial communities can provide insights into adaptations of microbial communities to factors such as pH, contaminants and heavy metals (Kaevska et al. 2016, Liao et al. 2018, and shed light on potential microbial risks. This study is a pioneer study on the microbial community composition and its anthropogenic signatures in the form of class I integron gene casettes (intl1) of road runoff effluents and the filter materials of two SQIDs along a heavily trafficked urban road in Munich, Germany. We collected water samples for over seven months, and sampled the filter media of the SQIDs. The aims of this study were to: (i) identify the major taxa of road runoff and treated effluent and (ii) identify microbial risk factors associated with the mobile genetic element intl1. This establishes the basis for evaluating the microbial relevance in road runoff, its impact on receiving water bodies and how this impact may be modulated by current treatment systems.

Study site
In this study, we monitored two different SQIDs (D1, D2; Fig. S1, Supporting Information) on a heavily trafficked road in Munich (Germany,48 • 10'47'N 11 • 32'25'E) with an annual average daily traffic of 24 000 vehicles per day. Devices D1 and D2 are pre-manufactured SQIDs (SediSubstrator XL 600/12, Fränkische Rohrwerke Gebr. Kirchner GmbH & Co. KG, Germany; Drainfix Clean 300, Hauraton GmbH & Co. KG, Germany). Both SQIDs were installed at the same time and drained road runoff from catchments in close proximity, which consequently showed the same runoff properties. The main difference between the devices were that D1 used a primary sedimentation stage and downstream media filtration stage using an iron-based filter medium with lignite addition and the filter medium was permanently submerged, while D2 used direct filtration with a carbonate containing sand, which ran dry after each rain event. After the treatment, the water was percolated into the groundwater. The catchment areas of the devices were 1660 m 2 for D1 and 165 m 2 for D2.

Sampling and characterization
Water samples before (n = 21) and after SQID treatment (n = 20) were collected during a seven-month period from April to October 2019 in order to evaluate different seasonal changes (spring, n = 19; summer, n = 13; autumn, n = 9). Three different types of water samples were collected based on the position of sampling: Influent (I): inflow of road runoff to the SQIDs; Effluent after sedimentation and adsorption (ESA): effluent of SQID D1; Effluent of Filtration (EF) filtrated water samples of the SQID D2. The samples were withdrawn volume proportionally using automatic samplers (WS 316, WaterSam, Balingen, Germany). Permanent flow measurement using electro-magnetic flow meters (Krohne Optiflux 2300 C or 1300 C, Krohne IFC 300 C, DN250 for D1, DN25 for D2) enabled the volume proportional sampling. Sampling was triggered when flow exceeded for 1 min 0.4 L/(s·ha) and stopped when flow was below the threshold value for 15 min. Discrete samples (400 mL) were withdrawn after approximately 0.2 L/m 2 runoff volume each. The discrete samples of each sampling point and runoff event were merged in composite samples for further analysis. The number of discrete samples per sampling point and runoff event depended on the runoff volume and ranged from 2 to 62. The samples were kept in coolers at 4±1 • C and transported to the lab within 60 h. Electric conductivity (EC) and pH of the samples were analyzed following the standard methods 2510 B and 4500-H+, respectively (Bruno 2017). Total concentrations of chromium (Cr), copper (Cu), nickel (Ni), lead (Pb) and zinc (Zn) were determined after aqua regia digestion according to EN ISO 15587-1:2002. Cr, Cu, Ni, and  Filter material samples (> 500 g) were withdrawn from the SQIDs after approximately 2.75 years of operation, labeled FD1 for SQID D1 and FD2 for SQID D2. The surface layer (0-5 cm) in flow direction of the filter materials, which commonly contain most of the contaminants (Muthanna et al. 2007, Hatt, Fletcher and Deletic 2008, Al-Ameri et al. 2018, was sampled using ethanol cleaned plastic spatulae and a stainless-steel soil sampler. In addition, we took samples for the microbial community analysis in the middle (5-10 cm) and deepest layer (10-15 cm) in the flow direction of the filter materials. The content of Cr, Cu, Ni, Pb and Zn in the filter media were analyzed after inverse aqua regia digestion adapted from DIN EN 13 346:2001 with a HNO 3 : HCl ratio of 3:1 using the aforementioned ICP-MS and ICP-OES devices. The LOQs of Cr, Cu, Ni, Pb, Zn in the filter media were 5.0, 5.0, 2.0, 10.0 and 1.0 mg/kg, respectively. All analysis results for water and filter media samples below LOQ were substituted by the half of the respective LOQ value.
To assess the overall pollution level of the water samples, a water pollution index (WPI GFS ) was determined based on the German insignificance threshold values for evaluation of locally restricted groundwater pollution (Geringfügigkeitsschwellenwerte ,  Table S1), which are used to evaluate if a negative anthropogenic effect on groundwater quality is present, following eq. 1. This method is adapted from Bartlett et al. 2012 where [C i ] is the concentration of the substance i present in the sample, C i,GFS is the minor threshold value of substance i, and n is the number of analyzed substances. The heavy metals Cr, Cu, Ni, Pb, and Zn were considered in this analysis.

DNA Extraction and 16S rRNA Gene Amplicon Sequencing
Water samples (45 mL) collected from the different devices were centrifuged at 5000 rpm for 10 min and the cell pellets were stored at −20 • C, while the filter media samples were directly stored at −20 • C until DNA extraction. The DNA from both water (i.e. cell pellets) and filter media samples (200 mg) was extracted using the FastDNA Spin Kit for Soil (MP Biomedicals, Solon, USA), following the manufacturers protocol. The DNA concentration of the individual extracts was quantified by using the dsDNA Broad Range Assay kit (DeNovix, Wilmington, USA), following the manufacturers protocol, then stored at −80 • C until sequencing. The 16S rRNA gene amplicon sequencing was performed at ZIEL using the primers 341F/806R targeting mainly bacteria (Institute for Food & Health at Technical University of Munich, Germany). All the data are generated using a MiSeq sequencer (Illumina technology, v3 chemistry) following the protocol of (Reitmeier et al. 2020)

Data analysis and quality control
All 16S rRNA gene amplicons were processed using the opensource bioinformatic pipeline DADA2 (version 1.14.1) (Callahan et al. 2016) for R (version 3.6.0) (R Development Core Team R 2011). Demultiplexing and quality filtering were carried out in DADA2 using customized settings (truncLen = c (290−200), trim-Left = c(14,12), maxN = 0, maxEE = c(2,6)) after the removal of the primers sequence. Error rates were subsequently estimated from a set of subsampled reads (1 million random reads), and chimeric sequences were identified and removed from the demultiplexed reads. The exact amplicon sequence variants (ASVs) were taxonomically classified with a naïve Bayesian classifier using the Silva v. 138 training set (https://benjjneb.githu b.io/dada2/training.html, accessed August 2020). Negative controls, (i.e. Nuclease-free water) were included at every step of processing, from DNA extraction through the library preparation. A subset of control samples were sequenced in sequencing runs to verify that methodological errors did not impact results. Samples that shared dominant taxa with negative controls were removed as false positives from the dataset (n = 5).

Quantitative Polymerase Chain Reaction (qPCR)
A quantitative Polymerase Chain Reaction (qPCR) protocol was performed to quantify the number of 16S rRNA and intl1 gene copies within samples. 16S rRNA gene was amplified by 16S Forward 5 -GACTCCTACGGGAGGCWGCAG-3 ; 16S Reverse: 5 -GTA TTACCGCGGCTGCTGG-3 (Jaric et al. 2013). Intl1 was amplified with the intl1 primers from (Barraud et al. 2010). The qPCR for 16S rRNA was carried out with a reaction mixture containing 10.5 μL GoTaq R qPCR Master Mix (2X) (Promega, Madison, USA), 0.2 μM of each primer, 7.5 μL nuclease free water and 1 μL of template for a total volume of 21 μL. The 16S rRNA qPCR program consisted of 2 min at 95 • C, 40 cycles with 5 s at 95 • C, 30 s at 60 • C, while for intl1 the program was 4 min at 95 • C, 40 cycles with 10 sec at 95 • C, 45 s at 64 • C. Both were performed using the CFX96 thermocycler (BioRad, Hercules, USA). Calibration curves for intl1 were obtained using serial dilutions of a purified PCR product (by NGSBeads, Steinbrenner, Wiesenbach, Germany, following the manual) derived from wastewater. Calibration curves for 16S rRNA were obtained by serial dilutions of a linearized plasmid (pGEM-T easy, Invitrogen, Carlsbad, USA) carrying a single amplicon variant. Specificity of PCR reactions was checked by melt curve analysis, and potential false positives were removed. All samples were analyzed in technical duplicates to obtain final copy numbers per sample by averaging. The global correlation coefficient between the duplicates was R 2 = 0.92 (n = 53), showing a strictly linear relationship across the measured range of 4 orders of magnitude.

Intl1 gene cassette sequencing
Genomic DNA from three effluent water and two filter material samples of the devices D1 and D2 was used for characterization of class 1 integron gene cassette arrays (intl1). The cassette arrays of Tn402-associated class 1 integrons were amplified using the primers HS458 and HS459 (Holmes et al. 2003). These primers target the integron recombination site and the 3' end of the cassette array, which normally terminates in the qacE /sul1 gene fusion. Sequencing of HS458/459 PCR products can thus recover resistance determinants. The library collection was carried out with the following cycling program: 94 • C for 3 min; 94 • C for 30 s, 55 • C for 30 sec, 72 • C 1 min 30 s for 35 cycles and 72 • C for 5 min. Amplicons were pooled and sequencing was performed using MinION (Oxford Nanopores Technologies, Oxford, UK) using the LSK-109 library preparation according to the manufacturers recommendations and a Flongle flow cell generating 622 526 reads with the high accuracy basecalling mode (MinKnow version 19.10.1).
The six-frame translations were also scanned against Pfam (El-Gebali et al. 2019) using HMMER (using defined trusted thresholds, the '-cut tc' option). All annotations were added to a FARAO annotation database (Hammarén, Pal and Bengtsson-Palme 2016). Lists of annotated integron regions were then produced by querying the FARAO database with different criteria.

Statistical analysis
Statistical analysis of the microbial community composition was performed by converting the ASV table produced by DADA2 into phyloseq objects using the 'phyloseq' package (v.1.24.2) in R (v 3.6.0) (R Development Core Team R 2011, McMurdie and Holmes 2013). The microbial diversity indices were analyzed using the 'vegan' and 'betapart' package from CRAN (Baselga andOrme 2012, Oksanen et al. 2013). The Shannon index was used for the alpha diversity while ASV richness estimate was determined by rarefying the amplicon dataset to the smallest sample (3538 sequences). Kruskal-Wallis was used to test significant differences between experimental conditions. Differential abundance analysis of taxa to identify the removal/replacement of microbes before and after the SQIDs was performed by DESeq2 (v 1.29.5) (Love, Huber and Anders 2014). To gain insight about the overall microbial retention exerted by the SQIDs, we partitioned the β-diversity into two components: turnover (β-sim) and nestedness (β-ness) (Baselga and Orme 2012). Multivariate statistics were investigated with generalized linear models (GLMs) for multivariate abundance data using the mvabund package (Wang et al. 2012). Predictive models were fitted using 'negative.binomial' family, often being appropriate for count data, with the mean-variance function tending to be quadratic rather than linear. Non-metric multidimensional Scaling (NMDS) was used to visualize the microbial community composition and how it aligned with different variables (heavy metals, intl1, sample type). The intl1 copies were further normalized by the 16S rRNA copy numbers. All qPCR data (16S rRNA, intl1) were log-transformed prior to statistical analysis. We used the BioEnv approach (Clarke and Ainsworth 1993) to examine the best subset of environmental variables to correlate with community dissimilarities. In addition, to explore the correlation between microbial community's relative abundances, heavy metals and intl1 gene abundances, Spearman correlations were calculated. To test whether heavy metal content could predict the bacterial composition, we assessed the significance of the correlation using the 'adonis2' function in vegan (Anderson 2001) (v 3.6.0).

Data availability
The sequence data (Microbial community and intl1 amplicon data) is deposited at ENA (

Physico-chemical properties of road runoff, effluent of the SQIDs and filter media
As already described for this site by Helmreich et al. (Helmreich et al. 2010), the concentrations of Cr, Cu, Ni, Pb, and Zn as well as the pollution level (WPI GFS ) of the analyzed road runoff (influent, I) and effluent samples of the two SQIDs (ESA, EF) showed strong seasonal variation with higher values observed in spring ( Table 1). The higher EC in the spring samples indicate the influence of de-icing salt (sodium chloride) applied on-site, which contribute significantly to the toxicity of road runoff (Bartlett et al. 2012). As a consequence of the neutral to slightly alkaline pH of the samples, heavy metals were predominantly found in the particulate phase in the influent of the SQIDs. The dissolved Pb concentrations were below LOQ, as were half of the dissolved Cr and Ni concentrations. Consequently, it was only possible to determine the dissolved fractions of Cu and Zn, which were in median 18% and 21%. The overall pollution level, as indicated by the WPI GFS of the SQID effluents, was lowered in EF. In ESA, 18% of Cu and 38% of Zn were dissolved. In EF larger dissolved fractions were observed: 63% Cu and 40% Zn . The filter material sample FD1 contained higher Ni contents than FD2, but showed lower values for the other metals, respectively.

Microbial parameters of road runoff and SQID systems
The investigated SQIDs were colonized by a diverse range of microbial taxa. We removed five false positive samples from 92 sequenced samples. Eventually, about 7538 unique amplicon sequence variants (ASV) were detected for water samples (I, ESA, EF) and 5599 in filter material FD1 and FD2 ( Table 2). The 16S copies as an approximation for cell counts ranged in the order of 10 8 -10 9 copies per ml water. The copy numbers in the filter material was in the range of 10 7 copies per gram material. The class I integron gene cassette intl1 copy numbers had high numbers in both filter material and water samples. Neither strong seasonal effects nor differences between the filter materials in terms of 16S copies were detected. The SQIDs treatment had no strong effect on microbial parameters (Table 3).

Microbial taxa of road runoff
In both systems, the most prevalent phyla consisted of Proteobacteria-mainly composed by Gammaproteobacteria and Alphaproteobacteria, followed by Actinobacteriota, and Bacteroidota (Fig. 1A). The main difference between the two devices was an increased proportion of Campilobacteriota for D1 that had itself established in the intermediate ESA (7%). At the genus level many genera ranged below 2% relative abundance (Fig. 1B).
Most of the dominant genera like Massilia, Alkanindiges, Sphingomonas, Hymenobacter, Acidovorax and Arthrobacter that were found in the influent were still present in ESA. The ESA water samples showed a dominance of Pseudarcobacter (8%). In contrast to the water samples, the filter media of the SQIDs were clearly distinct (with minor vertical changes between the filter horizons; Figure S2). For both filter media, the most prevalent phyla consisted of Gammaproteobacteria and Alphaproteobacteria, followed by Actinobacteriota, Bacteroidota, Acidobacteriota, Chloroflexi, Desulfobacterota and Firmicutes (Fig. 1A). On the genus level, Hydrogenophaga and Rhodoferax (4.7% and 4.1%, respectively) were the dominant taxa in FD1 column, while Arenimonas (3.4%) and Sphingomonas (2.8%) dominated FD2 (Fig. 1B)

Retention of microbes by SQIDs
By partitioning the β-diversity into loss of species (nestedness) and species turnover, we confirmed that there was a turnover of taxa (as ASV) between influent and effluent of D1 (turnover = 0.81, nestedness = 0.04) and D2 (turnover = 0.90, nestedness = 0.02), pointing to a replacement of species along the water flow of SQIDs rather than species loss (overall nestedness = 0.03). Differential abundances of microbial genera pointed to few differentially enriched genera for D1 and D2 (Fig. S3, Supporting Information). In D1, few genera showed up, e.g., a high enrichment of C39 (Rhodocyclaceae; log2 fold change of 28.5) was observed. In D2, a greater removal was detected for 15 different genera with up to 21.2 log2 fold change. On the ASV level, we identified several potential microbial risk factors, i.e. taxa that are derived from animal host systems and may be relevant for human health and hygiene (e.g. Erysipelothrix, Shigella, Escherichia; Table S2, Supporting Information). The majority of these taxa were mostly found at very low relative abundances in the road runoff (<0.08%). Among the potential bacterial pathogens, the genus Pseudomonas was most abundant in all samples followed by Corynebacterium in the effluent ESA.

Factors that influence the microbial community composition
Multivariate statistics separates the two effluent water samples EF and ESA and also point to single metals, pH and intl1 as additional influencing factors ( Fig. 2; Table S3, Supporting Information). Moreover, seasonal changes and heavy metals (as sum of Cr, Cu, Ni, Pb and Zn molarity), impacted the species composition of the effluent samples (GLM: LRT = 12 395, LRT = 9350, P = 0.006 and P = 0.03, respectively).

The influence of heavy metals on microbial taxa
To gain more insight into the role of heavy metals, we preselected the most predictive metals using bioenv, indicating an influence of Ni, Zn and Cu on the microbial community  Figure S1).
(adonis2 for the total model: R 2 = 0.28, P < 0.001). Likewise, the 16S-normalized intl1 abundance showed linear relationships with heavy metal concentrations Ni, Pb, and Zn with higher explanatory power for Ni and Zn (R 2 > 0.71, P < 0.001; Fig. 3). This was further explored by co-correlating the most abundant 50 genera with heavy metals and intl1 (Fig. 4). A total of seventeen genera showed positive correlations with metal concentrations, with Aquabacterium, Hydrogenophaga and Trichococcus, associated with almost all measured metals. Three out of the five heavy metals (Ni, Pb and Zn) showed the highest positive association with the relative abundance of genera (R 2 > 0.50, P < 0.05). On the other hand, Aeromonas, Aquicella, Legionella and Pseudomonas showed significant negative correlations with heavy metals.
To further test the potential link of intl1 and heavy metal resistance, we sequenced parts of the genes that were carried by the class 1 integrons in the systems. In total, 296 of the 622526 reads from the integrons (0.05%) contained 98 different resistance genes. Of these, 82 were metal or biocide resistance genes (BacMet), 7 were clinically relevant antibiotic resistance genes (ResFinder) and 11 were antibiotic resistance genes previously only encountered in functional metagenomics studies (FARME). The most common antibiotic resistance genes were aminoglycoside resistance genes aadA5 (found on six integron sequences) and aadA4 (three sequences) and fluoroquinolone efflux pump oqxB (four sequences). Four other genes (aac(3)-Ia, aac(3)-Ib, msr(D) and vat(E)) were found only once. Most of the identified genes were involved in metal resistance, most commonly to heavy metals such as Pb, Cd and Zn (Fig. 5). Cu and Ag resistance genes constituted around 13% of the identified genes, while antibiotic resistance genes accounted for 8.6% of the identified genes in total. Biocide resistance genes made up approximately one-third of the identified resistance genes. The most commonly encountered resistance genes (> 6  occurrences) were the metal resistance genes czrA, czcA and silP, the biocide resistance gene qacE, and the efflux pumps mexW and mexI that also facilitate multidrug resistance (Table S4).

DISCUSSION
The results revealed that SQIDs not only retain heavy metals from road runoff, but also change the microbial community composition and influence the mobile genetic elements. The overall analysis of the road runoff and the SQID samples indicated a predominance of Gammaproteobacteria, Actinobacteriota and Bacteroidota in the water and the respective filters media. These findings are consistent with previous reports, where these phyla have been identified in stormwater runoff as anthropogenic or erosion signatures (Leung, Chen and Sharma 2005, Shanks et al. 2013, McLellan, Fisher and Newton 2015. Similarly, the taxa that occurred in the SQID filter materials, like Desulfobacterota, Chloroflexi and Acidobacteriota, were all previously observed in an infiltration basin collecting highway runoff (Rotaru et al. 2012). Acidobacteriota are mainly found in low pH environments (Lauber et al. 2009) tolerating various pollutants such as PCBs, petroleum compounds (Abed et al. 2002, Sánchez-Peinado M del et al. 2010) and heavy metals (Gremion, Chatzinotas and Harms 2003). Only few taxa with pathogenic potential were present at low levels. Although SQIDs are not designed to retain microorganism, the high microbial abundances in road runoffs warrant further investigation,  especially if water treatment selects for mobile genetic elements and receiving waters are considered a critical resource.

Influence of heavy metals on the microbial community
Heavy metals with high concentrations in waters or soils show toxic effects to almost all microbes by affecting metabolic functions such as protein synthesis (Kandeler et al. 2000, Tang et al. 2019, thus leading to variations in microbial biomass and diversity (Kaurin, Cernilogar and Lestan 2018). Several studies have shown how Cu, Zn, Pb and other heavy metals severely inhibited microbial biomass and could cause a reduction of microbial α-diversity (Kandeler, Kampichler and Horak 1996). The most common conclusion is that only high concentrations of heavy metals can significantly decrease bacterial biomass, whereas mid-low concentrations of heavy metals can increase microbial biomass and stimulate microbial growth (Flieβbach, Sarig andSteinberger 1994, Chander, Brookes andHarding 1995). Our pH and metal measurements indicated that large fractions of the heavy metals are not readily bioavailable, nevertheless particulate-bound heavy metals are considered partly bioavailable (Zhang, Hua and Krebs 2015). Anoxic conditions in particular may favor metal reduction as a source of energy, which generally leads to the release of metal ions into the water (Teiri et al. 2016). The prevalence of Desulfobacterota, which are responsible for sulfate reduction processes in stormwater retentions ponds (D'Aoust et al. 2018), together with Chloroflexi that constitute a substantial proportion of the activated sludge community in wastewater treatment plants (Zhang, Xu and Zhu 2017), point to anoxic processes that may occur in the SQID systems. This is of special interest since (Rommel, Stinshoff and Helmreich 2021) showed that sediments trapped in SQIDs as well as filter media can contain reducible heavy metal fractions. Consequently, anoxic conditions could result in reduced treatment efficiency of SQIDs.

Heavy metal resistances are linked to intl1
Ni and Zn, which showed an influence on microbial community composition, are known to induce different resistance mechanisms in bacterial metabolism (Schmidt and Schlegel 1989, Nies 1992, Alboghobeish, Tahmourespour and Doudi 2014. In this context, one interesting case was Arcobacter (and Pseudoarcobacter;Pérez-Cataluña et al. 2018), which was abundant in water and filter material. Arcobacter is known to form biofilms in various stainless steel, Cu and plastic pipe surfaces and colonize water distribution systems (Assanta et al. 2002, Cervenka et al. 2008). In our case, Pseudoarcobacter strongly correlated with Ni and Zn and also intl1 gene abundance, thus suggesting the selection for bacterial resistance in Pseudoarcobacter. Both, metal and antibiotic resistances are commonly carried on mobile genetic elements. Integrons, in particular, have been recognized as marker for anthropogenic pollution (Gillings et al. 2015). Prior research from different heavy metal polluted scenarios showed the spread of resistances by horizontal gene transfer (Zafar et al. 2007, Rehman andAnjum 2011), and there have been signs of co-selection of several resistant genes linked to clinically relevant antibiotic resistances (Pal et al. 2015). For example, resistance to As, Mn, Co, Cu, Ag, Zn, Ciprofloxacin, βlactams, chloramphenicol and tetracycline can be achieved by reduction in membrane permeability (Silver andPhung 1996, Ruiz et al. 2003). Similarly, Cu, Co, Zn, Cd, tetracycline, chloramphenicol and β-lactams resistance can be achieved through rapid efflux of metals or antibiotics (Levy 2002, Nies 2003. Therefore, heavy metals have the potential to represent an extended selection pressure for the development of antibiotic resistance in microorganisms (Stepanauskas et al. 2005), and the transfer of these resistantances in the environment may pose potential risks to human health (Bengtsson-Palme, Kristiansson and Larsson 2018).

Stormwater quality improvement devices as hot spots for horizontal gene transfer
The intl1 gene cassette analysis highlighted the presence of heavy metal resistance in these microbial communities. In addition, the abundance of intl1 in the filter media compared to the influent suggests a strong selection pressure associated with a significant rate of horizontal gene transfer in the systems. The amount of bacteria carrying class 1 integrons is consistent with data reported in polluted water systems like WWTPs (Paiva et al. 2015). However, while previous studies have generally described high removal rates of bacteria carrying class 1 integrons (55%) after treatment process (Paiva et al. 2015), few studies have reported an increase in the abundance of the intI1 gene as in our study during the wastewater treatment (LaPara et al. 2011, Stalder et al. 2014. The variation of results observed among studies may be attributed to several factors such as selected resistant bacterial taxa, the climate and population conditions, occurrence of rain events before sampling as well as organic loading, pH and temperature. Horizontal gene transfer plays an important role in the evolution, diversity and recombination of multidrug resistant strains (Nakamura et al. 2004, Thomas andNielsen 2005). The class 1 integron has been associated with the presence of metal resistance genes (MRGs) and antibiotic resistance genes (ARGs) (Stokes andHall 1989, Li, Xia and. Our data suggests that SQIDs could be a high-risk environment for resistance development, similar to other hotspots, like manure, sewage and municipal solid waste (He et al. 2014, Su et al. 2015, Wang et al. 2019. Furthermore, the presence of different resistance genes (e.g. czrA, czcA and silP), including the high proportion of multidrug resistance genes (e.g. mexW, (Nikaido 2009, Chakraborty 2016) as well as the strong positive correlations to heavy metals, suggest that integrons contribute to the spread of MRGs and ARGs within road runoff drainage systems. While no typical antibiotic treatment related resistance genes (sul1, ampC, etc.) were identified, they may be present on other mobile genetic elements not targeted in this study.

Limitations and future directions
This study provides a first deeper description of road runoff microbial communities and contributes to our understanding of their potential environmental impact on the receiving water bodies. As a pioneer study, our study design was limited to two SQIDs and we could only monitor three seasons. Thus, we could not investigate if there are further effects by e.g. higher amounts of de-icing salts in winter, which potentially enhance mobility and bioavailability of the present heavy metals (Acosta et al. 2011, Bartlett et al. 2012, Schuler and Relyea 2018. Furthermore, we did not consider other systems such as infiltration basins or sand filters, and it remains an open question if our results are transferable to other road runoff drainage systems. However, it is obvious that runoff from a highly trafficked urban road carries a high microbial load with dominant signs of anthropogenic pollution. This comes with a relatively high risk related to the cycling of resistance genes and thus microbial risk mitigation practices should be considered in the future. Recently, it has become clear that microbes are critically linked to our changing environment, and that they have to be included in future policies (Cavicchioli et al. 2019). Future studies are therefore encouraged to assess the risks of discharge of microbes and their resistance genes from SQIDs and other SCMs into receiving environments.
Academy at the University of Gothenburg, the Swedish Research Council (VR; grant 2019-00299) under the frame of JPI AMR (EMBARK; JPIAMR2019-109), and the Centre for Antibiotic Resistance Research at the University of Gothenburg (CARe). CW was financially supported by TUM Junior Fellowship and the DFG project (WU 890/2-1). We would like to thank Dr Klaus Neuhaus of the ZIEL Core Facility Microbiome, Technical University of Munich, Freising, Germany, for sequencing.

Conflicts of Interests.
MSc Liguori, Dr Wurzbacher, and Dr Bengtsson-Palme declare no competing interests. MSc Rommel and Dr Helmreich informed the Bavarian Environment Agency, as well as the SQID companies FRÄNKISCHE Rohrwerke and Hauraton on the content of this article prior to submission.