Outflow risks of antibiotic-resistant bacteria in stormwater bioretention cells: understanding roles of adsorption and transmission

In recent years, antibiotic resistance caused by the abuse of antibiotics has been a critical public health challenge. Stormwater runoff, as an important carrier of non-point source pollution, has been proven to be one of the important paths for antibiotic resistance diffusion in environments (Almakki et al. 2019). Different from traditional stormwater contaminants (such as nutrients, heavy metals, or common pathogens), antibiotic-resistant bacteria (ARB) can carry antibiotic resistance, which have transmission risks due to the replicability of biological characteristics (Pachepsky et al. 2008). To mitigate this potential public health risk posed by antibiotic resistance diffusion, specific studies are needed to evaluate whether antibiotic resistance would present in outflow of the existing best management practices (BMPs).

As one of the common stormwater BMPs, bioretention cells have been widely used for the treatment of pollutants in stormwater runoff, including organic and inorganic chemicals (Sun et al. 2020; Kong et al. 2021; Biswal et al. 2022) and pathogens (Sharma & Malaviya 2021). The performance of bioretention cells was restricted to key design and operational parameters (Zhang et al. 2021). The literature has made some successful attempts to enhance pathogen removal through system design optimization, including media (e.g., iron or copper oxides) (Zhang et al. 2010; Li et al. 2016; Chandrasena et al. 2017; Xu et al. 2019), vegetations (Kim et al. 2012), and submerged zones (Rippy 2015). For operational parameters, Garbrecht et al. (2009) found that strategies to maximize infiltration in bioretention cells could be counterproductive to bacteria removal. Similarly, Bright et al. (2010) also found that a decrease in seepage rate was correlated with a decrease in effluent bacteria concentration in the bacteria-spiked stormwater sand columns. The aforementioned previous studies focused on Escherichia coli, enterococci, and coliphage, which do have transmission properties of antibiotic resistance. However, no studies have been found relating to the ARB outflows in stormwater bioretention cells.

Rusciano & Obropta (2007) reported that both adsorption and filtration were two main mechanisms for the reduction of faecal coliforms in bioretention cells. Sand and soil were two common substrate materials in bioretention cells (Palmer et al. 2013; Shrestha et al. 2018; Zuo et al. 2019). Zacharias et al. (2020) reported that 1–2 log unit reduction of ARB through retention soil filters could be detected in addition to the reduction within the municipal wastewater treatment plant. These studies indicated that ARB could also be reduced by bioretention cells through the adsorption, but there were no data on ARB reduction by the adsorption in bioretention cells until now. Thus, it was necessary to investigate the adsorption properties of the common substrate materials to ARB, to better understand the reduction of ARB in bioretention cells. Meanwhile, pathogens could not be killed through the adsorption of materials, which was not in line with the bactericidal mechanisms such as chlorine, UV, and advanced oxidation technologies (e.g., photocatalytic disinfection). Unlike pathogens without antibiotic resistance, the transmission of antibiotic resistance genes (ARGs) carried by ARB could frequently occur through the conjugation and/or transformation in environments (Dang et al. 2017; Wang et al. 2020), which implied that a potential risk of ARB outflow might exist in effluents from bioretention cells, where many dominant bacteria (like Ramlibacter and Nitrosomonadaceaea) have been found (Zuo et al. 2020).

Currently, reports on ARB outflow were given preference to conventional and advanced wastewater treatment processes, such as membrane bioreactor systems (Le et al. 2018), bio-electrochemical systems (Guo et al. 2017), and constructed wetlands (Huang et al. 2017), where concentrations of ARB in effluents declined relative to the raw influent at different degrees with the change of processes. Meanwhile, Zhang et al. (2018) found that functional bacteria, which were potential antibiotic resistance species were the main reason for the performance stabilization of up-flow anaerobic sludge blanket bioreactors. There was still a risk of releasing relatively more ARB in proportion to total bacteria into environments through the effluent during the wastewater treatment process (Huang et al. 2017). However, ARB outflow characteristics were still unclear, which impeded further assessment of the ecological risks caused by antibiotic resistance in the treated stormwater. On the other hand, previous studies have reported outflow modelling of the conventional chemical pollutants in bioretention cells (Li & Davis 2016; Fan et al. 2019). This indicated that a suitable model could also be absolutely necessary for the prediction of ARB outflow in stormwater bioretention cells. One recent study (Reddy et al. 2020) reported that the zero-order kinetic model best described the removal rate of Cu and Ni by sand, and the first-order kinetic model was only applicable for nitrate removal by iron filings, while the second-order kinetic model described the removal rates of other contaminants and filter media combinations. Furthermore, the first-order kinetics with plug-flow was often used to simulate the outflow concentrations of traditional pollutants in vertical flow wetland systems (very similar to stormwater bioretention system) (Saeed & Sun 2011; Weerakoon et al. 2020). However, it is not sure whether the mentioned model may be applicable to simulate ARB outflow in bioretention cells.

Therefore, this study aims to understand the outflow modelling of ARB in stormwater bioretention cells with the following four specific objectives:

• (1)

  To investigate ARB outflow profiles (at three different depths) by the detection of ARB levels in stormwater bioretention outflow over a 150-day operation time.

• (2)

  To study adsorption roles for both substrate and biofilms through the analysis of adsorption properties.

• (3)

  To explore the effect of the antibiotic resistance transmission process (conjugation transfer and transformation).

• (4)

  To fit the conventional first-order decay model (e.g., plug-flow) and polynomial equations to understand ARB outflow risks.

To our knowledge, this is the first study presenting ARB outflow risks in stormwater bioretention cells. The outcomes of this work provide promising data to design more efficient bioretention cells for the interception of antibiotic resistance diffusion risks through stormwater runoff.

E. coli K-12 with plasmid RP4 that carries ARGs including bla_TEM (Beta Lactamase resistance genes), tetR (Tetracycline resistance genes), and aphA (Aminoglycoside resistance genes) was selected as target ARB in this study. The strains were incubated in Luria broth medium (LB; pH of 7.4), shaken overnight (140 rpm) at 30 °C, and then centrifuged at 5,000 × g for 5 min. The supernatants were removed after the centrifugation, and the pellets were washed with phosphate-buffered saline (PBS, 1 × , pH of 7.2) and re-suspended in PBS. The main chemicals were obtained from Nanjing Jiaodeng Science Equipment Co. Ltd (China), including amylon (purity > 99%), ammonium chloride (NH₄Cl, purity > 99%), and potassium dihydrogen phosphate (purity 99%). The stock solutions of chemicals were prepared by dissolving the corresponding compound into sterile deionized water (autoclaving for 15 min at 121 °C) to yield the desired concentrations for chemical oxygen demand (30.0 mg/L), ammonia nitrogen (⁠⁠, 2.0 mg/L), and total phosphorus (TP, 1.0 mg/L) (the simulated stormwater quality).

Batch adsorption experiments were conducted as a function for both sorbent dosage and contact time using 250 mL Erlenmeyer flasks (after the sterilization) as reaction vessels. A known quantity of biofilms (or substrate depending on the experiment; collected right after the biofilm culture stage) was added into 50 mL solutions (the simulated stormwater) consisting of 10⁵ CFU/mL target ARB. The reaction vessels were placed on a rotary shaker at 200 rpm. All experiments were conducted at 25 °C. The water samples were taken at different time intervals (0, 30, 60, 90, 180, 240, 300, 360, and 480 min). These adsorption experiments were done in triplicate.

Transmission experiments were done to investigate the ARB conjugation transfer process and transformation process. Both target ARB (the donor of the anti-bacterial resistance gene) and E. coli HB101 carrying streptomycin sulphate resistance (the recipient) were used to establish the conjugation transfer system. Both the conjugation transfer system and experimental conditions were set based on the previous literature (Lu et al. 2018). The donor and recipient bacteria were placed in liquid LB and cultured overnight at 37 °C, followed by centrifugation for 5 min at 5,000 rpm. After the centrifugation, the supernatants were removed, and the residuals were washed with PBS (pH = 7.2) and re-suspended in calculated volumes of PBS to obtain 1 × 10⁸ CFU/mL bacterial concentrations. Then, the donor and recipient bacteria were mixed at a ratio of 1:1 and treated in solutions containing biofilms from different depths for 6 h. The solution sample without biofilms was used as the control.

A transformation system was established through the free plasmid (pWH1266) and the naturally competent bacterium Acinetobacter baylyi (A. baylyi ADP1). Both the transformation system and experimental conditions were in line with the one reported by Wang et al. (2020). Firstly, pWH1266 plasmid was prepared and suspended in elution buffer (10 mM Tris-HCl) and stored at −80 °C (concentration 210 ng/μL). One millilitre of A. baylyi ADP1 culture was cultured in 100 mL LB broth overnight at 30 °C. Then, the 500 μL overnight culture was inoculated into 50 mL liquid LB (1% mL v/v) in a 100 mL glass bottle. When the glass bottle was incubated in a constant temperature shaking incubator at 30 °C for 6 h, the culture reached the early stationary growth stage, and the OD600 was 1.1 (the bacterial concentration was 3 × 10⁸ CFU/mL). The bacteria were then collected by centrifugation, then precipitated, washed, and re-suspended using the PBS solution. The plasmid was added to A. baylyi ADP1 culture at the final concentration of 0.8 ng/μL (calculated to 8.34 × 10⁷ copies/μL), and then sodium acetate (100 mg/L) was added and the mixture used to treat solutions containing biofilms from different depths for 6 h. All transmission experiments were processed in triplicate.

• (a)

  Sample analysis

Based on our previous studies (Zuo et al. 2015), ARB quantity was determined by LB agar plates containing 50 mg/L of kanamycin sulphate, 10 mg/L of tetracycline hydrochloride, and 100 mg/L of ampicillin. Each sample was analysed in triplicate, and the average was used as the analysis data. Transconjugants (recipients that received the RP4 plasmid) were counted by using selective LB plates (containing ampicillin 100 mg/L, streptomycin 30 mg/L, kanamycin 50 mg/L, and tetracycline 10 mg/L), based on the literature (Rico et al. 2017; Jin et al. 2018). The recipient cells were determined by LB agar plates containing 30 mg/L streptomycins. The conjugative frequency was calculated by normalizing the transconjugant numbers to the recipient numbers. The transformation frequency for each transformation system was calculated through the number of transformants divided by the total number of bacteria, where the total number of bacteria was estimated by spreading the transformation systems onto LB agar without antibiotics. ORP was detected through a soil redox potential detector (SU-ORP, Beijing Allied Weiye Technology Co., Ltd, China) at 0, 3, 6, 12, 24, 48, 72, 96, and 120 h after the end of each experimental operation, and the average of ORP (31 operation times, 3 replication cells) was used for analysis in this study.

• (b)

  Analysis of the adsorption process

Adsorption kinetics (including the pseudo-first-order, pseudo-second-order) and isotherms (Langmuir and Freundlich isotherms) were used to explore the pathway and mechanism of adsorption reactions (Zuo & Balasubramanian 2013).

• (c)

  Outflow modelling

Based on results of the first order model produced in wetland systems (Saeed & Sun 2011; Weerakoon et al. 2020), and stormwater bioretention cells (Reddy et al. 2020), the first-order kinetics with plug-flow (Weerakoon et al. 2020) were chosen to model ARB outflow in this study.

In addition, the previous literature reported that the survival of bacteria in foods could be well fitted using primary growth equations (e.g., Baranyi model and Gompertz model) and polynomial equations (Tomac et al. 2013; Lee et al. 2020), where primary growth equations were applied to describe the growth of vertical inheritance of bacteria. However, the unavoidable was the ARB growth resulting from antibiotic resistance transmission, which suggested that the primary growth model could not be used to evaluate changes of ARB in bioretention cells. Thus, we used the first-order decay model and polynomial equation to simulate the outflow ARB concentrations using the inflow data.

ORIGIN (Version 8.0) was applied to conduct the data analysis graphing and model fitting, and coefficient of determination (R²), relative root-mean-square error (RRMSE), and Nash–Sutcliffe efficiency (NSE) were used in evaluating fitting performance, based on the previous studies (Saeed & Sun 2011; Zhang et al. 2021).

The previous literature reported that E. coli detachment from sand was nearly 100% of attached cells after one washing, whereas a total of less than 15% of cells were detached from soil after three washings (Mankin et al. 2007). Similarly, Chandrasena et al. (2016) claimed that. E. coli could be detected in the effluent of the first operation for stormwater bioretention cells, while Meng et al. (2018) found that Campylobacter spp. concentrations were often higher in outflow than inflow for stormwater wetlands. It could be due to that the conventional bioretention soil media, which typically consists of sand, sandy loam, loamy sand, or topsoil amended with compost, has limited capacity to remove and may leach some stormwater pollutants (Tirpak et al. 2021). However, in this study, ARB was not found in effluent before the 20th operation. This proved that ARB outflow profiles could be a different case in bioretention cells.

The pseudo-second-order equation was better in describing the adsorption kinetics of ARB by substrates and their surface biofilms (Table S1 of the Supplementary Information), with the high correlation coefficients (R² > 0.990) and low error (<0.002), compared with the pseudo-first-order equation. Meanwhile, C_e values of the pseudo-second-order equation were in good agreement with the experimental ones. This suggested that the adsorption process of ARB by substrate and their surface biofilms belonged to chemical adsorption. The rate constant of pseudo-second-order adsorption (k₂) was the largest for biofilms from the middle layer, and higher an order of magnitude than one of the substrates. On the other hand, the adsorption data were fitted by both Langmuir and Freundlich equations, and correlation coefficients are also provided (Table S2 of the Supplementary Information). The maximum amount of ARB adsorbed (Q_max) reached 10⁸ CFU/g of substrate and 10⁷ CFU/g of biofilms, respectively. Furthermore, along the depth direction, the maximum adsorption of ARB by biofilms showed a trend of first decrease and then increase. Biofilms from the bottom layer had the best adsorption capacity for ARB with Q_max of 10⁷ CFU/g.

The pseudo-first-order equation was commonly applied to fit the adsorption characteristics of bacteria onto sands. For example, the pseudo-first-order kinetic adsorption rate coefficient was an exponential function of the total interaction free energy between the bacteria and sediment from the Canadian River alluvium (Norman, OK) with 5.10 1/h (E. coli HB 101), 4.52 1/h (E. coli JM 109), 2.12 1/h (Pseudomonas fluorescens), 2.04 1/h (Pseudomonas putida), and 1.34 1/h (Pseudomonas sp.) (Chen et al. 2003). But the pseudo-second-order equation was dominant for the adsorption kinetics of ARB by substrate and biofilms, as an indication of a chemisorption mechanism because substrate in bioretention cells was mainly composed of sand, silt, and clay, where hydrogen bonding was the important factor in bacteria–clay mineral adsorption (Rong et al. 2008; Wu et al. 2012). This might result in the less easy desorption of ARB than conventional bacteria (like E. coli) from both substrate and biofilms. Both Langmuir and Freundlich equations fitted well to the ARB adsorption isotherms, implying that there were both monolayer and multilayer adsorptions present in the adsorption process of ARB by substrate and biofilms. The adsorption level (10⁸ pfu/g) of ARB by substrate (8:2 ratio of sand to soils) in this study agreed with the one (6.93 × 10⁸ cells/g) of bacteria by Fe (III)-coated sand (Mills et al. 1994) and the one (about 10⁹ pfu/g) of viruses (MS2 and X174) by the sand with iron oxide (Pecson et al. 2012). This suggested that adsorption of substrate and biofilms could be effective in preventing the outflow of ARB. Moreover, extracellular polymers (proteins, polysaccharides, and nucleic acids) might be responsible for the low adsorption levels of ARB by biofilms, which could be explained by the fact that organic matter played a suppressive role in the interfacial processes of the initial bacterial attachment to soil particles (Wu et al. 2012). Selecting substrates with good adsorption properties would be important to remove ARB, particularly the ones conducive to microbial attachment but not conducive to microbial growth, which needs further study based on new functional materials.

It is assumed that the role of substrate and biofilm adsorption is fully reflected in the bioretention cells. When estimated using the Q_max (Table S2), theoretically, the maximum number of ARB adsorbed were 1.23 × 10¹² CFU (the upper layer), 3.8 × 10¹² CFU (middle and its above layers), and 5.10 × 10¹² CFU (all of layers), these are about 400, 1,200, and 1,600 times of the total ARB number in a single influent (10⁹ CFU). This indicated that ARB should be detected in the upper effluent after about 400 operation times (more than 5 years), in the middle effluent after about 1,200 operation times, and in the bottom effluent after running about 1,600 times. However, the fact is that the actual ARB outflow characteristics (Figure 2) were not consistent with the theoretically estimated results, and the time of ARB detected was much earlier (such as around the 20th operation for the upper effluent) than the theoretical values. This should be attributed to antibiotic resistance transmission in bioretention cells.

The outflow order of ARB was not in line with frequencies of antibiotic resistance transmission (conjugation transfer and transformation) in substrate layers in this study. Transmission frequency was largest in the middle layer, which implied that the time of detecting ARB from the middle outlets should be the earliest, if only considering the perspective of antibiotic resistance transmission. However, the fact was that the detected ARB was found at first in the upper layers with lowest transmission frequencies. Adsorption experiments showed that ARB was firstly trapped and adsorbed in the upper layer, where there was the largest quantity of ARB (Figure S1 of the Supplementary Information). Moreover, Mohanty & Boehm (2014) found that the mobilization of deposited bacteria during interval infiltration was attributed to the exhaustion of attachment sites on biochar by the dissolved organic carbon leached from compost. This implied that the desorption resulting in microflora endogenous metabolism during interval periods could be another reasonable interpretation of the earliest detection of ARB in the upper outlets.

The previous study found that membrane permeability, gene expression, and reactive oxygen species (ROS) response were the dominant factors influencing the transmission of antibiotic resistance (Yu et al. 2021). Lu & Guo (2021) demonstrated that the stimulation of environmental conditions resulted in antibiotic resistance donors and recipient bacteria to produce intracellular ROS response, and then led to the enhancement of membrane permeability by exercising oxidative stress pressure, further increasing the risk of antibiotic resistance transmission. Changes in top-down environmental conditions (like ORP) of substrate layers should be responsible for the varying frequencies for both conjugation and transformation in different layers. A previous study found that removal in the bioretention cell with Lythrum salicaria L. was the highest (88.1%), which was consistent with ORP in the bioretention cells (Zuo et al. 2020), where ORP in different layers was not detected during the interval periods. High ORP (greater than 100 mV) represented an aerobic environment, while low ORP (less than −100 mV) reflected an anaerobic environment (Sun et al. 2012). The distribution of ORP at interval periods (Figure S2 of the Supplementary Information) indicated that both the upper and bottom layers were the aerobic environments, and the middle layer was hypoxia environments, which was in line with the maximum redox gradient between the surface and the bottom of the bed for continuous planted wetlands (407.7 ± 73.8 mV) (Corbella et al. 2014). Meanwhile, the rapid increase of ORP in the upper layer after 48 h of operation might be due to the decrease of substrate moisture (Zinger et al. 2021), implying that there would be more air in substrate layers. Jiang et al. (2007) found that air–water interface adsorption played an important role in E. coli transport in sand columns. In this study, the aerobic environments in the upper layer could contribute to the formation of an air–water interface, which might also result in the earliest outflow of ARB in upper outlets.

The highest conjugation frequency was found in the middle substrate layer (Figure 3), suggesting that the reductive conditions were more favourable for conjugation transmission of antibiotic resistance. In addition, frequencies for both conjugation and transformation in bioretention cells were in line with those in other environmental systems (e.g., activated sludge, soils, and ocean) (Zarei-Baygi & Smith 2021), which should be related to the similar dominant microflora in these systems. It is well known that the higher the frequency of conjugation transfer, the greater the risk of antibiotic resistance, which indicates that the middle substrate layer with the highest conjugation frequencies could be worth attention for the interception of antibiotic resistance transmission in bioretention cells through controlling aerobic conditions.

Although the third-order polynomial equation can describe outflow ARB acceptably in this study, it would be still essential to find out the desorption kinetics of ARB in substrate and biofilm, primary growth equations modified by considering antibiotic resistance horizontal transmission in the next work, and then to obtain outflow dynamic parameters of ARB in bioretention cells for the better blocking of antibiotic resistance risks. The more long-term operation will be more helpful in controlling risks of ARB in stormwater bioretention outflows.

This study uncovered the underlying ARB outflow processes to understand their outflow risks from stormwater bioretention cells. ARB in bottom effluents was detected after more than four months of operation, reflecting there was still a risk of antibiotic resistance in the effluents after the treatment of bioretention cells, although adsorption of the substrate and its surface biofilms were effective (Q_max reaching 10⁸ CFU/g of the substrate and 10⁷ CFU/g of biofilms) in preventing ARB outflow. ARB outflow was found at first in the upper layers (from the beginning of the 20th operation (i.e., 100 days)) with the lowest transmission frequencies of antibiotic resistance, instead of in the middle layer (from the beginning of the 21st operation) with the largest transmission frequencies, indicating that antibiotic resistance transmission was not the only decisive factor for ARB outflow risks. ARB outflow modelling for both upper (substrate layer thickness 150 mm) and middle (substrate layer thickness 450 mm) outlets could be well described by third-order polynomial equations with correlation coefficients 0.9067 (p = 0.0002) and 0.9780 (p < 0.0001), respectively, where there were both positive and negative relationships between outflow ARB and inflow ARB, reflecting the combined action of mechanisms blocking ARB outflow (e.g., substrate adsorption) and promoting ARB outflow (like transmission). Thus, the development of effective adsorption materials in the upper layers and the interception of antibiotic resistance transmission in the middle layers would be two potential approaches for the improvement of ARGs in stormwater bioretention cells.

This work was supported by grants from ‘National Natural Science Foundation of China’ (52170099) and ‘Jiangsu Provincial Department of Science and Technology’ (BK20220012).

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.