Estimated discharge of antibiotic-resistant bacteria from combined sewer overflows of urban sewage system

Yearly discharge of antibiotic-resistant bacteria (ARB) from combined sewer overflow (CSO) was estimated. The volume of CSO was estimated from operating data of the pumping station. In the target sewer catchment, 23% of the total of the volume of combined sewage was discharged untreated as CSO. Combined sewage contained 3-log larger E. coli than secondary treatment effluent although the abundance of antibiotic-resistant E. coli was not significantly different. In the target-combined sewer catchment, a yearly total of 4.8 × 1016 CFU of E. coli was discharged from 6.1 × 106 m3 of CSO, while 1.3 × 1012 CFU of E. coli from 2.1 × 107 m3 of effluent from the wastewater treatment plant (WWTP). This E. coli discharge was equivalent to 7.9 × 109 CFU/m3 from CSO, and 6.2 × 104 CFU/m3 from WWTP effluent. Consequently, a yearly total discharge of antibiotic-resistant E. coli from CSO was 3.7-log larger than the WWTP effluent. The small-flow CSO events, which had hourly flow rate smaller than five times of the average dry-weather flow, accounted for 43% of the total CSO volume, but 79% of the total discharge of antibiotic-resistant E. coli due to a small dilution factor with stormwater and frequent discharge. Reduction of small-flow CSO events would be important for effective reduction of ARB discharge from CSO.


INTRODUCTION
Increased emergence of antibiotic-resistant bacteria (ARB) is now a great concern for future human health. Water and soil environments are now recognized as reservoirs and sources of ARB of clinical concern 1 . To prevent prevalence of ARB through environmental pathways, it is important to control the release of ARB from anthropogenic activity into the environment 2 . Municipal wastewater is a major source of ARB in urban water environment in both developed and developing countries [3][4][5][6][7][8] . ARB in wastewater still remains at high abundance in secondary treatment effluents of WWTPs, and becomes the major source in water environment and drinking water resources [4][5][6][9][10][11][12][13][14] .
Combined sewer overflow (CSO) is also a possible significant pathway of ARB discharge from municipal wastewater. In a combined sewer system, wastewater and stormwater are collected in the same pipe. When the sewage volume in a combined sewer system can exceed the capacity of WWTP due to heavy rainfall or snowmelt, the combined sewage, which contains both wastewater and stormwater, is designed to overflow and discharge directly to nearby water bodies without treatment. These overflows, called combined sewer overflow (CSO), are a significant source of fecal pollution in water bodies in many urban areas [15][16][17][18][19] . According to USEPA 20 , discharge of CSO was estimated to be 20% of the total annual discharge of combined sewage in the United States. Moreover, climate change is predicted to substantially increase the volume and frequency of CSO in some areas due to a change in the precipitation pattern 21,22 . However, there are a limited number of studies on ARB discharge from CSO. Garner et al. 23 recently quantified antibiotic-resistant genes (ARG) at an urban stormwater outfall, and reported that a single CSO event could bring 1-to 3-log larger loadings of ARG than the dry-weather event. However, even though a single CSO event has a large discharge of ARB, the CSO events occur only occasionally. Moreover, the volume of CSO in each event has a large fluctuation, depending on the quantity of stormwater received by the combined sewer system. Concentrations of ARB in CSO also change, depending on the quantity of stormwater, because ARB in wastewater is diluted with stormwater in the combined sewer pipes. On the other hand, the WWTP effluent is discharged continuously throughout a year, although it may contain less ARB than CSO. For effective reduction of ARB discharge, it is important to know which of the CSO and WWTP effluents has a larger impact on ARB discharge into the water environment and drinking water sources.
The objective of this study is to estimate ARB discharge from CSO and wastewater treatment effluents of an urban-combined sewer system, in order to investigate the quantitative impact of CSO on ARB discharge into water environment. A combined sewer catchment, where the entire quantity of CSO and combined sewage was monitored at pumping stations, was chosen as the target area in order to obtain an accurate volume of the entire combined sewage and CSO from operating data of the pumping stations. Antibiotic resistance of E. coli, as a representative of fecaloriginated ARB, in the combined sewage and WWTP effluent, was monitored throughout a year. From these data, yearly discharge of antibiotic-resistant E. coli from CSO and WWTP effluents was estimated and compared with discussing their impact on ARB discharge from combined sewer systems to water environment.

RESULTS
Antibiotic-resistant bacteria in combined sewage and secondary treatment effluents The abundance of antibiotic-resistant E. coli in combined sewage and secondary treatment effluents had no significant change and seasonal trend ( Table 1). The annual average of the abundance of E. coli resistant to at least one antibiotic slightly increased from 37% in combined sewage to 42% in secondary treatment effluents; however, there was no statistically significant difference (p = 0.45). Many studies reported that ARB abundance showed no significant change or a slight increase through the treatment 9,10,12,13 . However, a change in the abundance of antibiotic resistance between combined sewage and secondary treatment effluents was highly fluctuated, and had no consistent trend in monthly data. The abundance of multiple antibiotic resistance in combined sewage and secondary treatment effluents was also highly fluctuated on a monthly basis, but not significantly different as the annual average (p = 0.67). The abundance values of E. coli with multiple antibiotic resistance were 26 and 23% in combined sewage and secondary treatment effluents as the annual average, respectively.
Although the abundance of antibiotic-resistant E. coli in combined sewage was comparable with secondary treatment effluents, combined sewage contained significantly higher concentration of E. coli than secondary treatment effluents (p < 0.001) ( Table 1). E. coli concentrations reportedly ranged 10 4 -10 5 CFU/mL in sewage 12,24,25 , and 10 1 -10 3 CFU/mL in secondary treatment effluents 12,24,26 . E. coli concentrations in this study were 3.2 × 10 4 -1.0 × 10 5 CFU/mL in combined sewage, and 2.8 × 10 1 -1.7 × 10 3 CFU/mL in secondary treatment effluents, which were within the typical range of those reported in other WWTPs 12,24,26 . At the target WWTP, secondary treatment effluent was discharged after chlorination. The log-removal value of E. coli by chlorination at the target WWTP was 3.8-log ( Supplementary Fig. S2). Therefore, combined sewage was estimated to contain 6.3-log larger E. coli than WWTP effluent at the target WWTP. Meanwhile, E. coli concentrations in combined sewage did not have significant seasonal change throughout a year. Standard deviation of E. coli concentrations in combined sewage was 0.16-log. E. coli in secondary treatment effluent had a slightly larger seasonal fluctuation than combined sewage, with 0.54-log of standard deviation. The observed fluctuations in this study had no consistent seasonal trend against temperature or precipitation. It is probably because they were originated from daily fluctuation rather than seasonal conditions. In this study, E. coli sampling was conducted once a month. No data were taken for daily and hourly fluctuation of E. coli concentrations in combined sewage and secondary treatment effluent. According to the past studies 12,24,26 , fluctuations of E. coli concentrations are reported to be small within 0.5-log of standard deviation in wastewater, 2-log of standard deviation in secondary treatment effluent. These were consistent with the observed fluctuations in this study. Therefore, E. coli concentrations in combined sewage and secondary treatment effluent in this study were within the typical range and had seasonal fluctuations reported at other WWTPs.
Quantity of combined sewer overflow In the target sewer catchment, 23% of total volume of combined sewage was discharged untreated as CSO. Although there are a limited number of studies that estimated CSO volume, it was comparable to estimation in the United States by USEPA (2004), which reported that 20% of combined sewage was discharged as CSO 20 . The volume of CSO and the number of CSO events fluctuated, depending on seasonal change in precipitation conditions. In the target catchment, the volume of CSO was large in August and December because of large precipitation (Fig. 1). Daily CSO volume was highly correlated with daily precipitation (Fig. 2). This indicates that the CSO volume is highly affected by precipitation in this combined sewer catchment. The frequency of CSO events and the flow rate of CSO in each event varies a lot, depending on not only precipitation but also rainfall intensity and duration. In the target catchment, the total volume of CSO in Table 1.  December were mostly small-flow and long overflows, while those in August were large-flow and short overflows. This seasonal difference in CSO flow rate has a large impact on ARB discharge to the receiving water body. ARB in combined sewage is mainly originated from municipal wastewater, which was constantly discharged independent of weather conditions. Meanwhile, the flow rate of stormwater was highly fluctuated by precipitation. Combined sewage contained a higher concentration of ARB when it received a smaller quantity of stormwater, and vice versa. Therefore, in this catchment, CSOs in December probably had a larger impact on ARB discharge than other months.
Quantity of antibiotic-resistant bacteria discharged from CSO and WWTP effluents Yearly quantity of antibiotic-resistant E. coli discharged from CSO and WWTP effluents was estimated from the abundance of antibiotic resistance, E. coli concentrations, and the flow rate of combined sewage and secondary treatment effluents. Yearly quantity of antibiotic-resistant E. coli discharged from CSO was much larger than the WWTP effluent (Fig. 3). Yearly quantity of E. coli with multiple antibiotic resistance also discharged more from CSO than WWTP effluents. The annual average of E. coli discharge  was equivalent to 7.9 × 10 9 CFU/m 3 from CSO, and 6.2 × 10 4 CFU/ m 3 from WWTP effluents. Yearly volume of CSO corresponded only to 29% of secondary treatment effluent. However, yearly quantity of E. coli discharged from CSO was 4.6-log larger than that from WWTP effluent. The main reason was because CSO contained much higher concentrations of E. coli than the WWTP effluent, while the abundance of antibiotic-resistant E. coli was not significantly different between combined sewage and WWTP effluent. As discussed above, CSO events with smaller flow rates had more impact on discharge of antibiotic-resistant E. coli. Figure 4 shows the contribution of CSO to its volume and ARB discharge by hourly flow rate normalized by dry-weather flow of combined sewage. The figure indicates that 43% of total volume of CSO was discharged by small-flow CSO events with normalized hourly flow rates <5, which means flow rates smaller than 5 times of the dry-weather flow of combined sewage. Meanwhile, the quantity of antibiotic-resistant E. coli, which was discharged from the small-flow CSO events, accounted for 79% of the total quantity of antibiotic-resistant E. coli discharged from CSOs. CSO events with normalized hourly flow rates <2 accounted for only 16% of the total CSO volume, but 42% of the total discharge of antibioticresistant E. coli. These results clearly indicated that small-flow CSO events have a larger impact on ARB discharge than large-flow CSO events.

DISCUSSION
Discharge of ARB from CSO occurs occasionally depending on precipitation conditions, while ARB in the WWTP effluent is continuously discharged independent of weather conditions. This study showed, on yearly quantity basis, that CSO had a much larger impact on discharge of antibiotic-resistant E. coli from a combined sewer system than the WWTP effluent. The major reason was because CSO contained a much larger concentration of fecal bacteria than secondary treatment effluent. This study targeted a typical combined sewer catchment in Japan; however, the basic conditions in this catchment are applicable to many combined sewer catchments in other countries. E. coli concentrations in combined sewer and secondary treatment effluents observed in this study were in the typical range reported in many other WWTPs, as described above. The abundance of antibiotic resistance in E. coli was not significantly different between combined sewage and secondary treatment effluents (p = 0.45).
Many of past studies in other countries also reported that the abundance of antibiotic-resistant E. coli did not significantly change. Therefore, the finding of this study on large impacts of CSO on ARB discharge are considered to be consistent in many combined sewer catchments in other countries. Nevertheless, the volume of CSO is highly dependent on catchment conditions. CSO volume varies among cities and countries mainly because of the difference in precipitation patterns. There are a limited number of studies on catchmentscale CSO volume. In this study, the volume of CSO corresponded to 29% of secondary treatment effluent. This was in similar proportion with estimation by USEPA 20 , in which 20% of combined sewage was discharged as CSO. As the least case in Sweden, the volume of CSO was only 0.4% of the total flow to the WWTP 22 . As event basis, CSO flow patterns are often affected by climate conditions. In tropical regions, where rainfall is often short and intense, CSO flows reach at a very high peak within half an hour, and end within an hour 27,28 , while CSO flows in temperate regions are rather smaller and longer to last for several hours or sometimes for days 16,17,23 . This suggests that CSOs in temperate regions could have more pollutant loads due to less dilution and longer duration. Compared with CSO quantity, E. coli concentrations in sewage have much smaller variations among locations. E. coli in sewage is reportedly 10 4 -10 5 CFU/mL, both in temperate and tropical countries 12,24,25,[29][30][31][32] . This indicates that CSO contains 2-to 5-log larger E. coli than that in WWTP effluents 12,24 . Therefore, even if the volume of CSO is much smaller than the WWTP effluent, CSO has a larger impact on E. coli discharge than the WWTP effluent in many combined sewer catchments. On the other hand, the abundance of antibiotic resistance in sewage is likely to be affected by economical and sanitation conditions in the country, rather than climate conditions. Urban sewage reflects gut microbiota and antibiotic resistome of the population 33,34 . Pärnänen et al. 35 recently reported that antibiotic resistome in urban sewage had systematic differences between Europe/North America and Africa/Asia/South America, and that low abundance of antibiotic-resistance genes was associated with low Human Development Index (HDI) of the country. Consequently, among three possible factors to affect ARB discharge from CSO, CSO flow patterns are mainly affected by climate conditions; abundance of antibiotic resistance by economical development conditions; E. coli concentrations have less variation among locations.
The results of this study suggest that it is important to control ARB discharge from CSO for reduction of ARB discharge from urban wastewater into water environment. Various countermeasures have been proposed and introduced to reduce pollutant loading of CSO on receiving water bodies, such as retention basins for stormwater storage, chemically enhanced primary treatment, etc. [36][37][38] . However, performance of these measures is limited when CSO has a large-flow rate. This study showed that small-flow CSO events contributed to a relatively larger portion of the total ARB discharge from CSO. It is because small-flow CSO contained a high concentration of E. coli due to less dilution with stormwater. Therefore, reduction of ARB from small-flow CSO events would be  important and efficient measures to reduce ARB discharge from a combined sewer catchment. In a case study in Lisbon 39 , installation of stormwater storage was predicted to reduce up to 40% of CSO volume. In the United States and France, retention treatment basin, which is stormwater tank with settling ability, was proved to be effective to reduce pollutant loading and fecal coliform from CSO 40,41 . These measures are also probably effective to reduce ARB discharge from CSO. In this study, the first flush from the sewer pipeline was not considered. The first flush possibly contains high loading of pollutants from wastewater and deposits in sewer pipes 42 . Sediments in sewer pipelines possibly contain ARB 43 , and could be flushed at the beginning of a runoff event. In such cases, more ARB could be discharged at the beginning of CSO discharge. Retention basins are effective to reduce the impacts of the first flush. Installation of these measures is expected also for control of ARB release to water bodies. This study also showed that ARB contained in the WWTP effluent still has a certain impact on ARB discharge from a combined sewer catchment, even though antibiotic-resistant E. coli discharge was 3.7-log smaller than CSO. Secondary treatment effluent is usually chlorinated before it is discharged into natural water bodies. However, it still contains a certain level of ARB 11,44 . Moreover, some reported that chlorination possibly increases the abundance of ARB because antibiotic resistance is induced via stress on its cells 11,45,46 . In some developing countries, chlorination is often emitted at WWTPs due to economical reasons. Abundance of antibiotic resistance at wastewater treatment in developing countries is reported to be higher than that in developed countries 3,13,47 . In such countries, therefore, ARB discharge from secondary treatment effluent would have a certain impact. As in the case of NDM-1, developing countries are recently concerned as the origin of ARB. ARB emerging in one country can spread to other countries and continents within a couple of years 48 . Chlorination is effective to reduce 3-log of ARB in secondary treatment effluent 44,49 . Disinfection of secondary treatment effluent is also an important measure to reduce ARB discharge from WWTP effluent in developing countries.

Combined sewer catchment
The target -combined sewer catchment was located in Kanazawa, Japan, and covered 29,900 of population and 4.04 km 2 of area. Yearly average precipitation in this area was 2400 mm. The combined sewage in the target catchment mainly consisted of municipal wastewater and stormwater runoff. The quantity of combined sewage was 28,300 m 3 /day as daily average. All the combined sewage from the coverage area was received at the three pumping stations. At the pumping stations, the combined sewage within the treatment capacity was sent to the WWTP for treatment. When the combined sewage exceeded the treatment capacity of the WWTP in wet-weather conditions, the exceeded quantity of combined sewage was discharged untreated as CSO from the pumping stations to the neighboring river. The WWTP received 4000 m 3 /h in wet-weather conditions, and 1200 m 3 /h in dry-weather conditions. The combined sewage received at the WWTP was treated by a conventional activated sludge process, and chlorinated before discharge.
Quantification and isolation of Escherichia coli E. coli was chosen as the target bacteria to represent ARB, which originated from the fecal sources. Fifty milliliters of water samples of influent wastewater (INF) and secondary treatment effluent (STE) before chlorination were collected monthly in the morning from July 2014 until June 2015 at the WWTP. After a sample was diluted with physiological saline water in an appropriate series, 1 mL of each diluted sample was filtered using 0.45μm cellulose-acetate membrane filters (A045H047W, Advantec Toyo, Tokyo, Japan). Filters were then placed on Chromocult® Coliform Agar ES (Merck KGaA, Darmstadt, Germany), and incubated at 37°C for 24 h. After counting all colonies, 20-25 colonies of E. coli were picked up from each sample into PERLCORE Trypto-Soy Broth (Eiken Chemicals, Tokyo, Japan), and incubated at 37°C overnight. Glycerol was added to a final concentration of 15-20%, and the cultures were frozen and stored at −80°C. The procedures until colony counting were conducted within 24 h after sampling.

Antibiotic-susceptibility test
Susceptibility of each E. coli isolate to six antibiotics in four classes was tested by Kirby-Bauer disk diffusion method: ciprofloxacin (CIP), norfloxacin (NFX) in fluoroquinolone class, tetracycline (TC) in tetracycline class, amoxicillin (AMX) in β-lactam class, kanamycin (KM) in aminoglycoside class, and sulfamethoxazole/trimethoprim (ST) in sulfonamide class. The isolated E. coli cultures were spread on Muller-Hinton agar (PERLCORE Sensitivity Test Agar, Eiken Chemicals, Tokyo, Japan), and antibioticsusceptibility test disks (KB Disk, Eiken Chemicals, Tokyo, Japan) were placed on the spread plates. Each antibiotic-test disk contained 5 μg of CIP, 10 μg of NFX, 30 μg of TC, 25 μg of AMX, 30 μg of KM, or a combination of 23.75 μg of sulfamethoxazole and 1.25 μg of trimethoprim. After incubation at 37°C for 18 h, the diameter of the inhibition zone around each antibiotic disk on the agar was measured. The resistance of the isolate to each antibiotic was determined from the criteria of zone diameter according to the protocol provided by the manufacturer (Supplementary  Table S1), which followed CLSI standard M100-S18 50 . According to the test protocol, the isolates, which had possible contamination, were excluded from the result. Hence, the effective number of isolates was 15-25 per sample, as shown in Table 1. The abundance of resistant isolates in each sample was calculated by where A r,i : abundance of isolates resistant to antibiotic i, R r,i : number of isolates determined as "resistance" to antibiotic i, N: the effective number of isolates in the susceptibility test. The abundance of multiple antibiotic resistance (MAR) was calculated as abundance of isolates non-susceptible to three or more antibiotic categories, according to the definition of multidrug resistance by Magiorakos et al. 51 . Paired-sample t test was used for monthly data of sewage and secondary treatment effluent in order to test the statistical difference by treatment of E. coli concentrations, abundance of antibiotic resistance, and multiple antibiotic resistance, respectively.

Volume of CSO
In the target-combined sewer catchment, all combined sewage was collected into the three pumping stations. Pumping stations Nos. 1 and 2 had two outlets each: one for WWTP and the other for CSO discharge. Pumping station No. 3 was used only for CSO discharge. Hourly operation data of the three pumping stations from July 1, 2014 until June 30, 2015 (8759 time points per outlet) were used to calculate hourly flow rate of the total combined sewage, sewage sent to WWTP for treatment, and CSO according to the following equations: where Q w(t) : hourly flow rate of sewage sent to WWTP at time t, q w(t),i : hourly flow rate of sewage sent to WWTP from pumping station i at time where Q o(t) = 0 at h. The average hourly flow rate of sewage in dry weather Q dry(h) is shown in Supplementary Fig. S1. The dilution factor of wastewater f d at time t was calculated as Estimation of E. coli and ARB discharge from CSO Concentrations of E. coli in CSO were estimated by considering dilution of wastewater with stormwater. Dilution factors of wastewater with stormwater were calculated hourly by Eq. 6. The quantity of E. coli discharged from CSO was then estimated from the hourly dilution factor of CSO, monthly data of E. coli in influent wastewater, and quantity of CSO, according to the equation below: where X CSO(t) : quantity of E. coli discharged from CSO at time t, C WW(t) : E. coli concentration in influent wastewater in the corresponding month m. Abundance of ARB in CSO was assumed to be equal to the influent wastewater. Impacts of the first flush from sewer pipelines were not considered in this study. The quantity of E. coli discharged via WWTP effluent was estimated from the monthly data of E. coli in secondary treatment effluent, quantity of WWTP effluent, and log-removal values of E. coli by chlorination, according to the equation below: where X DIS(t) : quantity of E. coli discharged via WWTP effluent at time t, C STE(t) : E. coli concentration in secondary treatment effluent in the corresponding month m, R Cl : log-removal value rate of E. coli in the chlorination step. Logremoval rate of E. coli in chlorination at the target WWTP was 3.8-log ( Supplementary Fig. S2).

DATA AVAILABILITY
The source data of the figures are available from Mendeley Data with the identifier https://doi.org/10.17632/78ds4m9nrf.1. Remaining data that support the findings of this study are available from the corresponding author upon reasonable request.