Pharmaceuticals in Tap Water: Human Health Risk Assessment and Proposed Monitoring Framework in China

Background: Pharmaceuticals are known to contaminate tap water worldwide, but the relevant human health risks have not been assessed in China. Objectives: We monitored 32 pharmaceuticals in Chinese tap water and evaluated the life-long human health risks of exposure in order to provide information for future prioritization and risk management. Methods: We analyzed samples (n = 113) from 13 cities and compared detected concentrations with existing or newly-derived safety levels for assessing risk quotients (RQs) at different life stages, excluding the prenatal stage. Results: We detected 17 pharmaceuticals in 89% of samples, with most detectable concentrations (92%) at < 50 ng/L. Caffeine (median–maximum, nanograms per liter: 24.4–564), metronidazole (1.8–19.3), salicylic acid (16.6–41.2), clofibric acid (1.2–3.3), carbamazepine (1.3–6.7), and dimetridazole (6.9–14.7) were found in ≥ 20% of samples. Cities within the Yangtze River region and Guangzhou were regarded as contamination hot spots because of elevated levels and frequent positive detections. Of the 17 pharmaceuticals detected, 13 showed very low risk levels, but 4 (i.e., dimetridazole, thiamphenicol, sulfamethazine, and clarithromycin) were found to have at least one life-stage RQ ≥ 0.01, especially for the infant and child life stages, and should be considered of high priority for management. We propose an indicator-based monitoring framework for providing information for source identification, water treatment effectiveness, and water safety management in China. Conclusion: Chinese tap water is an additional route of human exposure to pharmaceuticals, particularly for dimetridazole, although the risk to human health is low based on current toxicity data. Pharmaceutical detection and application of the proposed monitoring framework can be used for water source protection and risk management in China and elsewhere.


Sampling
The sampled cities were categorized into 4 groups according to their geographical locations: i) northern China: Beijing and Yancheng; ii) Yangtze River region: Nanjing, Hangzhou and Shanghai; iii) middle-southern China: Wuhan, Changsha and Xiamen; iv) Pearl River region in southern China: Guangzhou, Zhuhai, Macau, Shenzhen, and Hong Kong. Surface water is the dominant potable water source in the selected cities (>90% of total water supply), except for Beijing, which relies mainly on groundwater (67%) (NBSC 2009). Coagulation, sedimentation and chlorination are the most common processes in DWTPs but 18-49% of the water supplies in some sampled cities are further treated by ozonation, (bio-)activated carbon, and biofiltration (see Supplemental Material, Figure S1 and Table S2). We focused on household samples in relatively well-developed and densely-populated cities as pharmaceutical exposure could affect large populations in these locations. We also tried to maximize the geographical coverage of the samples by collecting tap water from areas with different water sources and treatment technologies under the constraint that samples had to be analyzed within 48 hours.

Analysis
The targeted pharmaceuticals were extracted with solid phase extraction methodology previously applied for sewage (Leung et al. 2012) with modifications for broadening the number of analytes and utilizing 9 isotopically-labeled standards instead of only 13 C-caffeine for reducing analytical uncertainties. Briefly, 500 mL of each sample was combined with 5 mL 5% (w/v) EDTA, acidified to pH 3-3.3 and then loaded on Hydrophilic-Lipophilic Balanced (HLB) cartridges preconditioned by methanol and water. After loading, the cartridge was rinsed with water and eluted with 4 mL methanol. The eluate was reduced to near-dryness (<0.1 mL) under a gentle stream of nitrogen, reconstituted to 0.5 mL with water and then centrifuged at 9000 rpm for 10 min. The final extract was spiked with 62.5 ng 13 C-phenacetin, 100 ng 13 C 3 -ibuprofen and 13 C 3 15 N-ciprofloxacin, and 50 ng of each remaining internal standard in order to compensate for matrix effects during instrumental quantification. For matching internal standards with analytes, we followed the quantitative methods applied in Gros et al. (2009) with slight amendments in order to minimize matrix effects. First, the slope difference of two calibration curves separately constructed in Milli-Q water and in tap water extract was calculated for each analyte. This difference was regarded as a matrix-induced interference factor and we then selected an appropriate internal standard for instrumental quantification in order to minimize the factor as close as 0 as possible. If the analyte was subject to limited matrix effects and the external calibration curve alone was the best quantification method, no internal standard was assigned. A 10 µL aliquot of extract was injected into an Agilent 1100 HPLC system (Palo Alto, CA, USA) and chromatographic separation was performed using an XBridge TM C18 column (2.1 x 50 mm, 5 µm, Waters Corporation). Analytes were ionized in electrospray ionization (ESI) source operated in positive and negative modes. Two mass transitions of each parent compound were monitored by an ABSciex 2000 QTRAP triple quadrupole tandem mass spectrometer (MS/MS) (Toronto, Canada) for quantification and confirmation in multiple reaction monitoring (MRM) mode except ibuprofen and the mass-labeled internal standards. Quantification was carried out by normalizing analyte peak area by the corresponding internal standard peak area in sample extracts and substituting into the linear equation of a seven-point external calibration curve (0-400 µg/L) constructed in Milli-Q water.

Quality assurance/quality control
Each individual sample was accompanied by a corresponding field blank (pure water fortified with ascorbic acid) and procedural blanks (n = 15) were analyzed with each sample batch. We found no background contaminations during sample collection, transportation and analysis. The matrix-matched limit of quantification (LOQ) was defined as the sum of the average and ten times the standard deviation of all procedural blank values and then corrected by the degree of matrix effects (Leung et al. 2012). LOQs ranged from 0.2 to 26.1 ng/L. Matrix-spiked absolute recoveries (n = 25, at 100 ng/L) ranged from 64.4% to 105%, with relative standard deviations mostly lower than 20% (Supplemental Material, Table S1).

Derivation of DWELs and risk assessment
The acceptable daily intake (ADI) or risk-specific dose (RSD) were derived using toxicological, microbiological or therapeutic approaches applied previously (Bruce et al. 2010;Schriks et al. 2010;Schwab et al. 2005).
For non-cancer effects, the no-observable-adverse-effect level (NOAEL) or lowest-observable-adverse-effect level (LOAEL) for different toxicity endpoints such as developmental and reproductive effects in humans or other mammals was extrapolated to an ADI by using equation S1, which includes five types of uncertainty factors (UFs): (UF1) extrapolation from LOAEL to NOAEL; (UF2) duration of exposure; (UF3) interspecies variation; (UF4) intraspecies variation; and (UF5) data quality (Schwab et al. 2005): The values and considerations of each uncertainty factor were consistent with those recommended by the U.S. EPA and in recent literature (U.S. EPA 2002; Schwab et al. 2005).
Carcinogenicity risk was assessed using slope factors (SFs), referring to the tumorigenic risk per increment of dose, of a linear non-threshold dose-response curve of the observed data extrapolated to a RSD associated with an incremental lifetime cancer risk of 10 -6 (equation S2).

RSD (µg/kg⋅d)=SF/1×10 -6 [S2]
This approach assumes that the entire range of human variation is taken into consideration and can protect public health at low doses (U.S. EPA 2005). If only evidence of carcinogenicity but no tumor incidence data was obtained from toxicity tests, a virtually safe dose (equivalent to ADI) was estimated based on the maximum tolerated dose (MTD) determined in a 90-day bioassay study corresponding to an incremental cancer risk of 10 -6 (Gaylor and Gold 1998; Bruce et al.

2010) (equation S3)
: For antibiotics, a microbiological ADI was also derived from MICs for the most sensitive human intestinal flora using equation S4 (Bruce et al. 2010;Schwab et al. 2005): where MIC 50 is the concentration inhibiting 50% of strains; MCC is the mass colonic content = 220 g/d; FA is the fraction of the oral dose available to microorganisms in the intestines; SF is the safety factor, normally equal to 1 if MIC 50 data is adequate; and BW is body weight = 60 kg (approximately average between Chinese female: 57 kg; and male: 66 kg; based on a marketing survey in China, Alvanon 2008).
If toxicological and microbiological data were deficient for a given compound, the lowest therapeutic dose was regarded as the LOAEL for derivation of a therapeutic ADI (Schwab et al. 2005). Supplemental Material, Figure S1. Locations of the 13 sampled cities in China. Table S2 for key to the location names and additional information about each location.  Davis et al. 1996 126829 -838710 labor, maternal peripartum death) a 95 th -percentile values recommended; b DWEL was calculated using the following equation: DWEL (ng/L)= [(ADI or RSD)×RSC DW ×BW×1000]/IngR DW , where RSC DW : relative source contribution of acceptable dose from drinking water, assumed to be 100% (most compounds) or 10% (caffeine only) for screening purposes; BW: body weight at each age-intervals; and IngR DW : daily ingestion rate of drinking water per day. The highest level of each pharmaceutical in tap water was compared to the corresponding DWEL for each age interval to determine RQs at different life-stages.