Article pubs.acs.org/est
Modeling the Photochemical Attenuation of Down-the-Drain Chemicals during River Transport by Stochastic Methods and Field Measurements of Pharmaceuticals and Personal Care Products Seiya Hanamoto, Norihide Nakada, Naoyuki Yamashita, and Hiroaki Tanaka* Research Center for Environmental Quality Management, Graduate School of Engineering, Kyoto University, 1-2 Yumihama, Otsu, Shiga 520-0811, Japan S Supporting Information *
ABSTRACT: Existing stochastic models for predicting concentrations of down-thedrain chemicals in aquatic environments do not account for the diurnal variation of direct photolysis by sunlight, despite its being an important factor in natural attenuation. To overcome this limitation, we developed a stochastic model incorporating temporal variations in direct photolysis. To verify the model, we measured 57 pharmaceuticals and personal care products (PPCPs) in a 7.6-km stretch of an urban river, and determined their physical and biological properties in laboratory experiments. During transport along the river, 8 PPCPs, including ketoprofen and azithromycin, were attenuated by >20%, mainly owing to direct photolysis and adsorption to sediments. The photolabile PPCPs attenuated significantly in the daytime but persisted in the nighttime. The observations were similar to the values predicted by the photolysis model for the photolabile PPCPs (i.e., ketoprofen, diclofenac and furosemide) but not by the existing model. The stochastic model developed in this study was suggested to be a novel and useful stochastic model for evaluating direct photolysis of down-the-drain chemicals, which occurs during the river transport.
■
INTRODUCTION In recent years, numerous hazardous down-the-drain chemicals such as pharmaceuticals and personal care products (PPCPs),1 endocrine-disrupting chemicals,2 perfluorinated compounds,3 fluorescent whitening agents,4 and nitrosamines5 have been increasingly detected in aquatic environment. Owing to variations in their use and removal in wastewater treatment, it is difficult to predict their concentrations by mathematical models as a time series. In addition, assessments of ecological and human health risks require concentrations not as a time series but as a probability distribution.6,7 Therefore, variations in concentrations of down-the-drain chemicals over time are evaluated as probability distributions, and stochastic models such as GREAT-ER are widely used.8−10 However, existing stochastic models do not account for the variation in direct photolysis by sunlight, despite its being an important factor in natural attenuation. Seasonal, day-to-day, and diurnal variations in sunlight intensity are very high,11 especially between day and night. Furthermore, several downthe-drain chemicals are photolabile in sunlight (e.g., pharmaceuticals, ketoprofen;12 personal care products, triclosan;13 fluorescent whitening agents, distyryl biphenyl;14 and nitrosamines, N-nitrosodimethylamine15). Therefore, temporal variations in direct photolysis should be reflected in the output of models (i.e., in the probability distribution of the concentrations in aquatic environment). Although, Robinson et al.16 incorporated seasonal variation in direct photolysis into GREAT-ER, no studies have attempted to incorporate diurnal © 2013 American Chemical Society
variation as far as we know. As existing stochastic models ignore the dependence on sunlight intensity between series of river reaches, using random values in independent calculations of each reach,10,17 incorporating the diurnal variation in direct photolysis into a stochastic model needs the model structure to be modified radically. The corroboration of mathematical models needs data in field conditions. Because many factors controlling natural attenuation are interrelated, isolation of natural attenuation is difficult in field studies.18 Although some studies measured natural attenuation of down-the-drain chemicals during river transport,18−28 few have examined temporal variations. Kari et al.27 and Poiger et al.28 surveyed diurnal variations in natural attenuation of photolabile chemicals in a river in Switzerland, but water plants blocked around 85% of the sunlight in the water column. In addition, the observed concentrations of the photolabile chemicals were almost uniform within a day because of their high adsorptivities to the river sediments.28 Therefore, natural attenuation of down-the-drain chemicals and its temporal variation still needs to be clarified through field studies. We conducted this study to develop and corroborate a stochastic model that can reflect the temporal variation in direct photolysis by sunlight on the probability distributions of concentrations. We developed our “photolysis model” for an Received: Revised: Accepted: Published: 13571
August 9, 2013 October 21, 2013 November 7, 2013 November 7, 2013 dx.doi.org/10.1021/es4035478 | Environ. Sci. Technol. 2013, 47, 13571−13577
Environmental Science & Technology
Article
compound except for direct photolysis (h−1), and t = travel time (h). This equation was derived from a mass balance approach on the assumption that the natural attenuation follows a first-order reaction. kp is estimated as
urban river in the city of Kyoto, Japan. For model corroboration, we monitored 57 PPCPs in the river. To identify PPCPs suitable for the model corroboration, we determined the direct photolysis, biodegradation, sorption to sediments, and indirect photolysis of the PPCPs in the laboratory.
■
⎧ UVBi(1 − R UVB )(1 − BUVB ) i i k p = φ⎨ ⎪ i UVB ⎩ t ⎪
METHODS Site Descriptions. The field surveys were conducted along a 7.6-km stretch of the River Katsura (Figure 1), between Kuze
× + ×
Lλ(1 − 10−α λili)ελ α λiDi
UVA i(1 − R UVA i)(1 − BUVA i ) UVA t 500
∑λ = 315
Lλ(1 − 10−α λili)ελ ⎫ ⎬ ⎪ α λiDi ⎭ ⎪
(2)
where φ = quantum yield of the compound (−), UVB and UVA = sunlight intensity at Earth’s surface in those wavelengths (W/m2), RUVB and RUVA = fraction of sunlight reflected at the surface of the water body in those wavelengths (−), BUVB and BUVA = fraction of sunlight shaded by water plants in those wavelengths (−), UVBt and UVAt = annual average sunlight intensity at Earth’s surface in those wavelengths (W/m2), Lλ = annual average sunlight intensity at Earth’s surface at wavelength λ (10−3 einsteins cm−2 h−1), αλ = decadic absorption coefficient of the water body at wavelength λ (m−1), l = path length of sunlight in the water body (m), ελ = molar absorption coefficient of the compound at wavelength λ (M−1 cm−1), and D = depth of the water (m). This equation considers the attenuation of sunlight in the atmosphere and water and was derived from equations of Zepp11 and Tixier.13 Parameters with high temporal variation were evaluated as probability distributions and the rest as constants. Details are described in the Supporting Information. The effects of the model parameters on the result were estimated by means of a Monte Carlo simulation. The photolysis model uses a double loop for the Monte Carlo simulation to evaluate diurnal and day-to-day variation separately (Figure 2).
Figure 1. Locations of the wastewater treatment plants and sampling sites on the River Katsura.
Bridge (site 4) and Miyamae Bridge (site 7, 34°54′29″N, 135°43′0″E), in the city of Kyoto. The UV intensity in Kyoto at midday is UVA = 37.6 ± 10.6 W/m2, UVB = 0.85 ± 0.23 W/m2 in August and UVA = 14.9 ± 4.4 W/m2, UVB = 0.22 ± 0.06 W/m2 in December. The average river depth along the stretch ranges from 0.3 to 2.0 m over the year, and the decadic absorption coefficient of surface water at Miyamae Bridge at 340 nm averages 1.7 m−1. There is little vegetation along the river (Supporting Information (SI) Figure S1). The stretch receives water from three wastewater treatment plants (WWTPs; sites 1−3) and two tributaries (sites 5 and 6), and there is no additional significant inflow in dry weather. The annual average flow rate at Miyamae Bridge is 28.7 ± 15.5 m3/s. Around 30% of the water consists of treated wastewater. WWTP T (sites 1 and 2) is the major source of most of the target PPCPs in the stretch (SI Figure S1), because it serves most of the population in the catchment. Information on WWTPs T and K is summarized in SI Table S1. Model Setup. We assumed the three WWTPs (sites 1−3), the two tributaries (sites 5 and 6), and the upper boundary of the stretch (site 4) to be the sources of down-the-drain chemicals discharged into the stretch. We divided the river network into six reaches at sources and junctions. Assuming the river to be a one-dimensional plug flow, we estimated the concentration at the end of a given reach i as Ci =
315
∑λ = 290
(Li + L 0i)exp{−(k p + ki)ti} i
Qi
(1)
where C = predicted concentration of a compound (ng/L), L = mass loading of the compound at the source (μg/s), L0 = mass loading of the compound from prior reach (μg/s), Q = flow rate (m3/s), kp = direct photolysis rate constant of the compound (h−1), k = natural attenuation rate constant of the
Figure 2. Schematic protocol for determining the concentration of each compound by the photolysis model. x is an integer between 1000 and 10 000, set before the model runs. 13572
dx.doi.org/10.1021/es4035478 | Environ. Sci. Technol. 2013, 47, 13571−13577
Environmental Science & Technology
Article
The outer loop evaluates the day-to-day variation (k = number of days), while the inner loop evaluates the diurnal variation by changing the time of day at which the water passes beneath the Kuze Bridge (i.e., j). Assuming that the model parameters, except for UVB and UVA, do not show significant diurnal variation, we considered the sunlight intensity in calculating both the diurnal and day-to-day variations but considered the other model parameters (e.g., flow rate, mass loading of a compound) only in day-to-day variation. The model was developed in Microsoft Excel (v. 2010). Field Study. Surface water samples were collected at 7 sites (Figure 1) 3 times in winter between November 2011 and January 2012 and 3 times in summer between July and September 2012. The samples were collected every 2 h 12 times at sites 1 and 2, the major sources of most of the PPCPs; every 2 h 12 times in winter and every 1 h 24 times in summer at Miyamae Bridge, the most downstream site; and twice in the daytime at a given interval at the other sites (site 3−6). The samples were collected at Miyamae Bridge by an automatic water sampler (6712 full-size portable sampler, Teledyne ISCO, Lincoln, NE, U.S.A.); at sites 1 and 2 by automatic water sampler in summer and by grab in a stainless steel bucket in winter; and at sites 3−6 by grab in a stainless steel bucket. The samples were stored in glass bottles with ascorbic acid at 1.0 g/L in darkness and taken to the laboratory. There was no significant loss of PPCPs (data not shown). The 57 selected PPCPs in the dissolved phase were concentrated by solid-phase extraction, measured by ultraperformance liquid chromatography/tandem mass spectrometry (LC-MS/MS), and quantified by the alternative surrogate method.29 The flow rate at each site and the sunlight intensity in Kyoto were obtained as described in the Supporting Information. We used the mass balance approach to estimate the attenuation of the PPCPs. The amount of a compound still remaining at the most downstream site (site 7) relative to the total inflow from the sources (site 1−6) is defined as mass recovery (eq 3). Because the flow rate at site 7 (Q7) was not always available, the mass balance of carbamazepine, which is persistent in aquatic environments30−32 and whose diurnal variation in mass loading discharged from WWTP T is low (SI Table S3), was used to estimate the flow rate (eq 4). Mass recoveries were calculated for each sampling time at Miyamae Bridge. Because the concentration and flow at sites 1 and 2 were determined every 2 h, the travel times between these sites and Miyamae Bridge on each sampling day were considered in the calculation. For sites 3−6, the average of the 2 samples collected on the sampling day was used for calculations. In total, 108 mass recoveries were calculated. (r )j =
(L 7)j 2 ∑i = 1 (Li)j − 5
6
+ ∑i = 3 (Li)avg
2
(Q 7)j =
Ultrapure water was sterilized by autoclave and the pH was adjusted to 7.3 with phosphate buffer (6.67 mM). All 57 PPCPs were added to give an initial concentration of 10 μg/L each. The mixture was poured into test tubes made of quartz glass and exposed to natural sunlight on the shore of Lake Biwa South (35°0′6″N, 135°53′33″E). One test tube was collected at each of 0, 2, 4, 6, 8, 10, 15, 20, 30, and 60 min after the start of the experiment, and the concentrations of the PPCPs in the tube were analyzed as described above. The change in concentrations in darkness was negligible (data not shown). p-Nitroanisole-pyridine solution was used as a chemical actinometer; concentrations were analyzed by ultraperformance LC coupled to a UV/vis absorbance detector (LC-UV, Acquity, Waters Co., Milford, MA, U.S.A.). The experiment was conducted 6 times on sunny days, starting at around noon. In addition, molar absorptivities in wavelengths between 290 and 500 nm were measured with a UV− vis spectrophotometer (UV-2500PC, Shimadzu, Kyoto, Japan), and quantum yields were calculated for the PPCPs. The methods and results of the biodegradation, sorption, and indirect photolysis experiments are described in the Supporting Information.
■
Table 1. Direct Photolysis Rate Constants and Quantum Yields for 18 PPCPs first-order reaction constant (h−1) average ± SD
PPCPsa enrofloxacin ketoprofen ciprofloxacin norfloxacin furosemide ceftiofur dipyridamole ofloxacin chloramphenicol oxytetracycline diclofenac tetracycline naproxen nalidixic acid sufapyridine ifenprodil propranolol sulfamerazine
× 100 (3)
16.14 12.69 7.26 6.73 3.49 2.73 2.50 2.12 1.25 1.16 1.13 0.96 0.42 0.40 0.39 0.32 0.31 0.24
± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
10.18 4.85 3.54 3.68 0.34 0.27 0.33 0.55 0.19 0.18 0.09 0.19 0.10 0.06 0.17 0.16 0.12 0.03
quantum yield average ± SD
CV (%)b
nc
± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
36 13 28 41 6 10 19 20 9 14 12 18 26 11 38 9 34 17
5 5 6 4 6 6 5 4 6 4 6 3 5 6 5 3 5 4
0.0961 0.7420 0.0141 0.0164 0.0141 0.0133 0.0005 0.0081 0.0114 0.0007 0.2105 0.0006 0.0085 0.0006 0.0046 0.9062 0.0090 0.0078
0.0349 0.1000 0.0039 0.0067 0.0009 0.0013 0.0001 0.0016 0.0010 0.0001 0.0248 0.0001 0.0022 0.0001 0.0018 0.0789 0.0030 0.0014
a Acetaminophen, antipyrine, atenolol, azithromycin, bezafibrate, caffeine, carbamazepine, clarithromycin, clenbuterol, clofibric acid, crotamiton, cyclophosphamide, N,N-diethyl-m-tolamide(DEET), diltiazem, disopyramide, ethenzamide, griseofulvin, ibuprofen, indometacin, isopropylantipyrine, mefenamic acid, metoprolol, pirenzepine, primidone, 2-quinoxaline carboxylic acid, roxithromycin, salbutamol, sulfadimethoxine, sulfadimidine, sulfamethoxazole, sulfamonomethoxin, sulpiride, theophylline, thiamphenicol, tiamulin, and trimethoprim degraded less than 20% on average during the experiment. Though tylosin degraded more than 20% on average during the experiment, the concentration change did not follow a first-order reaction (R2 > 0.90). b The coefficient of variation. cNumber of the experiment in which concentration change followed the first-order reaction (R >0.90).
6
∑i = 1 (Lci)j − 5 + ∑i = 3 (Lci)avg (Cc7)j
RESULTS AND DISCUSSION
Direct Photolysis Experiment. Eighteen out of fifty seven PPCPs degraded more than 20% on average during the
(4)
where (r)j = mass recovery of the compound at time j (%), Li = mass loading of the compound at site i (μg/s), Lc = mass loading of carbamazepine (μg/s), Cc = concentration of carbamazepine (ng/L), Q = flow rate (m3/s), and avg indicates the average load of the 2 samples. Laboratory Experiment. Direct photolysis experiments were conducted in accordance with the U.S. Environmental Protection Agency’s harmonized test guideline 835.2210.33 13573
dx.doi.org/10.1021/es4035478 | Environ. Sci. Technol. 2013, 47, 13571−13577
Environmental Science & Technology
Article
Figure 3. Mass recoveries of 20 PPCPs at Miyamae Bridge (site 7) relative to the total inflow from the sources (sites 1−6). Vertical bars denote 50th percentile; error bars denote 25th and 75th percentiles (n = 108: data from all samplings).
Figure 4. Diurnal variations in the mass recovery of 8 PPCPs at Miyamae Bridge (site 7) relative to the total inflow from the sources (sites 1−6). Plots denote means, and error bars denote SD (n = 3 for each plot).
experiment, and their concentration changes followed a firstorder reaction (R2 > 0.90), at least three experiments. The molar absorptivities and quantum yields were determined for the 18 PPCPs. The first-order reaction constants and quantum yields of the 18 PPCPs are summarized in Table 1, and molar absorptivities are shown in Figure S2. The quinolone antibiotics, tetracycline antibiotics, ketoprofen, furosemide, and diclofenac
showed high photodegradability. The coefficients of variation of the quantum yields of 12 PPCPs (n = 3−6) were ≤20%. The quantum yield of diclofenac (0.211 ± 0.025, n = 6) was similar to reported values (0.156 ± 0.055, n = 4).34−37 Field Study. The measured concentrations and frequencies of detection of the PPCPs at each sampling site are shown in SI Table S1. The source distribution and mass recovery of PPCPs 13574
dx.doi.org/10.1021/es4035478 | Environ. Sci. Technol. 2013, 47, 13571−13577
Environmental Science & Technology
Article
Figure 5. Comparison of observed and predicted mass recoveries of 3 PPCPs: (A) momentary values; (B) probability distributionsobserved and predicted by photolysis model (red vs blue lines) and predicted by the photolysis and existing stochastic models (blue vs green lines).
detected consistently at more than one of the sources are shown in Si Figure S1 and Figure 3, respectively. Carbamazepine and caffeine were excluded from Figure 3: carbamazepine was used to calculate mass recovery, and the tributaries and the upper boundary contributed substantially to the concentration of caffeine. Mass recoveries of crotamiton, clofibric acid, and sulfamethoxazole were around 100%, indicating no appreciable attenuation. On the other hand, mass recoveries of ketoprofen, azithromycin, and ofloxacin were
