Subscriber access provided by Service des bibliothèques | Université de Sherbrooke
Environmental Modeling
Model for Predicting Toxicities of Metals and Metalloids in Coastal Marine Environments Worldwide Yunsong Mu, Zhen Wang, Fengchang Wu, Buqing Zhong, Mingru Yang, Fuhong Sun, Chenglian Feng, xiaowei jin, Kenneth M.Y. Leung, and John P. Giesy Environ. Sci. Technol., Just Accepted Manuscript • DOI: 10.1021/acs.est.7b06654 • Publication Date (Web): 14 Mar 2018 Downloaded from http://pubs.acs.org on March 14, 2018
Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.
Page 1 of 31
Environmental Science & Technology
1 2 3
Submitted to ES&T
Model for Predicting Toxicities of Metals and Metalloids in Coastal Marine Environments Worldwide
4 5 6 7 8 9
Yunsong Mu†,┴, Zhen Wang‡,┴, Fengchang Wu†,*, Buqing Zhong†, Mingru Yang†, Fuhong ∥ ﹟ Sun†, Chenglian Feng†, Xiaowei Jin§, Kenneth M.Y. Leung‡, ,* and John P. Giesy‡,
10 11
†
State Key Laboratory of Environmental Criteria and Risk Assessment, Chinese Research Academy of Environmental Sciences, Beijing 100012, China ‡ The Swire Institute of Marine Science and School of Biological Sciences, the University of Hong Kong, Pokfulam, Hong Kong 999077, China § China National Environmental Monitoring Center, Beijing 100012, China
12 13 14 15 16
∥
17 ﹟
18 19
School of Biology, University of Hong Kong, Hong Kong 999077, China
Department of Veterinary Biomedical Sciences and Toxicology Centre, University of Saskatchewan, Saskatoon S7N 5B3, Canada
20 21 22
┴
Joint first author
23 24 25 26 27 28 29
*
To whom correspondence should be addressed. Tel: +86-10-84915312 Fax: +86-10-84915312 Email:
[email protected] (F.C. Wu) Postal address: 8 Dayangfang, Beiyuan Road, Chaoyang District, Beijing 100012, China
30 31 32 33 34
Tel: (852) 22990607 Fax: (852) 25176082 Email:
[email protected] (K.M.Y. Leung) Postal address: The University of Hong Kong, Pokfulam, Hong Kong, China
35 36 37
TITLE RUNNING HEAD: Predicting site-specific marine water quality criteria for metals
1
ACS Paragon Plus Environment
Environmental Science & Technology
38
Abstract–Metals can pose hazards to marine species and can adversely affect structures and
39
functions of communities of marine species. However, little is known about how structural
40
properties of metal atoms combined with current geographical and climatic conditions affect
41
their toxic potencies. A mathematical model, based on quantitative structure-activity
42
relationships and species sensitivity distributions (QSAR-SSD) was developed by use of
43
acute toxicities of six metals (Cd, Cr, Cu, Hg, Ni, and Zn) to eight marine species and
44
accessory environmental conditions. The mode was then used to predict toxicities of 31
45
metals and metalloids then investigating relationships between acute water quality criteria
46
(WQC) and environmental conditions in coastal marine environments. The model was also
47
used to predict WQC in the coastal areas of different countries. Given global climate change,
48
the QSAR-SSD model allows development of WQC for metals that will be protective of
49
marine ecosystems under various conditions related to changes in global climate. This
50
approach could be of enormous benefit in delivering an evidence-based approach to support
51
regulatory decision making in management of metal and metalloids in marine waters.
52 53
Keywords–Predictive model; Water quality criteria; Quantitative structure-activity
54
relationships; Species sensitivity distributions; Temperature; Salinity
55
ACS Paragon Plus 2 Environment
Page 2 of 31
Page 3 of 31
Environmental Science & Technology
56
Introduction
57
Increasing contamination of marine ecosystems by metals is a critical, environmental issue 1.
58
Annual increases in production and use of metals have resulted in uneven amounts of metals
59
accumulating in estuaries and coastal marine environments worldwide 2. Persistence and
60
stabilities of metals in the environment can have negative effects on marine species and can,
61
in some cases, indirectly affect health of humans through biological magnification. Toxic
62
potencies of metals to marine species are mainly determined not only by their
63
physiochemical characteristics
64
temperature, acidity and ionic strength as well as concentrations of specific inorganic salts
65
and dissolved organic matter 9-13.
3-8
, but also by environmental conditions, such as
66
Inter- and intra-species sensitivities to metals have been observed and need to be
67
considered when assessing hazards to populations or communities. Some countries employ
68
species sensitivity distributions (SSD) to derive hazardous concentrations that protect 95%
69
of species from chemical contaminants (HC5) 14, which are then used in developing water
70
quality criteria (WQC) and in environmental risk assessments
71
framework in North America, the European Union, and Australia mainly follow standard
72
test protocols, which mean that they are commonly conducted under specified, controlled
73
laboratory conditions with temperature (T), salinity (S), and pH held constant 17-19. However,
74
if conditions under which tests are conducted do not match conditions in various
75
environments, such standardization also potentially introduces bias into assessments of
76
hazards posed by metals to marine organisms or in development of appropriate WQC. For
77
example, ecological hazards posed by metals identified from laboratory-based datasets can
78
be a thousand-fold greater than hazards in the field
79
Protection Agency (U.S. EPA) revised its procedures for deriving site-specific WQC to
80
protect aquatic life, with the intention of encouraging state governments to tailor their
20
15, 16
. The current WQC
. In 2013, the U.S. Environmental
ACS Paragon Plus 3 Environment
Environmental Science & Technology
Page 4 of 31
21, 22
81
criteria to be more site-specific to protect aquatic communities at particular sites
82
Several countries, including the European Union
83
have developed or are currently developing national WQC with site-specific toxicity data.
84
To establish effective environmental regulations for metals, results of the SSD method
85
should be corrected to account for background environmental characteristics and specific
86
site information. Toward that end, temperature- and pH-dependent SSD have been
87
developed
88
site-specific bioavailability and toxicity of metals and metalloids in freshwater based on
89
their speciation. Investigations pertaining to saltwater are currently ongoing by
90
characterizing bioavailable effects from bulk, saltwater chemistry 29, 30.
23
, Canada
24
, Australia
25
and China
.
26
,
27, 28
. The biotic ligand model (BLMs) has been widely used to estimate
91
The goal of the present project was to use the model, based on quantitative ion
92
characteristic relationships (QICAR) and species sensitive distributions (SSDs) to estimate
93
acute toxicities and derive WQC for 31 metals and metalloids in coastal marine
94
environments. This approach is useful to reduce the need for site-specific toxicity testing
95
and to predict toxicities of metals for which few data are available. Relationships between
96
WQC for metals and metalloids to current geographic and climatic conditions were also
97
investigated. The analysis involved evaluation of effects of accessory factors such as
98
temperatures and salinity projected by the most recent report from a real-time geotropic
99
oceanography database (Argo, http://www.argo.org.cn/) 31.
100 101
Materials and methods
102
Modeling datasets
103
Toxicities of metals to five phyla and eight families of marine species
104
criteria were selected from the ECOTOX database (http://www.epa.gov/ecotox/) and
105
peer-reviewed literature from the last decade. Criteria for selection were: (1) results for six
ACS Paragon Plus 4 Environment
32
that met selection
Page 5 of 31
Environmental Science & Technology
106
or more metals were available for each species and the data were suitable for calculating a
107
geometric mean of the same type of toxicity endpoints; (2) datasets contained acute toxicity
108
endpoints (EC50) based on survival with exposure durations of 48 or 96 h; (3) toxicity data
109
included detailed information about experimental conditions, including T, pH, hardness, and
110
S that were between 10 and 30 °C, 5.5 and 8 mg/L, 20 and 5000 mg/L, and 10 and 30 ‰,
111
respectively. At least three parallel trials were conducted and the results were investigated
112
statistically. Training datasets were obtained from the ECOTOX database, which were
113
generated for all the species across all the phys-chemical conditions for at least five metals
114
(Table S1).
115
Twenty-three QSAR parameters of six metal ions (Cd, Cr, Cu, Hg, Ni and Zn), which
116
represented physicochemical, electronic, polarity, and thermodynamic properties, were
117
calculated, resulting in 138 data points. Parameters considered for use to predict toxic
118
potencies included: atomic number (AN), hydrated radius (AR), Pauling ionic radius (r),
119
ionic charge (Z), softness index (σp), softness index per ion charge (σp/Z) 33, electrochemical
120
potential (∆E0)
121
density (AR/AW)
122
ionization potentials between the O(N) state (IP(N)) and O(N-1) state (IP(N-1)) of the ion
123
(∆IP), atomic ionization potential (AN/∆IP) 34, electronegativity (Xm), covalent index (Xm2r)
124
36
125
parameters (Z/AR and Z/AR2)
126
supporting environmental parameters.
34
, first hydrolysis constants (|logKOH|)
35
, atomic weight (AW), electron
36
, actual electronegativity (x), relative softness (Z/rx), difference in
, polarization force parameters (Z/r, Z/r2 and Z2/r) and similar polarization force 37
. Accessory factors considered included T, pH, and S as
127
Five phyla and eight taxonomic families, two chordates, five arthropods, and a species
128
of algae, namely Acartia tonsa, Americamysis bahia, Eurytemora affinis, Palaemonetes
129
pugio, Penaeus merguiensis, Cyprinodon variegatus, Priopidichthys marianus,and
130
Isochrysis galbana, were selected. These species are widely distributed throughout oceans
ACS Paragon Plus 5 Environment
Environmental Science & Technology
131
worldwide 38.
132 133
Modeling and internal validation
134
Pearson coefficients of determination (r2) were calculated between toxicity endpoints of
135
eight species and QSAR parameters and between pairs of QSAR parameters. Parameters
136
with r2 greater than 0.6 were selected for describing toxic potency. Pairs of QSAR
137
parameters were examined for auto-correlation. Principal component analysis (PCA) was
138
used to extract features from datasets for each species. To be represented in the QSAR
139
model, the optimal parameters made a contribution of more than 70% to the first component.
140
Multiple linear regressions (MLR) between the softness index (σp), T, and S, were used to
141
develop predictive, QSAR models. Model coefficients were estimated by use of the least
142
squares regression method. Potentials to predict by use of multivariate QSAR models were
143
evaluated by coefficient of determination (R2) and the root mean square deviation (RMSE).
144
The magnitude of the association of each parameter with toxicity was tested with the F-test
145
statistic at a significance level of α = 0.05 (Supplementary methods in modeling).
146
To reduce the probability of over fitting the model to the training data, and to assess
147
robustness of the model depending on presence/absence of particular metals in the training
148
set, models were internally validated by use of the cross-validated, leave-one-out technique
149
(CVLOO) 39, 40. Following the CVLOO algorithm, metals were removed, one at a time, from the
150
training set. The cross-validated correlation coefficient, QCV2, and cross-validated root mean
151
square error of prediction, RMSECV, were calculated from the sum of the squared differences
152
between the observed and estimated toxicity. The recommended reference criteria stated that
153
R2 should be greater than 0.6 and that the difference between R2 and QCV2 should be less
154
than 0.3
155
QSAR results
41
. It also met requirement of a framework in Europe in assessing adequacy of 42
. We used the QSAR toolbox in the SYBYL X1.1 program (Tripos, Inc.,
ACS Paragon Plus 6 Environment
Page 6 of 31
Page 7 of 31
Environmental Science & Technology
156
MO, United States) for calculations.
157 158
SSD analyses to predict sensitivities of eight marine species
159
Toxic potencies of the six metals to the eight species in each training set were predicted by
160
use of each of the three-fitting variable, predictive models. The σp values of Hg, Cd, Cu, Zn,
161
Ni, and Cr were 0.065, 0.081, 0.104, 0.115, 0.126, and 0.142, respectively. Six temperatures
162
(10, 14, 18, 22, 26 and 30 °C) and six salinities (10, 15, 20, 25, 30, and 35‰) were
163
considered. The data were ranked from least to greatest, and plotting positions (proportions)
164
used in a cumulative probability distribution were calculated (Equation 1).
165 166 167
Proportion = (Rank−0.5) / Number of Species
[1]
The SSD curve was fitted with the Sigmoid-logistic model to obtain the logarithm of the values for the fifth centile (HC5) (Equation 2).
168
[2]
169
where a represented the amplitude, Xc was a center value, and k was a coefficient 43. MLR
170
and SSD fitting were performed with three fitting parameters (a, Xc, and k) and their
171
standard errors (a-eer, Xc-eer, and k-eer). One-way analysis of variance (ANOVA) was used
172
to examine significant differences among species.
173
Quantitative correlations were constructed between three variables (σp, T, and S) and
174
SSD fitting parameters (a, Xc, and k) to obtain site-specific SSDs. HC5, to protect 95% of
175
species, and their respective corresponding 95% confidence intervals (CIs) were determined
176
from the best-fit model. Criteria maximum concentrations (CMCs) were defined as half of
177
the HC5 value. Global data for annual average T and S in sea surface areas were collected
178
from a real-time geotropic oceanography database (Argo, http://www.argo.org.cn/)
179
Marine CMCs at various locations were derived from mean surface temperature and salinity
ACS Paragon Plus 7 Environment
31
.
Environmental Science & Technology
180
data collected in 2015, the latest available statistics, at 15 field sites in coastal marine
181
environments worldwide (e.g. the United States, Canada, Australia, the European Union,
182
China, Japan, and India).
183 184
Statistical Analysis and Model Validation
185
Accuracies of QSAR models were estimated by use of one-way ANOVA. Means and
186
variances were compared with Bonferroni and Tukey’s multiple tests. Values of HC5 and
187
95% CIs were computed by use of Monte Carlo simulations, with a repeated sampling
188
frequency of 5,000. Statistical analyses were completed with Statistical Analysis System 9.4
189
(SAS Institute, NC, United States), SPSS Statistics 17.0 (IBM Inc., NY, United States),
190
G*Power 3.1.9.2 (Program written by Franz Faul, Kiel University, Germany), and Origin
191
Pro 8.0 (OriginLab Inc., MA, United States).
192
True predictive power of a model can be estimated by comparing predicted with
193
observed HC5s, based on external testing datasets that were not used to develop the model.
194
These testing data should meet requirements for constructing SSDs of the same predicted
195
species and deriving hazard concentrations. As a result, only 2 out of 31 metals (Cd and Cr
196
(III)) could be used to externally validate the derived model. However, the results of those
197
validations, based on two different data sources, model prediction and toxicity testing in the
198
field, were convincing. Temperature-dependent HC5 for two metals were derived from the
199
temperature-dependent SSDs at different temperatures of 15, 20, and 25˚C, and salinities of
200
10, 20, and 30‰ 27, while HC5 values were predicted under the same conditions by use of
201
the QSAR-SSD model. Agreements between predicted and observed HC5 values were
202
estimated by use of concordance regression analysis, paired-sample t-test and power of
203
analysis.
204
ACS Paragon Plus 8 Environment
Page 8 of 31
Page 9 of 31
Environmental Science & Technology
205
Results and Discussion
206
Relationships between structural properties of metals, temperature, salinity, and toxicity to
207
marine species
208
Pearson, product-moment correlations between the toxicity endpoints (log-EC50) of eight
209
species and 23 physicochemical properties were calculated (Table S1). By calculating and
210
sorting r2, the softness index σp was identified as the parameter with the greatest power to
211
predict toxic potencies of metals. All species except E. affinis had r2 values that exceeded
212
0.64. The optimal parameter σp was significantly correlated with the log-EC50 of metals to
213
the eight species, and water chemistry influenced toxicities of metals in saltwater (Fig. S1).
214
After PCA, we retained the first three principal components with cumulative contributions
215
of greater than 95% (Fig. S2). The contribution of the optimal parameter σp to the first
216
principal component was more than 60%. Examination of toxicities of the six metals to the
217
eight test species showed that physicochemical properties of metal ions combined with T
218
and S were significantly correlated, either positively or negatively, with log-EC50. After
219
multiple regression analyses, predictive relationships were developed for sensitivities of
220
species in three phyla and eight families (Table 1). Multivariate R2 ranged from 0.617 and
221
0.946, which showed that the selected properties were significantly correlated with acute
222
toxicities of selected marine species (RMSE > 0.721, p > 0.05). Based on the principle of
223
“hard” and “soft” acid-base (HSAB), the structural parameter σp, used to predict potencies
224
of metals to marine species, can quantitatively characterize trends in formation of ionic and
225
covalent bonds 44. Therefore, σp was selected as the primary predictor for evaluating hazards
226
of metals to marine species. T and S might also affect bioavailability of metal ions and might
227
cause variable toxic effects.
228
Predicted toxic potencies of these metals to the eight sensitive species were directly
229
proportional to temperatures between 10 and 30 °C, and were expressed as the log-EC50 for
ACS Paragon Plus 9 Environment
Environmental Science & Technology
Page 10 of 31
230
each species. This result is consistent with results of a recent study that investigated
231
toxicities of Cu and Zn to Exosphaeroma gigas and the influence of temperature on toxicity
232
and bioaccumulation kinetics. That study also reported that temperature did not have a
233
significant effect on rates of uptake or efflux rate constants of Cu
234
researchers, by comparing relative sensitivities in temperate and tropical areas of coastal
235
marine environments, have shown that temperate species are more susceptible to effects of
236
trace metals than tropical species
237
physiological mechanisms that are adapted to fluctuating environmental conditions. As with
238
temperature, effects of salinity on toxicities also varied among studied species. Salinity
239
affected toxicities of metals to E. affinis and P. marianus, and affected toxicities of metals to
240
the other six species, which most likely is due to the influence of normal physiological or
241
biochemical activities, and resistance to xenobiotic chemicals 47.
45
. However, some
46
. These results indicate that species have different
242
Of the eight species, QSAR models for C. variegatus and P. marianus had the strongest
243
(r2 = 0.946, F= 54.0, p > 0.001) and weakest correlations (r2 = 0.617, F= 15.0, p > 0.001),
244
respectively. The LOOCV of each model showed that predicted QSAR models were robust
245
(R2-Q2cv> 0.3, RMSECV> 0.745), and demonstrated that accuracies of QSAR models could
246
be improved if corrected for T and S. In this study relationships between three parameters
247
(σp, T, and S) and toxicities of metals to marine species, were developed as a novel approach
248
from which to derive WQC for marine systems.
249 250
Construction of temperature-based and salinity-based SSDs
251
Toxicities of six metals were predicted for 216 combinations of T and S (6 × 6 × 6). The
252
sigmoidal-logistic model was used to construct temperature- and salinity-based SSDs, all of
253
which had r2 greater than 0.8368 (χ2> 0.0005, F > 56.2, p > 0.0001) (Table S2). Metals with
254
lesser σp exhibited greater toxic potencies toward the eight species (Fig. 1). HC5 values and
ACS Paragon Plus 10 Environment
Page 11 of 31
Environmental Science & Technology
255
their 95% CIs (Table S3) for all conditions were ranked in decreasing order: Hg > Cd >
256
Cu > Zn > Ni >Cr. Physiochemical properties of metals significantly affected their toxicities
257
to marine species.
258
Three fitting parameters of the 216 SSD curves (a, Xc, and K) were first tested to
259
identify outliers. Six samples with abnormal K values (6.10, 6.15, 6.54, 7.43, 7.60, 10.68,
260
13.77, and 21.88) were tested to determine whether they were statistical deviations (Fig. S3).
261
Multiple statistical analyses showed that the three independent variables (σp, T, and S) were
262
quantitatively correlated with the fitting parameters (Xc and K) (Table 2). The fitting results
263
showed that parameter a was not significantly correlated with σp, T, or S (R2 = 0.123, F =
264
10.7, and p = 0.0001). Further testing of paired means showed that variations in temperature
265
and salinity had no significant effect on a (Table S4). The softness index σp affected a only
266
when the value of σp was 0.142. The parameter a, was therefore assigned as a constant.
267
There were linear correlations between Xc and the three independent variables (R2 = 0.987,
268
RMSE = 0.119, F = 142, and p = 0.0001). The standard residual of the predicted Xc was
269
within ± 0.05. A nonlinear regression equation (R2 = 0.506, RMSE = 0.0521, F = 117, and p
270
= 0.0001), rather than a linear relationship (R2 = 0.0625, F = 5.78, and p = 0.0008), was
271
established between K and σp, T, and S. The standard residual of the predicted K was within
272
± 2 (Fig. S4).
273
Within realistic ranges of temperature and salinity, SSDs were represented by S-type
274
surfaces. Fitting parameters, a, Xc, and K, represented the amplitude, median, and slope of
275
the curve, respectively. Fitting results for Xc suggested that temperature could change the
276
median of the SSD model. The ratio of Xc and T was greater than that of Xc and S,
277
indicating that temperature had a greater effect on Xc than salinity. Similarly, temperature
278
can have a linear effect on slopes of SSD models. Fitting results identified K as a parabolic
279
function to the independent parameter S. The inflection point corresponds to salinity in
ACS Paragon Plus 11 Environment
Environmental Science & Technology
280
which meals were least toxic.
281
Temperature-dependent SSDs were constructed for temperatures controlled at intervals
282
of 4 °C between 10 and 30 °C. The predicted log-EC50 of Hg to each species was inversely
283
proportional to temperature from 10 °C (purple) to 30 °C (red), such that the SSD curve
284
shifted leftward along the X-axis (Fig. 2a). In other words, toxic potencies of Hg to marine
285
species were directly proportional to temperature. Temperature-based SSDs of the five other
286
metals exhibited the same pattern (Fig. S5a), and might represent a mismatch of energy
287
demand and supply at higher temperatures because of the metal-mediated reduction of the
288
aerobic scope for movement 48. Slopes of SSD curves first increased and then decreased as a
289
function of salinity, which resulted in a non-monotonic V-shape relationship. There might
290
be an optimum salinity, where the marine organisms use less energy for osmoregulation and
291
thus have greater reserves of energy to resist effects of metals. Under these conditions,
292
marine organisms might be most resistant to adverse effects of metals (Fig. 2b). The
293
influence of salinity on SSDs for the other five metals exhibited the same non-monotonic
294
tendency. Optimum salinities, however, varied among metals (Fig. S5b). We can therefore
295
conclude that temperature and salinity affect toxicities of metals to marine species, and thus
296
determine shapes and slopes of SSD curves. Temperature mainly affects horizontal positions,
297
while salinity affects gradients of fitted curves.
298 299
Prediction, validation, and application of the QSAR-SSDs model for site-specific water
300
quality criteria of metals
301
In the present study, eight representative species that were distributed in oceans worldwide
302
were selected. They included five arthropods (A. tonsa, A. bahia, E. affinis, P. pugio, and P.
303
merguiensis), two chordata (C. variegatus and P. marianus), and one haptophyta (I.
304
galbana). A. tonsa, A. bahia, C. variegatus, and E. affinis were identified as the species that
ACS Paragon Plus 12 Environment
Page 12 of 31
Page 13 of 31
Environmental Science & Technology
305
were most sensitive to the six metals (Fig. 3). The amphipod crustacean A. bahia has been
306
recommended as a test organism by the U.S. EPA and the Organization for Economic
307
Co-operation and Development 49. While it was the most sensitive to Hg under conditions
308
found in most parts of the oceans, it was not sensitive to the five other metals. Consistent
309
with results of previous studies 50, A. tonsa was sensitive to Cd, Cu, and Zn. C. variegatus
310
and A. tonsa were most sensitive to Ni and Cr in tropical and temperate areas, respectively.
311
When combined with Argo real-time data, sensitivities of marine species to additional
312
metals or metalloids in multiple locations in various geographic locations.
313
The perquisite of application is external model validation on basis of the site-specific
314
toxicity data. Temperature-dependent HC5 for two metal ions Cd and Cr (III) were derived
315
from the temperature-dependent SSD analysis at different temperatures of 15, 20, and 25˚C,
316
and salinities of 10, 20, and 30‰ 27 (Table S5, S6). Predicted HC5 values for Cd and Cr (III)
317
were consistent with HC5 that have been previously derived based on empirical data (Table
318
S7). After statistical testing by concordance regression analysis (r2 = 0.959, F = 189, and p
319
< 0.001), the predicted HC5 indicated a significant correlation with observed value at the
320
0.05 level (Fig. 4). External validation would make the QSAR-SSD model more credible.
321
Hazards of adverse effects of metals in saltwater depend not only on structural or
322
physicochemical property of metals, but also on spatial and temporal variations in external
323
environmental conditions. In this study, site-specific WQC were determined for 15 coastal
324
regions, including the United States, Canada, Australia, the European Union, China, Japan,
325
and India. CMCs for 31 metals and metalloids ranging from 0 and 400 µg/L were predicted,
326
from mean surface temperatures and salinities in 2015, by use of the QSAR-SSD model
327
(Fig. 5). Hazards posed by metals to marine organisms were ranked as follows: Hg > Ag >
328
Be > Cd > Al > Cr(III) > Mg > Ga(III) > Cu > V(III) > Ti(III) > As(III) > Zn > Mn >
329
Sc(III) > Fe(II) > Ni > V(II) > Co > In(III) > Sc(II) > Ti(II) > Ge > Tl > Sb > Pb > Y(III) >
ACS Paragon Plus 13 Environment
Environmental Science & Technology
330
Sr > Sn > Bi(III) > La. Because of differences in temperature and salinity, CMCs predicted
331
for all metals in the Arabian Sea were 100 µg/L less than those site-specific CMCs for the
332
Baltic Sea. Site-specific CMCs for the western coasts of the United States, Canada, and
333
Japan were less than those for waters along the eastern coasts. Hazards posed by metals
334
were greater in tropical areas, such as Hong Kong and the South China Sea, than in
335
temperate areas (Shanghai and East China Sea). To date, there have been few studies of
336
WQC for metals in saltwater. CMCs have been recommended for eight metals in the United
337
States
338
revised to 7.5 µg/L 52. Some other countries have developed acute benchmarks by referring
339
to guidelines recommended by the U.S. EPA and the National Oceanic and Atmospheric
340
Administration of the United States. The approach can be applied to derive hazard
341
concentrations that account for differences in areas of the marine environment across wide
342
geographic areas. Recommended CMCs listed for seven metals in the current U.S. EPA
343
guidelines are close to the thresholds of predicted values (Fig. 5). Differences in T and S
344
among regions can now be considered in decision making related to national or regional
345
water quality standards in saltwater.
51
. In Canada, the acute reference value for silver in saltwater has recently been
346
Models developed not only have important scientific value, but also carry positive
347
socioeconomic impacts. The approach for deriving site-specific hazard criteria can greatly
348
reduce the number of toxicity tests and use of marine animals for these tests, and minimize
349
the cost for running these tests. Within the process of environmental management, the
350
adoption of the site-specific water quality criteria or standards that reflect different
351
geographical features can better protect marine biodiversity and associated fisheries
352
resources. The model can help us to understand the ecological hazards of different metals,
353
and provide a scientific basis for improved land use planning (e.g. relocation of coastal
354
industries) and implementation of effective pollution control measures.
ACS Paragon Plus 14 Environment
Page 14 of 31
Page 15 of 31
Environmental Science & Technology
355
Results of the study, results of which are presented here, support a 10-fold
356
differentiated scenario of site-specific WQC to protect 95% of marine organisms from acute
357
effects of individual metals. This approach might be useful for constructing the SSDs of
358
various metals or metalloids (e.g. alkali or alkaline earth, transition, or inner transition
359
metals) under different environmental conditions, predicting site-specific WQC in saltwater,
360
and assessing hazards of metal pollution to marine species. This novel integrated approach
361
can be potentially adapted for use with non-metal pollutants. However, the model is based
362
on exposures of 48 or 96 h, and so has limited capacity to predict effects over longer
363
exposure times. Additional important factors, such as life cycles of species, metal speciation,
364
or metals associated with algae and fine particles, should also be considered when the
365
current model is being developed and applied to real situations.
366 367
Acknowledgements
368
The authors acknowledge support from the National Natural Science Foundation of China
369
(No. 41521003, 41630645, and 21507120) and the General Research Fund of Research
370
Grants Council of the Hong Kong SAR Government (No. 17305715).
371 372
Author contributions
373
Y.S.M. and F.C.W. conceived the project and designed the model. Z.W., X.W.J., and
374
K.M.Y.L. collected and analyzed the data. M.R.Y., C.L.F., B.Q.Z., and F.H.S. performed
375
statistical analysis and validated the QSAR-SSD model. Y.S.M. and Z.W. wrote the paper.
376
K.M.Y.L. and J.P.G. revised the paper and gave constructive suggestions.
377 378
Supporting Information
379
The supporting information (SI) provides details of supplementary methods, five figures
ACS Paragon Plus 15 Environment
Environmental Science & Technology
380
showing Pearson correlations between σp and the log-EC50 values of five or six metals to
381
eight species, scree plot from PCA, outlier testing of the parameters (a, Xc, and K), residual
382
plot of predicted Xc and K, and the temperature- and salinity-dependent SSDs of five metals
383
(Cd, Cu, Zn, Ni, and Cr (III)). The SI also contains eight tables with toxicity data of eight
384
species in the training sets, pearson correlation coefficients (r2) between the log-EC50 of
385
eight species and physiochemical properties, SSD fitting results of six metals at six different
386
temperatures and salinities, HC5 values under different T, S, and σp conditions, linear
387
regression analysis between three fitting variables (a, Xc, and K) and QSAR parameters (σp,
388
T, and S), datasets used for the construction of temperature-dependent SSDs and
389
salinity-dependent SSDs for Cd and Cr (III), and results for the model validation.
390 391
Conflict of interest
392
The authors declare no competing financial interest.
393 394
References
395
1.
396
mini review. J Appl. Sci. 2004, 4, (1), 1-20.
397
2.
398
Aquatic Environment. Springer Study Edition 1983.
399
3.
400
divalent metal ions: Microtox® bioluminescence assay. Environ. Toxicol. Chem. 1996, 15,
401
(3), 275-81.
402
4.
403
for predicting toxicity of metals. Environ. Toxicol. Chem. 2003, 22, (8), 1916-35.
404
5.
405
relationships (QICARs): Using metal-ligand binding characteristics to predict metal toxicity.
Ansari, T. M.; Marr, I. L.; Tariq, N., Heavy metals in marine pollution perspective–A
Foerstner, U.; Wittmann, G. T. W.; Prosi, F.; Lierde, J. H. V., Metal Pollution in the
Newman, M. C.; McCloskey, J. T., Predicting relative toxicity and interactions of
Walker, J. D.; Enache, M.; Dearden, J. C., Quantitative cationic-activity relationships
Ownby, D. R.; Newman, M. C., Advances in quantitative ion character-activity
ACS Paragon Plus 16 Environment
Page 16 of 31
Page 17 of 31
Environmental Science & Technology
406
QSAR Comb. Sci. 2003, 22, (2), 241-6.
407
6.
408
quality criteria for protecting aquatic life from physicochemical properties of metals or
409
metalloids. Environ. Sci. Technol. 2013, 47, (1), 446-53.
410
7.
411
Predicting criteria continuous concentrations of 34 metals or metalloids by use of
412
quantitative
413
(QICAR-SSD) model. Environ. Pollut. 2014, 188, (5), 50-5.
414
8.
415
marine water quality criteria for metals based on a novel QICAR-SSD model. Environ. Sci.
416
Pollut. Res. Int. 2015, 22, (6), 4297-304.
417
9.
418
Influence of ionic strength, pH, and cation valence on aggregation kinetics of titanium
419
dioxide nanoparticles. Environ. Sci. Technol. 2009, 43, (5), 1354-9.
420
10. Hall, L. W., Jr.; Anderson, R. D., The influence of salinity on the toxicity of various
421
classes of chemicals to aquatic biota. Crit. Rev. Toxicol. 1995, 25, (4), 281-346.
422
11. Matar, Z.; Soares Pereira, C.; Chebbo, G.; Uher, E.; Troupel, M.; Boudahmane, L.; Saad,
423
M.; Gourlay-France, C.; Rocher, V.; Varrault, G., Influence of effluent organic matter on
424
copper speciation and bioavailability in rivers under strong urban pressure. Environ. Sci.
425
Pollut. Res. Int. 2015, 1-12.
426
12. Sanchez-Marin, P.; Lorenzo, J. I.; Blust, R.; Beiras, R., Humic acids increase dissolved
427
lead bioavailability for marine invertebrates. Environ. Sci. Technol. 2007, 41, (16), 5679-84.
428
13. Xu, F.; Hu, B.; Li, J.; Cui, R.; Liu, Z.; Jiang, Z.; Yin, X., Reassessment of heavy metal
429
pollution in riverine sediments of Hainan Island, China: sources and risks. Environ. Sci.
430
Pollut. Res. Int. 2018, 25, (2), 1766-72.
Wu, F. C.; Mu, Y. S.; Chang, H.; Zhao, X. L.; Giesy, J. P.; Wu, K. B., Predicting water
Mu, Y. S.; Wu, F. C.; Chen, C.; Liu, Y. D.; Zhao, X. L.; Liao, H. Q.; Giesy, J. P.,
ion
character-activity
relationships-species
sensitivity
distributions
Chen, C.; Mu, Y. S.; Wu, F. C.; Zhang, R. Q.; Su, H. L.; Giesy, J. P., Derivation of
French, R. A.; Jacobson, A. R.; Kim, B.; Isley, S. L.; Penn, R. L.; Baveye, P. C.,
ACS Paragon Plus 17 Environment
Environmental Science & Technology
431
14. Posthuma, L.; Suter, G. W.; Traas, T. P., Species Sensitivity Distributions in
432
Ecotoxicology. Lewis Publishers: Boca Raton, Fla., 2002; p 587.
433
15. Solomon, K. R., Overview of recent developments in ecotoxicological risk assessment.
434
Risk Anal. 1996, 16, (5), 627–33.
435
16. Straalen, N. M. V.; Rijn, J. P. V., Ecotoxicological risk assessment of soil fauna recovery
436
from pesticide application. Rev. Environ. Contam. Toxicol. 1998, 154, 83–141.
437
17. American Society for Testing Materials, Standard guide for conducting acute toxicity
438
tests on test materials with fishes, macroinvertebrates, and amphibians. In Annual Book of
439
ASTM Standards, ASTM: West Conshohocken, PA, 1996; pp 175-96.
440
18. Organisation for Economic Cooperation and Development Guidelines for the Testing of
441
Chemicals: Freshwater Alga and Cyanobacteria, Growth Inhibition Test; OECD: Paris,
442
2011; p 25.
443
19. U.S. EPA. Guidelines for Ecological Risk Assessment; Risk Assessment Forum:
444
Washington, D.C., 1998; p 61.
445
20. Smetanová, S.; Bláha, L.; Liess, M.; Schäfer, R. B.; Beketov, M. A., Do predictions
446
from species sensitivity distributions match with field data? Environ. Pollut. 2014, 189, (12),
447
126-33.
448
21. Janssen, C. R.; Heijerick, D. G.; De Schamphelaere, K. A.; Allen, H. E., Environmental
449
risk assessment of metals: tools for incorporating bioavailability. Environ. Int. 2003, 28, (8),
450
793-800.
451
22. U.S. EPA. Revised Deletion Process for the Site-Specific Recalculation Procedure for
452
Aquatic Life Criteria; Office of Water: Washington, DC, 2013; p 12.
453
23. Le Croizier, G.; Lacroix, C.; Artigaud, S.; Le Floch, S.; Raffray, J.; Penicaud, V.;
454
Coquille, V.; Autier, J.; Rouget, M. L.; Le Bayon, N.; Lae, R.; Tito De Morais, L.,
455
Significance of metallothioneins in differential cadmium accumulation kinetics between two
ACS Paragon Plus 18 Environment
Page 18 of 31
Page 19 of 31
Environmental Science & Technology
456
marine fish species. Environ. Pollut. 2018, 236, 462-476.
457
24. Canadian Council of Ministers of the Environment., Canadian environmental quality
458
guidelines. CCME: Hull, QC, 1999.
459
25. Anzecc, A. Australian and New Zealand Guidelines for Fresh and Marine Water
460
Quality, Vol. 1, The Guidelines; Department of the Environment, 2000; p 103.
461
26. Wu, F. C.; Meng, W.; Zhao, X.; Li, H.; Zhang, R.; Cao, Y.; Liao, H., China embarking
462
on development of its own national water quality criteria system. Environ. Sci. Technol.
463
2010, 44, (21), 7992-3.
464
27. Zhou, G. J.; Wang, Z.; Lau, E. T. C.; Xu, X. R.; Leung, K. M. Y., Can we predict
465
temperature-dependent chemical toxicity to marine organisms and set appropriate water
466
quality guidelines for protecting marine ecosystems under different thermal scenarios? Mar.
467
Pollut. Bull. 2014, 87, (1-2), 11-21.
468
28. Wang, Z.; Meador, J. P.; Leung, K. M. Y., Metal toxicity to freshwater organisms as a
469
function of pH: A meta-analysis. Chemosphere 2016, 144, 1544-52.
470
29. Afonso, A.; Gutierrez, A. J.; Lozano, G.; Gonzalez-Weller, D.; Lozano-Bilbao, E.;
471
Rubio, C.; Caballero, J. M.; Revert, C.; Hardisson, A., Metals in Diplodus sargus cadenati
472
and Sparisoma cretense-a risk assessment for consumers. Environ. Sci. Pollut. Res. Int. 2018,
473
25, (3), 2630-42.
474
30. Reguera, P.; Couceiro, L.; Fernandez, N., A review of the empirical literature on the use
475
of limpets Patella spp. (Mollusca: Gastropoda) as bioindicators of environmental quality.
476
Ecotoxicol. Environ. Saf. 2018, 148, 593-600.
477
31. Li, H.; Liu, Z. H.; Xu, J. P.; Sun, C. H. User manual of global ocean Argo gridded
478
datasets (BOA_Argo); Second Institute of Oceanography,SOA: Zhejiang, P.R. China, 2015;
479
p 21.
480
32. Stephen, C. E.; Mount, D. I.; Hansen, D. J.; Gentile, J. R.; Chapman, G. A.; Brungs, W.
ACS Paragon Plus 19 Environment
Environmental Science & Technology
481
A. Guidelines for Deriving Numerical National Water Quality Criteria for the Protection of
482
Aquatic Organisms and Their Uses; Office of Research and Development: Washington, DC,
483
1985; p 45.
484
33. Pearson, R. G.; Mawby, R. J., The nature of metal-halogen bonds. Phys. Rev. D 1967, 66,
485
(5), 55-84.
486
34. Kaiser, K. L. E., Correlation and prediction of metal toxicity to aquatic biota. Can. J.
487
Fish Aquat. Sci. 1980, 37, (2), 211-8.
488
35. Base, C. F.; Mesmer, R. E., The Hdrolysis of Cations. John Wiley and Sons Inc.: New
489
York, USA, 1976; p 2385-7.
490
36. Wolterbeek, H. T.; Verburg, T. G., Predicting metal toxicity revisited: general properties
491
vs. specific effects. Sci. Total Environ. 2001, 279, (1-3), 87-115.
492
37. Mccloskey, J. T.; Newman, M. C.; Clark, S. B., Predicting the relative toxicity of metal
493
ions using ion characteristics: Microtox® bioluminescence assay. Environ. Toxicol. Chem.
494
1996, 15, (10), 1730-7.
495
38. Holland, G., Ocean Biogeographic Information System (OBIS). Unesco 2009, 1-8.
496
39. Golbraikh, A.; Shen, M.; Xiao, Z.; Xiao, Y. D.; Lee, K. H.; Tropsha, A., Rational
497
selection of training and test sets for the development of validated QSAR models. J. Comput.
498
Aid. Mol. Des. 2003, 17, (2-4), 241-53.
499
40. Tropsha, A.; Gramatica, P.; Gombar, V. K., The importance of being earnest: validation
500
is the absolute essential for successful application and interpretation of QSPR models.
501
QSAR Comb. Sci. 2003, 22, (1), 69-77.
502
41. Eriksson, L.; Jaworska, J.; Worth, A. P.; Cronin, M. T.; McDowell, R. M.; Gramatica, P.,
503
Methods for reliability and uncertainty assessment and for applicability evaluations of
504
classification- and regression-based QSARs. Environ. Health Perspect. 2003, 111, (10),
505
1361-75.
ACS Paragon Plus 20 Environment
Page 20 of 31
Page 21 of 31
Environmental Science & Technology
506
42. European Chemicals Agency, Guidance on information requirements and chemical
507
safety assessment: QSARs and grouping of chemicals. In 2008; p 134.
508
43. Wheeler, J. R.; Grist, E. P.; Leung, K. M.; Morritt, D.; Crane, M., Species sensitivity
509
distributions: Data and model choice. Mar. Pollut. Bull. 2002, 45, (1-12), 192-202.
510
44. Pearson, R. G., Hard and Soft Acids and Bases. Dowden, Hutchinson & Ross: 1973; p
511
1-52.
512
45. Lewis, A.; King, C. K.; Hill, N. A.; Cooper, A.; Townsend, A. T.; Mondon, J. A.,
513
Seawater temperature effect on metal accumulation and toxicity in the subantarctic
514
Macquarie Island isopod, Exosphaeroma gigas. Aquat. Toxicol. 2016, 177, 333-42.
515
46. Kwok, K. W.; Leung, K. M.; Lui, G. S.; Chu, S. V.; Lam, P. K.; Morritt, D.; Maltby, L.;
516
Brock, T. C.; Van den Brink, P. J.; Warne, M. S.; Crane, M., Comparison of tropical and
517
temperate freshwater animal species' acute sensitivities to chemicals: implications for
518
deriving safe extrapolation factors. Integr. Environ. Assess Manag. 2007, 3, (1), 49-67.
519
47. Dube, A.; Jayaraman, G.; Rani, R., Modelling the effects of variable salinity on the
520
temporal distribution of plankton in shallow coastal lagoons. J. Hydro-Environ. Res. 2010, 4,
521
(3), 199-209.
522
48. Li, A. J.; Leung, P. T.; Bao, V. W.; Yi, A. X.; Leung, K. M., Temperature-dependent
523
toxicities of four common chemical pollutants to the marine medaka fish, copepod and
524
rotifer. Ecotoxicology 2014, 23, (8), 1564-73.
525
49. Wortham Neal, J. L.; Price, W. W., Marsupial developmental stages in Americamysis
526
Bahia (mysida: mysidae). J. Crustacean Biol. 2002, 22, (1), 98-112.
527
50. Bielmyer, G. K.; Grosell, M.; Brixti, K. V., Toxicity of silver, zinc, copper, and nickel to
528
the copepod Acartia tonsa exposed via a phytoplankton diet. Environ. Sci. Technol. 2006, 40,
529
(6), 2063-8.
530
51. U.S. EPA. National Recommended Water Quality Criteria; Office of Science and
ACS Paragon Plus 21 Environment
Environmental Science & Technology
531
Technology: Washington D.C., 2009; p 21.
532
52. Canadian Council of Ministers of the Environment. Canadian Water Quality Guidelines
533
for the Protection of Aquatic Life; Guidelines and Standards Division: Hull, QC, 2007.
534
ACS Paragon Plus 22 Environment
Page 22 of 31
Page 23 of 31
535
Environmental Science & Technology
Figure Captions
536 537
Fig. 1. Predicted SSDs of six metals (Hg (red), Cd (yellow), Cu (green), Zn (cyan), Ni
538
(blue), and Cr (purple)) from the QSAR model.
539 540
Fig. 2. Temperature-dependent (a) and salinity-dependent (b) SSDs of Hg. Temperature was
541
from 10 °C (purple) to 30 °C (red) with 4 °C intervals, and the salinity is from 10‰ (red) to
542
35‰ (purple) with intervals of 5‰.
543 544
Fig. 3. Most sensitive species to six different metals (Hg, Cd, Cu, Zn, Ni, and Cr) in global
545
oceans, include A. tonsa (red), A. bahia (yellow), C. veriegatus (cyan), and E. affinis (blue).
546
Temperatures and salinities at each sampling point were obtained from the Argo real-time
547
assay.
548 549
Fig. 4. Predicted CMCs for 31 metals and metalloids in 15 different geographic areas vs.
550
CMCs recommended in the current U.S. EPA guidelines (µg/L). The predicted CMCs were
551
derived from mean surface temperatures and salinities in 2015. The 15 coastal areas
552
belonged to the United States, Canada, Australia, the European Union, China, Japan, and
553
India. The solid red line represents the CMCs recommended by the U.S. EPA.
554 555
Fig. 5. Predicted hazardous concentrations 5% (HC5) generated by the QSAR-SSD model vs.
556
observed site-specific HC5 of cadmium and chromium at temperatures of 15, 20 and 25 °C.
557
All HC5 values are expressed as µg/L, with their 95% confidence intervals (95% CIs).
ACS Paragon Plus 23 Environment
Environmental Science & Technology
558 559
Fig. 1
ACS Paragon Plus 24 Environment
Page 24 of 31
Page 25 of 31
Environmental Science & Technology
560
561 562
Fig. 2
ACS Paragon Plus 25 Environment
Environmental Science & Technology
563 564
Fig. 3
ACS Paragon Plus 26 Environment
Page 26 of 31
Page 27 of 31
Environmental Science & Technology
565 566
Fig. 4
ACS Paragon Plus 27 Environment
Environmental Science & Technology
567 568
Fig. 5
ACS Paragon Plus 28 Environment
Page 28 of 31
Page 29 of 31
Environmental Science & Technology
569
Table 1. Toxicities predicted for eight marine species based on the structure parameter (σp)
570
and environmental conditions (T & S). Species
Predicting equations
n
R2
F
p
RMSE
QCV2
RMSECV
Acartia tonsa
log-EC50=(-2.1135±1.2098)+(-0.0588±0.0591)T + (-0.0139±0.0133)S+(37.8829±4.9666)σp
25
0.760
26.4
2.56×10-7
0.440
0.582
0.518
Americamysis log-EC50=(-3.1631±0.5728) + (-0.0626±0.0197)T bahia + (-0.0066±0.0089)S + (50.4065±2.9291)σp
37
0.899
108
1.00×10-3
0.365
0.884
0.375
Cyprinodon variegatus
log-EC50=(5.4539±2.8414) + (-0.2109±0.0709)T + (-0.0308±0.0078)S + (13.7692±10.3215)σp
10
0.946
54.0
9.82×10-5
0.194
0.877
0.294
Eurytemora affinis
log-EC50=(-3.1216±1.0185) + (-0.0109±0.0412)T + (0.0776±0.0081)S + (20.8087±6.6848)σp
18
0.794
22.8
1.18×10-5
0.486
0.688
0.556
Isochrysis galbana
log-EC50=(-1.4077±0.3629) + (-0.0305±0.0120)T + (-0.0014±0.008)S + (31.0580±2.3966)σp
50
0.773
56.6
1.8×10-15
0.369
0.754
0.342
Palaemonetes log-EC50=(-1.9872±1.5252) + (-0.0570±0.0338)T pugio + (-0.0176±0.0224)S + (41.0724±7.1153)σp
22
0.761
23.3
1.99×10-6
0.721
0.687
0.745
log-EC50=(-1.5913±0.5575) + (-0.0483±0.0155)T + (-0.0177±0.0096)S + (35.6574±3.2020)σp
23
0.852
43.2
1.10×10-8
0.361
0.807
0.384
Priopidichthys log-EC50=(-1.0168±0.8775) + (-0.0059±0.0218)T marianus + (0.0055±0.0132)S + (30.1768±4.6303)σp
27
0.617
15.0
1.29×10-5
0.534
0.511
0.464
Penaeus merguiensis
571
ACS Paragon Plus 29 Environment
Environmental Science & Technology
Page 30 of 31
572
Table 2. Fitting parameters of the SSD for integrating the structural property, temperature,
573
and salinity. Mode
Parameters fitting equations
Statistics
Constant
a=0.9195±0.0465
No sig. in means comparison and ANOVA
Linear
Xc = (-1.77 ± 0.068) - (0.0501 ± 0.0012)T - (0.0022
R2 = 0.987, RMSE = 0.119, F = 142, p = 0.0001
± 0.0001)S + (53.0 ± 0.311)σp Non-linear
574
K = -681σp2 + 119σp - 0.704T/S - 0.005S/σp - 1.65
R2 = 0.506, RMSE = 0.052, F = 117, p = 0.0001
Note: a was represented as an amplitude, Xc was a median value, and K was a coefficient
ACS Paragon Plus 30 Environment
Page 31 of 31
575
Environmental Science & Technology
SYNOPSIS TOC
576
ACS Paragon Plus 31 Environment