Distribution of Organophosphate Esters between ... - ACS Publications

Subscriber access provided by ORTA DOGU TEKNIK UNIVERSITESI KUTUPHANESI

Article

Distribution of organophosphate esters between the gas and particle phase – model predictions vs. measured data Roxana Sühring, Hendrik Wolschke, Miriam L. Diamond, Liisa M. Jantunen, and Martin Scheringer Environ. Sci. Technol., Just Accepted Manuscript • DOI: 10.1021/acs.est.6b00199 • Publication Date (Web): 04 May 2016 Downloaded from http://pubs.acs.org on May 10, 2016

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 free 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 accessible to all readers and 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.

Environmental Science & Technology 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 21

Environmental Science & Technology

1

TOC/Abstract art:

2

4

Distribution of organophosphate esters between the gas and particle phase – model predictions vs. measured data

5

Roxana Sühring1,3,6, Hendrik Wolschke1,2, Miriam L. Diamond3, Liisa M. Jantunen4 , Martin Scheringer*1,5,7

3

6 7 8 9 10 11 12 13 14

1

15

Abstract

16

Gas-particle partitioning is one of the key factors that affects the environmental fate of semi volatile

17

organic chemicals. Many organophosphate esters (OPEs) have been reported to primarily partition to

18

particles in the atmosphere. However, because of the wide range of their physicochemical properties,

19

it is unlikely that OPEs are mainly in the particle phase “as a class”.

20

We compared gas-particle partitioning predictions for 32 OPEs made by the commonly used OECD

21

POV and LRTP Screening Tool (“the Tool”) with the partitioning models of Junge-Pankow (J-P) and

22

Harner-Bidleman (H-B), as well as recently measured data on OPE gas-particle partitioning.

23

The results indicate that half of the tested OPEs partition into the gas phase. Partitioning into the gas

24

phase seems to be determined by an octanol-air partition coefficient (log KOA) < 10 and a subcooled

25

liquid vapour pressure (log PL) > –5, as well as the total suspended particle concentration (TSP) in the

26

sampling area. The uncertainty of the physicochemical property data of the OPEs did not change this

Leuphana University Lüneburg, Institute of Sustainable and Environmental Chemistry, Scharnhorststraße 1, 21335 Lüneburg Helmholtz-Zentrum Geesthacht, Centre for Materials and Coastal Research, Institute of Coastal Research, Department for Environmental Chemistry, Max-Planck-Strasse 1, 21502 Geesthacht, Germany 3 University of Toronto, Department of Earth Sciences, 22 Russell Street, Toronto, Canada M5S 3B1 4 Air Quality Processes Research Section, Environment and Climate Change Canada, Egbert, ON L0L 1N0, Canada 5 Institute for Chemical and Bioengineering, ETH Zürich, 8093 Zürich, Switzerland 6 Centre for Environment, Fisheries and Aquaculture Science (Cefas), Lowestoft, Suffolk NR33 0HT, United Kingdom 7 RECETOX, Masaryk University, 625 00 Brno, Czech Republic 2

1 ACS Paragon Plus Environment


27

estimate. Furthermore, the predictions by the Tool, J-P- and H-B-models agreed with recently

28

measured OPE partitioning.

29

1. Introduction

30

Organic flame retardants (FRs) have been used in increasing amounts since the 1950s in a variety of

31

polymer-based industrial and consumer products to comply with flammability standards. Some FRs,

32

such as penta- and octa- polybrominated diphenyl ethers (PBDEs) have been restricted from new

33

production and application under the Stockholm Convention due to their hazardous properties,

34

including high persistence and bioaccumulation potential (1) (2). Other brominated FRs (BFRs) such

35

as deca-PBDE are partially restricted or voluntarily phased-out by the industry (3). However, the

36

overall demand and production of FRs is still increasing (4). Therefore, the restrictions and phasing-

37

out of PBDEs since the early 2000s have led to the introduction of a variety of halogenated and non-

38

halogenated alternatives (5) (6).

39

One group of substitutes are organophosphate esters (OPEs), which comprise a diverse set of

40

compounds varying widely in physicochemical properties. Their production and use have been

41

increasing, in part, as a result of restrictions on PBDE use (7). In 2004 the global annual use was

42

207,200 tonnes (8), which makes OPEs high-production volume chemicals. OPE uses extend beyond

43

that of FRs. Whereas chlorinated OPEs are mainly used as flame retardants, non-chlorinated OPEs are

44

also used as plasticizers, antifoaming agents and additives in hydraulic fluids (9).

45

Whether or not OPEs are suitable replacements for hazardous FRs, such as PBDEs with known

46

carcinogenic or endocrine disrupting properties, is controversial. OPEs are, in general, regarded as

47

less persistent in the environment – one of the key criteria for evaluating the potential environmental

48

hazard of a chemical. However, our recent study suggested that many OPEs have medium to high

49

persistence

50

Ethylhexyldiphenyl phosphate, Isodecyl diphenyl phosphate, Tris(2,3-dichloro-1-propyl)phosphate,

51

Tris(cresyl) phosphate, Triphenyl phosphate, 1-Propanol, 3-bromo-2,2-bis(bromomethyl)-, 1,1',1''-

(>

1700

h)

and

that

Tetrakis(2-chloroethyl)dichloroisopentyldiphosphate,


Page 2 of 21

Page 3 of 21


52

phosphate, and potentially Tris(2-butoxyethyl) phosphate and 1,3-Dichloro-2-propanol phosphate

53

(BCMP-BCEP, EHDPP, IDDPP, MC984, TCP, TPhP, TTBNPP, TBEP, TDCPP) have a persistence similar to

54

that of PBDEs they are replacing (10). Furthermore, increasing amounts of OPEs have been detected

55

in remote areas (11–13) such as the Arctic and Antarctic, which indicates their persistence (POV) and

56

long-range transport potential (LRTP).

57

To assess potential risks posed by OPEs, it is essential to establish suitable measurement and

58

estimation methods that provide information on where and how OPEs are distributed. Most studies

59

of OPEs have focused on the particle phase (12, 14, 15), because many OPEs have been reported to

60

mostly partition to particles in the atmosphere (14, 15). Sorption to particles has, furthermore, been

61

hypothesised as one potential explanation for reduced degradation and thereby increased POV and

62

LRTP of OPEs in the environment (16). However, considering the range of physicochemical properties

63

of OPEs, notably the vapour pressure (PL) and octanol-air partition coefficient (KOA), from which gas-

64

particle partitioning of semi-volatile chemicals can be estimated (17, 18), it seems unlikely that all

65

OPEs are particle-sorbed.

66

Revisiting the literature that has reported that OPEs are primarily in the particle phase, we found that

67

these studies either investigated selected OPEs with similar properties (e.g. only halogenated OPEs),

68

screened selected samples for OPEs in the gas phase, or that the detection limits were significantly

69

higher for the gas than particle phase (12, 14, 15). Furthermore, studies using the OECD POV and LRTP

70

Screening Tool (“the Tool”) for the estimation of the environmental behaviour predicted that about

71

half of the tested OPEs should be primarily in the gas phase (10, 19). This could, of course, be an error

72

of the model or an error in the physicochemical properties used as input parameters.

73

Considering the importance of an accurate and reliable assessment of LRTP and persistence for the

74

risk assessment of OPEs, we assessed the hypothesis that OPEs primarily partition to the particle

75

phase. Due to the importance of the OECD POV and LRTP Screening Tool as tool for assessment of POV

76

and LRTP in chemical risk assessment (20) we decided to use this model, which derives gas-particle



Page 4 of 21

77

partitioning from KOA to predict the fraction of OPEs in the particle phase (fpart). Here, fpart is defined as

78

the ratio of the concentration of a chemical (x) in the particle phase (cpart(x), ng m–3) to its total

79

concentration in particle and gas-phases (∑ cpart(x), cgas(x), ng m–3), with part =

80

Our aim was to investigate the hypothesis that all OPEs are in the particle phase. We first assessed

81

the impact of uncertainty of various model input parameters on predicted fpart, identified the factors

82

determining fpart, and then evaluated and compared these predictions to other gas-particle

83

partitioning models (Harner-Bidleman (17)) (Junge-Pankow (18)), as well as measured values of fpart

84

from previous studies.

85

2. Methods

.

86

2.1 Compounds evaluated

87

The 32 compounds evaluated for gas- particle partitioning (Table 1, S1) were taken from those

88

assessed by Zhang & Sühring et al. (2016) (10), Liagkouridis et al. (2015) (19), measured in the

89

atmosphere by Möller et al. (2011) (15), and measured during a one-year weekly stationary sampling

90

at Büsum, Northern Germany (21). We included OPEs with a wide range of physicochemical

91

properties: log PL ranged from –5.6 for BCMP-BCEP, Resorcinol bis(diphenyl phosphate) (PBDPP) and

92

1-Propanol, 3-bromo-2,2-bis(bromomethyl)-, 1,1',1''-phosphate (TTBNPP) to 1.7 for Trimethyl

93

phosphate (TMP) and log KOA ranged from 4.3 for TMP to 22 for Tetrakis(2,6-dimethylphenyl)-m-

94

phenylene biphosphate (PBDMPP) (see Supporting Information Tables S2–7 for a detailed description

95

of all input data).

96 97 98 99

2.2 Log KOA based fpart predicted using the OECD Tool, fpart,T


Page 5 of 21


100

The OECD POV and LRTP Screening Tool (“the Tool”) (Version 2.2 (22)) calculates fpart,T based on the

101

octanol-air (KOA) partition coefficient. This approach assumes that absorption into the organic film

102

coating of atmospheric particles is the driving factor for the gas-particle partitioning of contaminants

103

(23), rather than the adsorption to active sites hypothesised by Junge and Pankow (24, 25). Using

104

generalised parameters for the global environment fpart,T is represented in the Tool source code by the

105

following equation(22)

106

, = φ ∗ ∗ 0.42 ∗ %∗&'* 1 − φ ∗ %∗&' + φ ∗ ∗ 0.42 ∗ %∗&' (1)

107

The volume fraction of aerosol particles in air (φ ) in the Tool is set by default to 2 ∗ 10,$$ -.. The

108

default settings in the Tool use coarse particles with an aerosol deposition velocity of 10.8 m h–1,

109

usually attributed to a particle diameter of 2.5–10 µm (26). Assuming an average density of aerosol

110

particles of 1.5 g cm–3 this value reflects a particle concentration in air of 30 µg m–3, which is the

111

target value for PM 2.5 in the Canada-wide Standard for Particulate Matter and Ozone (27). KOA in the

112

Tool is calculated as the ratio of the octanol-water partition coefficient (KOW) and the air-water

113

partition coefficient (KAW), with = '/0 . The gas constant (R) is defined as 8.314 ' -67 and the

114

temperature (TK) is set to 298.15 K.

115

Data for the Tool input parameters were extracted from a total of 50 publications (1979–2015) and

116

sorted into three datasets for the evaluation of the Tool (see Supporting Information Tables S2 – S7

117

for a detailed list of used data and sources). To test the hypothesis that uncertainty of input data is

118

responsible for the discrepancy between modelled and measured partitioning behaviour of OPEs, we

119

collected input data with high variability. We explicitly included data published by industry and

120

government authorities, as well as data from sources dating back more than 30 years. The aim of this

121

approach was to “simulate” datasets that would be used by different stakeholder groups such as

122

regulators, industry, the scientific community and NGOs and to test whether the use of different

123

datasets and sources would have a significant effect on the predicted fpart.

$

$

$

-.

'

- .45

10



124

Input all: Values of KOW, KAW and PL were obtained from modelled and measured data from all 50

125

publications (1979–2015) and data provided for registration under the European chemicals regulation

126

REACH. The median values of the combined dataset were used as input data for the Tool (a full list of

127

the data used and their sources is presented in Tables S1-S4). LogKOA was calculated as

128

89: = 89:; − 89:; (2)

129

Modelled Data: Values of KOW, KAW and PL estimated by EPI Suite, SPARC and Absolv were taken from

130

the studies by Zhang and Sühring et al. (2016) (10) and Liagkouridis et al. (2015) (19). This dataset

131

was used to specifically assess the uncertainty of modelled physicochemical properties. LogKOA was

132

calculated from log KOW and logKAW (2).

133

Outlier Removed: These are the partition coefficients, KOW, KAW, from the input-all dataset with

134

outliers removed. Outliers were defined as data points with a value below Q1 – 1.5 IQR or above Q3 +

135

1.5 IQR, respectively, with Q1 = the first quartile or 25th percentile, Q3 = the third quartile or 75th

136

percentile, and IQR being the interquartile range defined as Q3 – Q1. LogKOA was calculated as noted

137

above.

138

Predictions by the Tool (fpart,T) were compared with those calculated from the Junge-Pankow model

139

(fpart,J-P) and the Harner-Bidleman model (fpart,H-B). Input data to test these models were estimated by

140

the three most commonly used models for physicochemical properties, EPISuite, SPARC and Absolv,

141

as well as measured data.

142

2.3 Data evaluation for the Tool

143

To evaluate the impact of uncertain input parameters on the prediction of fpart,T, a sensitivity analysis

144

was performed for the Outlier Removed dataset, as well as a Monte Carlo uncertainty analysis for

145

compounds of the Outlier Removed dataset with low or intermediate fpart,T (fpart,T< 0.9).

146

The sensitivity analysis was performed as described by Morgan and Henrion (1992) (28). The

147

sensitivity of fpart,T to a specific input parameter was tested by changing the input parameters one-by6 ACS Paragon Plus Environment

Page 6 of 21

Page 7 of 21


148

one and evaluating the impact on fpart,T. A factor of 1.1 was chosen for the change of parameter

149

values. The sensitivity (S) of fpart to an individual parameter was quantified as

part,T @ >part ,& ∆@

(3)

151

with S defined as sensitivity, Bpart,T defined as change in fpart,T, C as the initial input parameter value

152

and ∆C as the change in input parameter value.

153

A Monte Carlo uncertainty analysis was performed on all compounds with partitioning behaviour

154

contrary to the hypothesis of partitioning into the particle phase, i.e. with fpart,T < 0.9. The Monte

155

Carlo calculation assumed that the values of all chemical properties are distributed log-normally

156

around their “true value”. Dispersion factors describe the width of the log-normal distribution. In the

157

default Monte Carlo analysis implemented in the Tool, this width is estimated by multiplying or

158

dividing the median by a factor of 5 for the partition coefficients (22). The Monte Carlo uncertainty

159

analysis then generates random values from the distribution of each input parameter, using the

160

defined dispersion factors and assuming uncertainties in chemical input parameters were

161

uncorrelated. The model was run for a set number N of combinations of the dispersed input

162

parameters (22). For this study the Q1 and Q3 determined in the Outlier removed dataset were used

163

as dispersion factors, as well as the (higher) default Tool dispersion factors and N was set to 200

164

(Tables S9 and S10 for calculated dispersion factors, initial and median fpart from Monte Carlo analysis

165

for default and calculated dispersion factors).

166

2.4 Junge-Pankow (fpart,J-P) and Harner-Bidleman (fpart,H-B) predictions and evaluation

167

Predictions of fpart,T from the Tool were compared to those from the Junge-Pankow (J-P) and Harner-

168

Bidleman (H-B) models. The J-P model (18) assumes adsorption to the active sites of an aerosol

169

particle according to the subcooled liquid vapour pressure (PL, Pa) of the compound, and the surface

170

area of the adsorbing aerosol particle per volume air (θ, cm2 aerosol cm–3 air) (18). fpart(J-P) is then

171

calculated as



D5EF,G,4 = 4

172

HI J HI

(4)

173

where fpart,J-P is the fraction of chemical adsorbed to air particles and c is a constant. The constant (c)

174

depends on the desorption temperature from the particle surface, the volatilisation temperature of

175

the compound, and the active sites of the aerosol. We used default values of c = 0.172 Pa cm–1 and

176

K = 1.1 ∗ 10,L H-. for urban air (23). PL data were obtained using EPI Suite’s MPBPVP v1.43 (29),

177

SPARC (30), as well as measured data (31) and literature data (7, 32) (Tables S6,7).

178

The Harner-Bidleman model (17) estimates fpart, similarly to the Tool, based on KOA and assuming

179

primary absorption into the organic layers of aerosols. fpart,H-B is calculated as

H-M

'P ∗&Q4

D5EF,N,O = '

180

P ∗&Q4$

(5)

log D = log + log U − 11.91 (6)

181

with

182

where fpart,H-B is the fraction of chemical adsorbed to air particles, KP (m3 µg–1) is the particle-gas

183

partition coefficient, TSP is the total suspended particulates concentration (µg m–3), and fOM is the

184

fraction of organic matter (OM) on the aerosol particles (gOM gTSP–1). We used default values for urban

185

air of TSP of 55 µg m–3 and fOM of 0.40 gOM gTSP–1 (23).

186

2.4 Measured fpart,M

187

Finally, fpart,T, fpart,H-B and fpart,J-P were compared to measured fpart,M values obtained by Wolschke et al.

188

(2016) (21) who measured weekly OPE air concentrations over one year at the sea-side town of

189

Büsum in northern Germany. The temperature varied between –7 °C and 20°C over the course of the

190

sampling period and TSP varied from 2 µg m–3 to 65 µg m–3 (median 16 µg m–3) (a summary of the

191

sampling and analysis techniques is presented in the SI).

192

The set of compounds for comparison of fpart,T, fpart,H-B, fpart,J-P and fpart,M included nine compounds:

193

TBEP, TCEP, TCP, TCPP, TDCPP, TEHP, TiBP, TnBP and TPhP. This reduced dataset covered a wide range


Page 8 of 21

Page 9 of 21


194

of physicochemical properties with log PL between –4.5 for TDCPP and 0.9 for TCEP and log KOA

195

between 7 for TIBP and 14 for TEHP.

196

3. Results and Discussion

197

3.1 fpart predictions by the Tool

198

The Tool calculates KOA based on the partition coefficients KAW and KOW. Data for log KAW and log KOW

199

for each compound were on average within two log units but deviated by up to eight and nine log

200

units, respectively, for log KAW (TDBPP) and log KOW (MC984) (Tables S1, S2). Datasets from models

201

(EPISuite, SPARC and Absolv) were on average lower for log KAW and higher for log KOW than data from

202

literature sources (Table S5).

203

The approximation of KOA based on other partition coefficients in itself introduces uncertainty.

204

Furthermore, scatter of KAW and KOW data can potentially add additional uncertainty and could,

205

therefore, impact the uncertainty of fpart,T.

206

For most chemicals, values of fpart,T obtained from the three data sets were consistent (Table 1, Figure

207

1): All input data sets led to fpart,T > 0.9 for 18 compounds and fpart,T < 0.1 for 13 compounds (Table 1).

208

The results were more diverse for only for seven OPEs; fpart,T estimated from the Modelled Data set

209

for isomers of TCP (TmCP, ToCP and TpCP) as well as BEHP, EHDPP, IDDPP and TEHP were considerably

210

lower (0.35, 0.70, 0.24, 0.67 and 0.87) than the results of the Input all and Outlier Removed data sets

211

(1.00, 0.94, 0.97, 0.98 and 1.0) (Table 1).

212

These results do not support the contention that OPEs partition only to the particle phase.

213

Specifically, compounds with a log KOA between 10 and 12 were partially in the gas and partially in the

214

particle phase: fpart,T for BDCPP was 0.32–0.48, for DCP was 0.14–0.28, for TDCIPP was 0.27, and for

215

TDCPP was 0.31–0.38. BEHP, EHDPP, IDDPP and the TCP isomer in the Modelled Data set were also

216

predicted to have appreciable fractions in both phases (as described above) (Figure 1, Table 1).



217 218

Figure 1: fpart,T predicted by the Tool plotted against log KOA obtained from the three input data sets (Input all,

219

Modelled Data and Outlier Removed). The dotted lines mark the areas with compounds predominantly in the

220

particle phase (fpart,T > 0.9) and compounds predominantly in the gas phase (fpart,T < 0.1). Compounds with more

221

than 10% in both phases are labelled individually.

222 223 224 225 226 227 228 229

Table 1: Predictions of fpart,T for 32 OPEs obtained from the OECD Tool Compound

Input all

Outlier removed

Modelled input data

BCMP-BCEP

1.00

1.00

1.00

BDCPP

0.48

0.48

0.32


Page 10 of 21

Page 11 of 21


BEHP

0.92

0.94

0.74

BPA-BDPP (BADP)

0.97

0.97

1.00

DCP

0.15

0.14 -3

0.28 -3

1.8 x 10

1.8 x 10

EHDPP

0.97

0.97

0.24

IDDPP

0.98

0.98

0.67

MC 984

1.00

1.00

0.99

PBDMPP

1.00

1.00

1.00

PBDPP (RDP)

1.00

1.00

1.00

TBEP

0.98

0.98

0.92

TCEP

4.5 x 10-4

4.5 x 10-4

3.4 x 10-4

TCIPP

5.8 x 10

1.6 x 10

TCP

0.74

0.66

0.54

TDBPP

1.00

1.00

1.00

TDCIPP

0.28

0.28

0.26

TDCPP

0.38

0.38

0.31

TDMPP

1.00

1.00

0.97

TEHP

1.00

-3

-3

1.00 -5 -5

2.8 x 10

-4

DOPO

2.7 x 10

-3

0.87

8.0 x 10

-6

3.3 x 10-6

-5

2.9 x 10

TEP

1.0 x 10

TIBP

9.1 x 10

8.1 x 10

TiPP

2.9 x 10-5

2.0 x 10-5

2.1 x 10-5

TmCP

1.00

1.00

0.35

TMP

-7

1.8 x 10

-3

-4

1.8 x 10

-7

2.1 x 10-7

-4

4.2 x 10

TnBP

1.0 x 10

6.0 x 10

ToCP

1.00

1.00

0.35

TpCP

1.00

1.00

0.35

TPhP

0.059

0.018

0.061

TPPP

1.00

1.00

0.98

TTBNPP

1.00

1.00

1.00

TTBPP

1.00

1.00

1.00

-4

230 231 232 233 234 235

3.2 Evaluation of the impact of uncertainty of input data on the prediction of OPE environmental

236

behaviour with the Tool



237

To evaluate the sensitivity of the tested compounds to individual changes of log KOW and log KAW as

238

well as the uncertainty of fpart,T, sensitivity and Monte Carlo uncertainty analyses were performed. For

239

the sensitivity analysis log KOW and log KAW data from the Outlier Removed data set were used. The

240

Monte Carlo uncertainty analysis was performed for all compounds of the Outlier Removed dataset

241

with fpart,T < 0.9.

242

fpart values of 22% of the 32 OPEs (7 compounds) displayed sensitivities from less than 5% (absolute)

243

to as high as 108% to individual changes of log KAW and log KOW by a factor of ±1.1 (±10%). Within this

244

group of 7 compounds, sensitivity was greatest for changes in log KAW. Very high sensitivities were

245

observed for compounds with predicted fpart,T ≤ 10–3, while compounds with predicted fpart close to 1

246

displayed the lowest sensitivities to log KOW and log KAW. However, high absolute sensitivity values for

247

compounds with fpart,T ≤ 10–3 did not change their general partitioning behaviour of fpart,T < 0.1. Effects

248

of changing log KOW and log KAW on the partitioning behaviour were only observed for compounds

249

with fpart,T between 0.1 and 0.9 (Table S8).

250

The Monte Carlo analysis revealed that uncertainties in values of log KOW and log KAW did not alter the

251

general partitioning tendency of the tested compounds using either the 25th-75th percentile or the

252

default dispersion factors (Table S9 and S10). Interestingly, even OPEs with intermediate fpart values

253

(between 0.1 and 0.9) such as BDCPP were not greatly affected by the variation: the predicted value

254

for fpart was 0.46 and the Monte Carlo analysis returned 0.4 and 0.52 for the 25th and 75th percentile.

255

In summary the results of the sensitivity as well as the uncertainty analysis indicated that the Tool’s

256

prediction of OPE partitioning between gas and particle phases did not significantly change with input

257

data uncertainty. Compounds with initial low or non-distinct gas-particle partitioning properties (fpart

258

between 0.1 and 0.9) were found to be more affected by changes in input data than compounds with

259

fpart close to 1. However, this sensitivity did not cause any significant changes in the fpart prediction.

260

3.3 Comparison of Tool (fpart,T), J-P model (fpart,J-P) and H-B model (fpart,H-B) predictions


Page 12 of 21

Page 13 of 21


261

Using the Input all or Outlier removed dataset, fpart predictions were statistically indistinguishable for

262

the two KOA-based models, the Tool and the H-B model (Table S11). The J-P model, in general,

263

predicted similar partitioning behaviour to the Tool and H-B model for 27 out of 32 compounds (81

264

%). For the remaining 6 compounds (DCP, TMP, TnBP, ToCP PBDMPP and TDBPP) the Tool/H-B model

265

and J-P model predictions differed considerably (Table S11).

266

An explanation for the considerable differences could be that all of these compounds had either a log

267

KOA between 10 and 12 (Figure 1) or a log PL between –5 and –2 and were therefore predicted to

268

partition between gas and particle phases, which means that fpart is relatively sensitive to differences

269

in input data or estimation method (Figure 1).

270

Unexpectedly, fpart predictions using the J-P model changed significantly when physicochemical input

271

data estimated by SPARC or EPISuite were used (t-test at p < 0.05). Furthermore, J-P results based on

272

EPISuite input data indicated a significantly different partitioning behaviour (t-test at p < 0.05) for 15

273

OPEs (47 %) compared to the Tool and H-B model (Table S11).

274

Interestingly, most (13) of the J-P model fpart,J-P predictions with significant differences compared to

275

the other prediction methods were lower than the Tool and H-B model. Previous studies reported

276

that the J-P model tended to overestimate fpart compared to measurements for several classes of

277

semi-volatile organic compounds (33–36).

278

3.4 Comparison of modelled partitioning with measured partitioning of OPEs (fpart,M)

279

Neither the sensitivity nor the uncertainty analysis provided compelling evidence that the uncertainty

280

of chemical property data was responsible for the observed discrepancy between the fpart values from

281

the Tool and the prevalent hypothesis in the literature that OPEs partition primarily to the particle

282

phase. We therefore re-assessed the initial hypothesis. To this end, we compared the fpart,T, fpart,H-B and

283

fpart,J-P values with measured fpart,M data. We used the dataset of Wolschke et al. (2016) (21), who

284

reported a year of weekly OPE air concentrations for 13 OPEs. The median TSP for the Wolschke et al.

285

(2016) (21) data set was 16 µg m–3. For comparison we calculated all fpart based on this value. 13 ACS Paragon Plus Environment


Page 14 of 21

286

Plotting fpart results of calculated vs. measured datasets yielded unexpected results. Calculated values

287

of fpart,T, fpart,H-B and fpart,J-P correlated well with fpart,M values (21), which reported a significant fraction

288

of OPEs in the gas-phase (Figure S1, Table 2), whereas they did not agree with the hypothesis that

289

OPEs primarily partition into the particle-phase (Figure S1, S2).

290

Furthermore, fpart,T, fpart,H-B, fpart,J-P and fpart,M all showed strong positive correlations with log KOA and

291

negative correlations with log PL (Table 2, Figures S3 and S4). The weakest correlation resulted from

292

the H-B model for log PL and the J-P model for log KOA with Pearson correlation coefficients, r, of –

293

0.40 for log PL vs. fpart,H-B and r = 0.65 for log KOA vs. fpart,J-P (Table 2). The resulting plots of fpart vs. log

294

KOA show the S-shape that is typical for the relationship between log KOA and particle partitioning with

295

0% partitioning to particles at log KOA < 10 and 100% partitioning into the particle phase at log KOA >

296

12 (Figure 1).

297

Table 2: Pearson correlation coefficient, r, for fpart results of measured and modelled datasets as well as with the

298

partition coefficients log KOW and log KOA and subcooled liquid vapour pressure log PL.

OECD POV and LRT Screening Tool fpart,M

fpart Modelled Data

fpart fpart fpart Input all Outlier Removed H-B model

fpart J-P model

log PL log KOA

–0.90 0.89

–0.70 0.84

–0.76 0.85

–0.75 0.86

–0.40 0.83

–0.91 0.65

fpart,M

1

0.95

0.83

0.80

0.87

0.90

299 300

The strong correlations between predicted and measured fpart for the nine compared OPEs (presented

301

in Figure S1) are primarily related to the agreement of predictions and measurements of fpart for OPEs

302

with distinct predicted partitioning (fpart below 0.1 or above 0.9). OPEs with predicted fpart between

303

0.1 and 0.9 displayed high variability in measured as well as modelled fpart. To identify the driving

304

factors for the variation in observed fpart for these compounds, namely TDCPP, TPhP and TCP, we

305

assessed the impact of various factors on the partitioning behaviour of OPEs.

306


Page 15 of 21


307

EPISuite, SPARC and Absolv are commonly used to estimate physicochemical properties. In our recent

308

paper (Zhang and Suhring et al. 2016) we found that predictions for physicochemical properties could

309

vary significantly between EPISuite and SPARC, whereas Absolv and EPISuite produced similar

310

datasets. We therefore, assessed the correlations of fpart estimates from the Tool, the H-B model and

311

the J-P model with measured fpart,M data using only measured physicochemical properties and those

312

estimated by EPISuite and SPARC as model inputs.

313

For all three models input data estimated with SPARC led to highest correlations with fpart,M (Figure 2).

314

The Tool had the overall best agreement with fpart,M followed by the H-B model predictions (Figure 2).

315

Unexpectedly, fpart,J-P correlated poorly with the measured results (fpart,M) when either EPISuite or

316

SPARC input data were used (Figure 2). This was unexpected because, with median values from

317

SPARC, EPISuite and literature, fpart,J-P was in good agreement with fpart,M (Table 2). Furthermore, using

318

measured PL as input also gave good agreement of fpart,J-P and fpart,M (Table 2, Figure 2).

319

Contrary to the Tool predictions, the J-P model predictions seemed to be strongly affected by the

320

variability in input data – specifically PL. These strong differences in J-P model predictions based on

321

different input data sets emphasize the need for reliable and evaluated physicochemical property

322

data in order to obtain consistent and reproducible results that can be used in chemical risk

323

assessment.

324 325

Figure 2: fpart,M vs. fpart,H-B and fpart,J-P (measured PL) (left) and fpart,M vs. fpart,J-P (right) using physicochemical

326

properties estimated by EPISuite and SPARC as well as measured PL.



327 328

3.5 Driving factors for partitioning behaviour of OPEs

329

All tests assessing potential errors and sources of uncertainty in the Tool and H-B model – the

330

evaluation of the impact of input data uncertainty, the comparison of model results with measured

331

fpart,M and correlation with physicochemical properties – led to the same conclusion: OPEs with a log

332

KOA < 10 partition into the gas-phase. This is in contrast to previous reports on OPE behaviour in the

333

environment (12, 13, 14, 15, 37).

334

To some extent, this discrepancy may be explained by the choice of analytes in the literature, e.g.

335

primarily halogenated OPEs with high log KOA. However, that does not explain the frequent reports of

336

e.g. TCEP in the particle phase of atmospheric samples around the world (12, 13, 15, 37, 38), because

337

according to the Tool, H-B and J-P model TCEP would primarily partition into the gas phase.

338

Our results indicate that uncertainty (Monte Carlo analysis) of physicochemical properties such as log

339

KAW, log KOW and log PL was not responsible for significant differences in the predicted partitioning

340

behaviour between gas- and particle phases in the Tool and H-B model. Therefore, our next step was

341

to identify and evaluate potential external drivers. External factors could be meteorological factors

342

such as temperature.

343

The temperature in the Tool is set to 298.15 K by default. However, the partitioning behaviour, as

344

calculated by the Tool, does not depend on changes in temperature (Table S12). This lack of

345

temperature dependence in the Tool could be a source of error for fpart predictions. Both log KOA as

346

well as PL are highly temperature dependent (39). The derived partition coefficient log KOA in the Tool

347

should represent this temperature dependence. Otherwise, fpart could be systematically under

348

predicted in regions with temperatures consistantly below 20°C (298.15 K) such as the Arctic. A

349

correlation of the fpart values reported by Wolschke et al. (2016) (21) with temperatures at the time of

350

sampling yielded, however, a correlation coefficient, r, (Pearson) below 0.30, perhaps because of the

351

narrow temperature range between 3 and 17oC (Table S13). The parameter with the highest 16 ACS Paragon Plus Environment

Page 16 of 21

Page 17 of 21


352

correlation with fpart,M was the particle concentration in air, with r up to 0.45 for compounds with

353

measured (and predicted) fpart > 0.8 (Table S13)..

354

Thus, we next assessed the impact of particle concentration (coarse, ø 2.5–10 μm) in the atmosphere

355

(TSPcoarse) on fpart,T. TSPcoarse varies by several orders of magnitude between different countries or

356

regions; with annual mean values between 2 μg m-3 in Powell River, Canada, to 153 μg m-3 in Delhi,

357

India (40). Daily concentrations have even been reported to reach up to 1000 μg m-3 in Beijing, China

358

(41). Locally, TSPcoarse concentrations can be increased at low temperatures or decreased by rain/snow

359

events (26).

360

In the Tool TSPcoarse is expressed as a volume fraction. The default value is 2 * 10–11, which represents

361

an average background particle concentration, equal to 30 pg m–3. However, OPEs are usually

362

produced and emitted and urban or industrial areas, with TSPcoarse in the 10–10 range or higher,

363

whereas deposition in the Arctic would suggest TSPcoarse volume fractions of 10–12. The default TSPcoarse

364

used in the Tool is therefore likely not representative for the conditions in these environments.

365

To assess the impact of varying particle concentrations in the atmosphere on fpart,T we used TSPcoarse

366

volume fractions between 10–12 and 10–9 on the Outlier Removed data set as well as the Modelled

367

Data set. This caused major changes in the partitioning behaviour of OPEs with log KOA between 6 and

368

12, whereas the partitioning of compounds with log KOA below 7 or higher than 12 was not affected

369

(Figure 3). Interestingly, differences in the general predicted partitioning behaviour between the

370

Outlier Removed data set and Modelled Data set (Table 1) disappeared at TSPcoarse above 10–11 (Tables

371

S14, S15). For TSPcoarse of 10-12 in the Outlier removed and modelled dataset, the fraction of OPEs

372

partitioning into the gas phase was 44% and 66%, respectively. At TSPcoarse of 10–11, this fraction was

373

reduced to 41% and 53%. At TSPcoarse of 10-10 and 10-9 the datasets produced the same predictions for

374

the number of OPEs in the gas phase with 28% and 25%, respectively (see Supporting Information

375

Tables S14 and S15 for detailed information).



376 377

Figure 3: Log KOA vs. fpart,T for the Outlier Removed dataset for different values of TSPcoarse; the dotted lines mark

378

the areas of distinct partitioning behaviour for all scenarios (above fpart,T = 0.9/ below fpart,T = 0.1). Compounds

379

with major fpart,T changes depending on TSPcoarse are labelled individually.

380

Our results indicate that roughly half of the OPEs into the gas phase regardless of whether

381

predictions are made with the Tool, the H-B model, or J-P model, and are consistent with

382

measurements of fpart by Wolschke et al. (2016) (21). However, the lack of temperature dependence

383

of fpart,T estimated by the Tool is a potential source of error and could lead to a systematic

384

underestimation of fpart in cold regions.

385

Of the commonly used models for estimating physicochemical properties and partition coefficients,

386

SPARC led to fpart estimates closer to the measured data than EPISuite. The unexpected discrepancy

387

between fpart,J-P estimates using modelled PL and measured fpart,M should be further investigated,

388

especially considering that J-P predictions using measured PL showed strong correlations with fpart,M.

389

Changes in fpart,T based on differences in TSP emphasize the importance of sampling location in

390

addition to time (e.g. before or after rain events) on fpart. In conclusion, our results show that OPEs 18 ACS Paragon Plus Environment

Page 18 of 21

Page 19 of 21


391

with log KOA values below 10 or a log PL above –5, such as BDCPP, DCP, DOPO, TCEP, TCIPP, TDCIPP,

392

TDCPP, TEP, TIBP, TIPP, TMP, TnBP and TPhP, should not be limited to the particle phase.

393

Acknowledgements

394 395

We would like to thank Xianming Zhang from the University of Toronto and Zhiyong Xie from the Helmholtz-Zentrum Geesthacht for their help in data analysis and evaluation.

396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431

4. References (1) Darnerud, P. Toxic effects of brominated flame retardants in man and in wildlife. Environment international 2003, 29 (6), 841–853; DOI 10.1016/S0160-4120(03)00107-7. (2) Stockholm Conference of Parties. Report of the Conference of the Parties of the Stockholm Convention on Persistent Organic Pollutants on the Work of Its Fourth Meeting: Stockholm Convention on Persistent Organic Pollutants: Geneva, 4 to 8 May 2009 2009. (3) United States Environmental Protection Agency (USEPA). DecaBDE phase-out initiative. Announcement on the EPA website. (4) China Market Research Reports. Global and China Flame Retardant Industry Report, 2014-2016 by Research In China at China Market Research Reports. http://www.chinamarketresearchreports.com/114859.html (accessed January 8, 2016). (5) Bergman, A.; Rydén, A.; Law, R. J.; Boer, J. de; Covaci, A.; Alaee, M.; Birnbaum, L.; Petreas, M.; Rose, M.; Sakai, S.; van den Eede, N.; van der Veen, I. A novel abbreviation standard for organobromine, organochlorine and organophosphorus flame retardants and some characteristics of the chemicals. Environment international 2012, 49, 57–82; DOI 10.1016/j.envint.2012.08.003. (6) Stapleton, H. M.; Klosterhaus, S.; Keller, A.; Ferguson, P. L.; van Bergen, S.; Cooper, E.; Webster, T. F.; Blum, A. Identification of flame retardants in polyurethane foam collected from baby products. Environmental science & technology 2011, 45 (12), 5323–5331; DOI 10.1021/es2007462. (7) Reemtsma, T.; Quintana, J. B.; Rodil, R.; Garcı´a-López, M.; Rodrı´guez, I. Organophosphorus flame retardants and plasticizers in water and air I. Occurrence and fate. TrAC Trends in Analytical Chemistry 2008, 27 (9), 727–737; DOI 10.1016/j.trac.2008.07.002. (8) EFRA - Cefic. Marketstatistic No 31 No. 31, 2007. (9) Marklund, A.; Andersson, B.; Haglund, P. Screening of organophosphorus compounds and their distribution in various indoor environments. Chemosphere 2003, 53 (9), 1137–1146; DOI 10.1016/S0045-6535(03)00666-0. (10) Zhang, X.; Sühring, R.; Serodio, D.; Bonnell, M.; Sundin, N.; Diamond, M. L. Novel flame retardants: Estimating the physical-chemical properties and environmental fate of 94 halogenated and organophosphate PBDE replacements. Chemosphere 2016, 144, 2401–2407; DOI 10.1016/j.chemosphere.2015.11.017. (11) Möller, A.; Sturm, R.; Xie, Z.; Cai, M.; He, J.; Ebinghaus, R. Organophosphorus flame retardants and plasticizers in airborne particles over the Northern Pacific and Indian Ocean toward the Polar Regions: evidence for global occurrence. Environmental science & technology 2012, 46 (6), 3127– 3134; DOI 10.1021/es204272v. 19 ACS Paragon Plus Environment


432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475

(12) Salamova, A.; Hermanson, M. H.; Hites, R. A. Organophosphate and halogenated flame retardants in atmospheric particles from a European Arctic site. Environmental science & technology 2014, 48 (11), 6133–6140; DOI 10.1021/es500911d. (13) Jantunen, L.M., Gawor, A., Wong, F., Bidleman, T., Wania, F., Stern, G.A., Pucko, M., Hung, H. Organophosphate ester flame retardants and plasticizers in the Canadian Arctic: Workshop on Brominate & Other Flame Retardants, Indianapolis, USA., 2014. (14) Bergman, A.; Östman, C.; Nybom, R.; Sjödin, A.; Carlsson, H.; Nilsoson, U.; Wachtmeister, C. A. Flame Retardants and Plasticisers on Particulate−in the Modern Computerized Indoor Environment. Organohalogen Compounds 1997, 33, 414–419. (15) Möller, A.; Xie, Z.; Caba, A.; Sturm, R.; Ebinghaus, R. Organophosphorus flame retardants and plasticizers in the atmosphere of the North Sea. Environmental pollution (Barking, Essex : 1987) 2011, 159 (12), 3660–3665; DOI 10.1016/j.envpol.2011.07.022. (16) Liu, Y.; Liggio, J.; Harner, T.; Jantunen, L.; Shoeib, M.; Li, S.-M. Heterogeneous OH initiated oxidation: a possible explanation for the persistence of organophosphate flame retardants in air. Environmental science & technology 2014, 48 (2), 1041–1048; DOI 10.1021/es404515k. (17) Bidleman, T. F. Atmospheric processes. Environ. Sci. Technol. 1988, 22 (4), 361–367; DOI 10.1021/es00169a002. (18) Pankow, J. F. An absorption model of gas/particle partitioning of organic compounds in the atmosphere. Atmospheric Environment 1994, 28 (2), 185–188; DOI 10.1016/13522310(94)90093-0. (19) Liagkouridis, I.; Cousins, A. P.; Cousins, I. T. Physical-chemical properties and evaluative fate modelling of 'emerging' and 'novel' brominated and organophosphorus flame retardants in the indoor and outdoor environment. The Science of the total environment 2015, 524-525, 416–426; DOI 10.1016/j.scitotenv.2015.02.106. (20) UNEP. Stockholm Convention on Persistent Organic Pollutants – Report on the assessment of chemical alternatives to endosulfan UNEP/POPS/POPRC.8/INF/28, 2012. (21) Wolschke H, Sühring R, Mi W, Möller A, Xie Z, Ebinghaus R. Atmospheric occurrence and fate of organophosphorus flame retardants and plasticizer at the German coast. Atmospheric Environment 2016, 137, 1-5; DOI: 10.1016/j.atmosenv.2016.04.028. (22) Wegmann, F.; Cavin, L.; MacLeod, M.; Scheringer, M.; Hungerbühler, K. The OECD software tool for screening chemicals for persistence and long-range transport potential. Environmental Modelling & Software 2009, 24 (2), 228–237; DOI 10.1016/j.envsoft.2008.06.014. (23) Lohmann, R.; Lammel, G. Adsorptive and Absorptive Contributions to the Gas-Particle Partitioning of Polycyclic Aromatic Hydrocarbons: State of Knowledge and Recommended Parametrization for Modeling. Environ. Sci. Technol. 2004, 38 (14), 3793–3803; DOI 10.1021/es035337q. (24) Junge, C. E. in Environmental modeling. Fate and transport of pollutants in water, air, and soil; Environmental science and technology; Wiley: New York, 1996. (25) Pankow, J. F. The calculated effects of non-exchangeable material on the gas-particle distributions of organic compounds. Atmospheric Environment (1967) 1988, 22 (7), 1405–1409; DOI 10.1016/0004-6981(88)90165-5. (26) Götz, C. W.; Scheringer, M.; MacLeod, M.; Wegmann, F.; Hungerbühler, K. Regional differences in gas–particle partitioning and deposition of semivolatile organic compounds on a global scale. Atmospheric Environment 2008, 42 (3), 554–567; DOI 10.1016/j.atmosenv.2007.08.033.


Page 20 of 21

Page 21 of 21

476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517


(27) Canadian Council of Ministers of the Environment (CCME). CANADA-WIDE STANDARDS for PARTICULATE MATTER (PM) and OZONE: Endorsed by CCME Council of Ministers, June 5-6, 2000, Quebec City. 2000. (28) Morgan, M. G.; Henrion, M.; Small, M. Uncertainty. A guide to dealing with uncertainty in quantitative risk and policy analysis; Cambridge University Press: Cambridge, New York, 1992. (29) United States Environmental Protection Agency (USEPA). EPI Suite- Estimation Programs Interface Suite: http://www.epa.gov/oppt/exposure/pubs/episuite.htm, 2013. (30) Hilal, S. H.; Saravanaraj, A. N.; Whiteside, T.; Carreira, L. A. Calculating physical properties of organic compounds for environmental modeling from molecular structure. Journal of computeraided molecular design 2007, 21 (12), 693–708; DOI 10.1007/s10822-007-9134-y. (31) Brommer, S.; Jantunen, L. M.; Bidleman, T. F.; Harrad, S.; Diamond, M. L. Determination of Vapor Pressures for Organophosphate Esters. J. Chem. Eng. Data 2014, 59 (5), 1441–1447; DOI 10.1021/je401026a. (32) van der Veen, I.; Boer, J. de. Phosphorus flame retardants: properties, production, environmental occurrence, toxicity and analysis. Chemosphere 2012, 88 (10), 1119–1153; DOI 10.1016/j.chemosphere.2012.03.067. (33) Cotham, W. E.; Bidleman, T. F. Polycyclic aromatic hydrocarbons and polychlorinated biphenyls in air at an urban and a rural site near lake michigan. Environmental science & technology 1995, 29 (11), 2782–2789; DOI 10.1021/es00011a013. (34) Harner, T.; Bidleman, T. F. Octanol−Air Pareeon Coeﬃcient for Describing Parecle/Gas Partitioning of Aromatic Compounds in Urban Air. Environ. Sci. Technol. 1998, 32 (10), 1494– 1502; DOI 10.1021/es970890r. (35) Chen, L.; Mai, B.; Xu, Z.; Peng, X.; Han, J.; Ran, Y.; Sheng, G.; Fu, J. In- and outdoor sources of polybrominated diphenyl ethers and their human inhalation exposure in Guangzhou, China. Atmospheric Environment 2008, 42 (1), 78–86; DOI 10.1016/j.atmosenv.2007.09.010. (36) Cincinelli, A. Atmospheric Occurrence and Gas-Particle Partitioning of PBDEs in an Industrialised and Urban Area of Florence, Italy. Aerosol Air Qual. Res. 2014; DOI 10.4209/aaqr.2013.01.0021. (37) Salamova, A.; Ma, Y.; Venier, M.; Hites, R. A. High Levels of Organophosphate Flame Retardants in the Great Lakes Atmosphere. Environ. Sci. Technol. Lett. 2014, 1 (1), 8–14; DOI 10.1021/ez400034n. (38) Ciccioli, P.; Cecinato, A.; Brancaleoni, E.; Montagnoli, M.; Allegrini, I. Chemical Composition of Particulate Organic Matter (POM) Collected at Terra Nova Bay in Antarctica. International Journal of Environmental Analytical Chemistry 1994, 55 (1-4), 47–59; DOI 10.1080/03067319408026208. (39) Wania, F.; Mackay, D. Tracking the Distribution of Persistent Organic Pollutants- Control strategies for these contaminants will require a better understanding of how they move around the globe. Environmental science & technology 1996, 30 (9), 390A–396A; DOI 10.1021/es962399q. (40) World Health Organization. Ambient Air Pollution Database, WHO, May 2014. http://www.who.int/phe/health_topics/outdoorair/databases/cities/en/ (accessed August 28, 2015). (41) Clean Air Alliance of China (CAAC). Air Pollution Episode Planning Gains Momentum in China - Clean Air Alliance of China. http://en.cleanairchina.org/product/6185.html (accessed January 8, 2016).


Distribution of Organophosphate Esters between ... - ACS Publications

Recommend Documents