Estimating n-octanol-water partition coefficients for neutral highly hydrophobic chemicals using measured n-butanol-water partition coefficients

Abstract

Direct measurement of the n-octanol partition coefficients (K_OW) for highly hydrophobic organic chemicals is extremely difficult because of the extremely low concentrations present in the water phase. n-Butanol/water partition coefficients (K_BW) are generally much lower than K_OW due to the increased solubility of solute in the alcohol saturated aqueous phase, and therefore become easier to measure. We measured the K_BW for 25 neutral organic chemicals having measured log K_OWs ranging from 2 to 9 and 4 additional highly hydrophobic chemicals, with unmeasured K_OWs, having estimated log K_OWs ranging from 6 to 18. The measured log K_BW and log K_OW values were linearly related, r²= 0.978, and using the regression developed from the data, K_OWs were predicted for the 4 highly hydrophobic chemicals with unmeasured K_OWs. The resulting predictions were orders of magnitude lower than those predicted by a variety of computational models and suggests the estimates of K_OW in the literature for highly hydrophobic chemicals (i.e., log K_OW greater than 10) are likely incorrect by several orders of magnitude.

Introduction

Regulatory authorities and other stakeholders across the globe are tasked with assessing the risks of new and existing chemicals to human health and the environment. In many circumstances, these assessments must be made without the benefit of measured values for important environmental fate and effect processes, such as soil/sediment partitioning, or bioaccumulation by fish. For neutral organic chemicals, the n-octanol/water partition coefficient (K_OW) has been shown to be quantitatively related to many processes important to risk assessment. Early bioaccumulation models for fish were based entirely on K_OW (e.g., log BCF = 0.85 log K_OW −0.70; (Veith et al., 1979)), and K_OW has remained a fundamental parameter for estimating bioaccumulation in more recent and more sophisticated models (Connolly, 1991; Thomann et al., 1992; Arnot and Gobas, 2004; Barber et al., 2017). K_OW is also used within PBTK (Physiologically Based ToxicoKinetic) models to predict blood-water partitioning (Poulin and Krishnan, 1995) and partitioning among tissues (Bertelsen et al., 1998). K_OW has also been shown to be the largest source of uncertainty in applying food-web models to predict residues in fish tissues for chemicals with K_OW larger than 10⁴ (Burkhard, 1998; Ciavatta et al., 2009).

Measurement of K_OWs for highly hydrophobic organic chemicals, i.e., log K_OW greater than 10, is difficult, in part, due to the exceeding low concentrations of the chemicals in the aqueous phase. For example, decachlorobiphenyl (PCB-209) with a log K_OW of 8.22 (Tolls et al., 2003) can be measured successfully using a slow stir procedure (OECD, 2006) with an aqueous phase sample of 0.5 L, based on a water concentration mean of 32 ppb and a detection limit of approximately 0.2 ppb. Assuming the same analytical sensitivity, a chemical with a log K_OW of 12.22 would require the equilibration, extraction, and analysis of an aqueous phase sample of 5,000 L-- clearly, a very difficult if not impossible experimental measurement to perform.

Measurement of K_OWs for highly hydrophobic organic chemicals using the generator column technique (Devoe et al., 1981; Tewari et al., 1982; Woodburn et al., 1984; US-EPA, 1996) has similar difficulties. In general, these methods use HPLC pumps and column hardware, and are configured for flowrates of 5 mL/min and less. With a 5 mL/min flowrate, collection of a 5,000 L sample would take 694 d. Scaling columns up in size to 0.1 and 1.0 L/min flowrates would result in collection times of 34.7 and 3.47 d for a 5,000 L sample, respectively. In addition, as noted by Kuramochi et al (Kuramochi et al., 2007), micro-emulsions can be a problem with the generator column technique for chemicals with higher K_OWs. These issues make the experimental measurements with generator column problematic for highly hydrophobic chemicals.

Another issue with both the slow stir and generator column techniques is the low solubility of highly hydrophobic chemicals in n-octanol saturated with water. With lower solubility in the alcohol phase, concentrations in the aqueous phase are further lowered. For example, assuming similar K_OWs between two chemicals, one with ppm solubility and another with ppb solubility in n-octanol, the concentrations in the aqueous phase are three orders of magnitude lower for the latter chemical from solubility considerations alone.

Because of these measurement difficulties, most highly hydrophobic organic chemicals lack measured K_OW values, forcing assessments of environmental behavior and bioaccumulation potential to be based on computationally derived estimates. Unfortunately, estimated K_OWs for highly hydrophobic chemicals vary substantially across commonly used K_OW estimation algorithms. For example, decabromodiphenyl ethane (BDEthane-209) has predicted K_OW values spanning 6 orders of magnitude: log K_OW of 7.72 (OPERA (https://comptox.epa.gov/dashboard/)), 7.87 (Tetko et al., 2005), 8.70 (Wania and Dugani, 2003), 9.16 (Yue and Li, 2013), 9.68 (NICEATM via https://comptox.epa.gov/dashboard/), 10.56 (Admire et al., 2015)), 11.1 (ACD via http://www.chemspider.com/Chemical-Structure.82781.html (accessed 15:39, Apr 6, 2017)), 12.4 (Chemaxon via CSID:82781, http://www.chemspider.com/Chemical-Structure.82781.html (accessed 15:39, Apr 6, 2017)), 12.3 (Carreira et al., 1994; Hilal et al., 1995), and 13.6 (US-EPA, 2012). From these estimates, BDEthane-209 is most likely highly hydrophobic with a log K_OW of greater than 9. Nevertheless, the resulting uncertainty in selecting a K_OW is folded into the uncertainties associated with environmental behavior and bioaccumulation models used in risk assessments of such chemicals.

The intractability of using slow-stir methodologies to directly measure K_OW for highly hydrophobic chemicals and the issues associated with scaling the generator column methodology to high flowrates, e.g., 0.10 to 10 L/min, has led to the exploration of other approaches that use empirical chemical measurements to estimate K_OW. One such approach is to use a correlation between HPLC retention times/capacity factors and measured K_OWs (OECD, 2004). Currently, due to the lack of chemicals with measured log K_OWs greater than 10, this method requires extrapolations well beyond the chemicals with known K_OWs, i.e., the retention times/capacity factors are much longer/larger than those for chemicals with known K_OWs, and leads to large uncertainties in the forecasted K_OWs.

Recently, Jonker (Jonker, 2016) developed a method using polydimethylsiloxane-coated solid-phase microextraction fibers for measuring K_OW. In the method, the K_OW is determined from slow-stirring dual-flask/solid phase extraction system where concentrations of the chemical in the fiber and octanol phases after equilibration are measured. With chemical concentration in the fiber and fiber-water partition coefficient, the concentration in the aqueous phase is estimated and subsequently, used to calculate the K_OW of the chemical. The method requires the fiber-water partition coefficient and for highly hydrophobic chemicals, their determination will be challenging. Additionally, the method required equilibration times of 18–20 weeks. Overall, the fiber method extends the range where experimental measurements are possible but does have limitations (Jonker, 2016).

Another approach is to use extrathermodynamic relationships (linear free-energy relationships) (Wold, 1974; Leffler and Grunwald, 2013) between partition coefficients in different solvents. Collander (Collander, 1951) demonstrated that relationships of the form:

$$\log K_{\mathit{solvent}-1∕\mathit{water}}=A+B\times K_{\mathit{solvent}-2∕\mathit{water}}$$ (1)

exists for numerous solvent pairs where K_{solvent-1/water} and K_{solvent-2/water} are solvent-1/water and solvent-2/water equilibrium partition coefficients for organic chemicals. Leo et al. (Leo et al., 1971; Leo and Hansch, 1971) provided further follow-up comparisons and developed 20 solvent pair relationships with n-octanol/water partition coefficients (K_OW) being the independent variable in all cases (solvent-2/water in equation 1). Like the HPLC method, the lack of chemicals with measured log K_OWs greater than 10 limits this approach for estimating the K_OWs of highly hydrophobic chemicals. Therefore, further development of the Collander relationship will simply provide an extended approach for estimating highly hydrophobic chemical’s log K_OWs via measured log K_BWs.

In the relationships developed by Leo et al. (Leo et al., 1971; Leo and Hansch, 1971), the slopes of the n-butanol/water (K_BW) and n-pentanol/water (solvent-2/water) against K_OW (solvent-1/water) were 1.43 and 1.24, respectively. With slopes greater than 1.0, the n-butanol/water and n-pentanol/water partition coefficients are less than K_OWs. As discussed by Chiou (Chiou, 2003), n-butanol is much more polar than n-octanol and therefore has a greater solubility in water which causes nonpolar solutes to have a lower affinity for water-saturated n-butanol compared to water-saturated n-octanol resulting in K_BW < K_OW. In Figure 1, log K_OW data from BioByte (BioByte, 2017), Leo and Hansch (Leo and Hansch, 1971), and Westall (Westall, 1983) are plotted against their log K_BWs, and range of log K_OWs extends from −3.21 to 4.92. The data in Figure 1 spans a wide variety of chemical classes and includes sugars, alcohols, phenols, aromatic and aliphatic acids, amines, drugs, and chlorinated benzenes.

Fig. 1.

Log K_OW vs. Log K_BW for existing literature data. Regression coefficients (± standard error).

The K_BW-K_OW correlation shown in Figure 1 demonstrates a strong linear relationship over a series of chemicals with log K_OWs < 5. However, the range of the available data does not provide much support for confidently applying this regression to chemicals with log K_OWs > 5. Based on this, the objective of this study was 4-fold. The first objective was to measure additional K_BWs for chemicals with measured log K_OWs ranging from 2 to 9 and for chemicals with estimated log K_OWs exceeding 10. The second objective was to expand the K_BW-K_OW correlation shown in Figure 1 using the newly measured K_BWs and subsequently evaluate the relationship. The third objective was to measure K_OWs for a few highly hydrophobic chemicals to further evaluate the K_BW-K_OW relationship. The fourth objective was to make predictions for highly hydrophobic chemicals, i.e., estimated log K_OW > 10, with K_BW-K_OW relationship and compare these predictions to estimates from log K_OW calculators.

Materials and Methods

Reagents and Glassware Cleaning

See Supplementary Data: Tables S1-S4 for details on all chemicals and solvents used. Prior to use, n-octanol and n-butanol were further purified by distillation, with distillate collection between the alcohol boiling point and 5 °C beyond.

To eliminate cross contamination and minimize background, after washing with soap and water, all glassware was combusted at 450 °C for 4 h prior to use. The Teflon stir-bars used were cleaned with a Dionex 300 Accelerated Solvent Extractor using dichloromethane:acetone (1:1, v/v).

Slow Stir Measurement Methodology

n-Butanol and n-octanol/water partition coefficients were measured following the OECD slow stir technique (OECD, 2006). Glass jacketed reservoirs (2 L and 5 L Radnoti reservoirs, Covina, CA, USA) were set up on magnetic stir plates with water jacket spouts connected to a recirculating bath with a controlled temperature of 25.00±0.05 °C. Deionized (DI) water (saturated with n-butanol or n-octanol) was added to the reservoirs and allowed to reach temperature. Chemical dissolved in n-butanol or n-octanol (saturated with DI water) was then slowly added to the reservoirs carefully avoiding emulsion formation. After chemical was added to the reservoirs, slow-stirring was started with a vortex at the water/n-alcohol interface of less than 2.0 cm in depth and the reservoirs were covered with aluminum foil to avoid any possible photodegradation. Start times for sample collections were based solely on an estimated chemical equilibration time. Prior to sampling the reservoirs, stirring was stopped and the reservoirs were allowed to rest until quiescent conditions were obtained.

Sampling of the n-alcohol phase was done using a Pasteur pipette through the neck of the reservoir. After sampling, a subsample of n-alcohol phase was diluted using toluene or ACN, and the dilution was spiked with surrogate chemical.

For the aqueous phase, sampling was performed by draining 10–20 mL of the aqueous phase through sampling line and value to waste, and then, collecting 0.1–2 L of sample into a flask. The collected aqueous sample was analyzed using one of two methods. The first method involved the direct dilution of a subsample of the aqueous solution with ACN for analysis with HPLC instrumentation. This method was used for the polyaromatic hydrocarbons, see Supplementary Data: Table S5 for specific chemicals analyzed using this method.

The second method involved the collection of the aqueous sample into a tared volumetric flask containing the surrogate standard after evaporation of its solvent. The pre- and post-sample collection mass of the volumetric flask was used to determine the sample volume. Aqueous phase samples were liquid-liquid extracted 3x with hexane in the volumetric flasks. Prior to extraction, DI water saturated with Na₂SO₄ was added (1% sample volume) to the flask to prevent emulsion formation during extraction. Extraction flasks were mixed vigorously using a Teflon stir-bar for at least 30 minutes and then allowed to settle. The hexane layer was pipetted off into a 60 mL test tube and evaporated using a N-Evap 111 nitrogen evaporator (Organomation, Berlin, MA) with a heated water bath maintained at a temperature range of 50–90 °C or a Rocket solvent evaporator (Thermo Scientific, Waltham, MA). Once hexane reached a volume < 1 mL, it was transferred to a micro-GC-vial via Pasteur pipette. The 60 mL test tube was rinsed 3x with hexane and pooled into the micro-vial. Samples were concentrated under N₂ to a final volume of 20–200 µL, spiked with appropriate internal standard, and vortex mixed.

Quantification

Depending upon the chemical, analytes were quantified using one of the following systems: Agilent 6890N/5975C Gas Chromatography/Mass Spectrometry (GC/MS), Thermo Scientific Q-Exactive Orbitrap coupled with a Trace 1310 Gas Chromatograph (QE-GC/MS), Thermo Scientific DFS High Resolution GC/MS, or an Agilent 1100 series High Performance Liquid Chromatography (HPLC) coupled to fluorescence (FLD) or Diode Array (DAD) detector. Selected-ion monitoring (SIM) for target and confirming masses was used for mass spectrometer detection. Quantification was performed using the isotopic dilution method for chemicals that required liquid extraction. For the samples resulting from the direct dilution of the aqueous phase, they were analyzed using HPLC. See Supplemental Data: Table S5 for additional analytical details.

The n-butanol/water partition coefficients (log K_BW) and n-octanol/water partition coefficients (log K_OW) were computed using the equation:

$$\log K_{\mathit{AW}}=\log \frac{C_A}{C_W}$$ (2)

In equation 2, log K_AW is the alcohol/water partition coefficient that represents either n-butanol or n-octanol, C_A is the concentration in the alcohol phase, and C_W is the concentration in the water phase. n-Butanol and n-octanol/water partition coefficients are plotted against collection days, and for a successful log K_AW measurement experiment, the slope of the regression of log K_AW against time must be not significantly different from zero (α= 0.05) with a minimum of 4 collections over time (criteria in OECD 123 methodology (OECD, 2006)). Additional samples were collected where necessary to satisfy the equilibrium criterion.

An aqueous phase saturated with n-butanol or n-octanol was used as the method blank. Blanks were processed with each slow stir collection to monitor possible hexane extraction and sample preparation contamination. Aqueous phase sample concentrations were method blank corrected using the blank for the corresponding collection days.

Generator Column Measurement Methodology

An additional approach for measuring n-octanol/water partition coefficients was using the generator column technique (Woodburn et al., 1984; US-EPA, 1996). Stainless steel HPLC columns (25 cm in length with 2.0 µm frits) were packed with diatomaceous earth (Celite 545, Celite 560 or sieved >100 mesh Celite 560) with ½ cm of 100–120 mesh glass beads on the downstream end of the column. After measuring the concentration of the chemical in the water saturated n-octanol solution, the columns were coated with the solution using a gentle vacuum to draw the solution through the column. For each measurement, columns were placed in series in a water jacket connected to a recirculating bath with a temperature of 25.0±0.05 °C. Using Gilson HLPC 303 and 306 pumps, water saturated with n-octanol was passed through the columns and the exiting water passed through activated octadecyl solid phase extraction (SPE) columns (Bakerbond 7020–06). After sample collection, SPE columns were dried with nitrogen gas, spiked with the analytical surrogate solution, and then, eluted using hexane followed by toluene. The eluted extracts were concentrated using Rocket Evaporator to approximately 1 mL and then, to 200 µL using N-Evap nitrogen evaporator. The extracts were transferred to micro-GC-vials via Pasteur pipette with 3x rinses of hexane. Extracts were concentrated to a final volume of 20–200 µL, spiked with appropriate internal standard, and vortex mixed. Analytes were quantified using selected-ion monitoring with GC/MS systems described above.

Partition coefficients were computed using equation 2. For an individual chemical, partition coefficients were measured using different columns and different flowrates through the columns to demonstrate equilibrium conditions with the measurements (See Supplemental Data: Table S8 for experimental conditions).

Results and Discussion

In addressing the first objective of this study, we measured the K_OW for three chemicals (benz[a]anthracene, benzo[a]pyrene, and PCB-209) for the purpose of insuring comparability between our measurement methods and the literature since we intended to combine our measure K_BW values with K_OW values. The measurements of PCB-209 by the two methods were essentially identical (average of 8.17 by slow stir and 8.18 by generator column; Table S6) and both values agree well with the literature value of 8.22 (Table 1). For further method validation, our measured slow stir value for benz[a]anthracene was 5.77 versus a literature average of 5.85. For benzo[a]pyrene, our measurement; is higher than the literature average (6.77 vs. 6.09; Table 1), though, our slow stir value is in good agreement with the recently reported slow stir value of 6.69±0.02 by Jonker (Jonker, 2016). Taken together we believe these results affirm the comparability of measurement methods, and support the use of literature K_OW values in building a K_BW-K_OW relationship.

Table 1.

n-Butanol/water (K_BW) and n-octanol/water partition coefficients (K_OW).

Chemical	Measured Log KBW		Measured Log KOW		Estimated Log KOWs
	This Study a	Literature	This Study a	Literature b, d	Eq. Regression c	EPISuite	SPARC	ACD
Benzene	1.92 (0.02; 2)	1.85 [1], 1.90 [2]		2.13	1.90	1.99	2.06	2.22
Chlorobenzene	2.43 (0.02; 2)	2.38 [2]		2.84	2.81	2.64	2.53	2.81
Bromobenzene	2.53 (0.02; 2)			2.99	2.98	2.88	2.67	2.99
Naphthalene	2.76 (0.02; 3)			3.40 (0.17; 33)	3.39	3.17	3.41	3.45
Acenaphthene	3.21 (0.04; 2)			3.97 (0.21; 6)	4.18	4.15	4.06	4.19
Fluorene	3.28 (0.02; 2)			4.14 (0.12; 11)	4.31	4.02	4.20	4.16
Phenanthrene	3.46 (0.01; 3)			4.49 (0.14; 22)	4.63	4.35	4.74	4.68
Anthracene	3.56 (0.06; 3)			4.63 (0.25; 21)	4.82	4.35	4.69	4.68
PCB-1	3.64 (0.06; 3)			4.54 (0.15; 4)	4.95	4.40	4.65	4.45
Fluoranthene	3.73 (0.13; 3)			5.00 (0.28; 17)	5.11	4.93	5.29	5.17
Pyrene	3.74 (0.05; 3)			5.10 (0.21; 14)	5.12	4.93	5.25	5.17
Chrysene	4.07 (0.02; 4)			5.69 (0.27; 13)	5.72	5.52	5.90	5.91
Benz[a]anthracene	4.11 (0.02; 4)		5.76 (0.14; 2)	5.85 (0.08; 7) e	5.79	5.52	5.85	5.91
Hexachlorobenzene	4.19 (----; 1)			5.53 (0.25; 17)	5.93	5.86	6.04	4.89
Benzo(g,h,i)perylene	4.21 (0.20; 3)			6.62 (0.51; 7)	5.97	6.70	7.04	6.89
Indeno(1,2,3-cd)pyrene	4.23 (0.16; 3)			6.65 (0.43; 4)	6.00	6.70	7.09	6.89
Benzo[a]pyrene	4.28 (0.03; 4)		6.77 (0.21; 2)	6.09 (0.31; 7) e	6.08	6.11	6.54	6.40
PCB-77	4.68 (0.04; 3)			6.58 (0.34; 3)	6.80	6.34	6.28	6.00
PCB-155	4.72 (0.10; 3)			7.23 (0.28; 4)	6.87	7.62	7.66	6.83
Di-n-hexyl phthalate	4.91 (0.04; 2)			6.82 (0.10; 6)	7.20	6.57	6.29	6.95
PCB-209	5.20 (0.04; 5)		8.18 (0.12, 4)	8.22 (0.17; 10) e	7.72	10.2	10.5	8.10
BDE-209	5.47 (0.04; 4)		8.37 (0.21, 2)		8.21	12.1	12.0	11.2
Di-n-octyl phthalate	5.52 (0.04; 2)			8.15 (0.04; 3)	8.28	8.54	8.03	9.08
Di-n-decyl phthalate	5.95 (0.05; 2)			9.05 (0.31, 2)	9.06	10.5	9.71	11.2
TBPH	6.21 (0.19; 2)		9.21 (0.19; 2)		9.52	12.0	10.9	10.8

^aaverage (propagated error, n).

^baverage (standard deviation, n).

^cPredicted using Equation 3 and measured K_BWs.

^dSee Supplementary Data: Table S7 for sources of the literature log K_OW values.

^eMeasured log K_OW from this study not included in literature average. 1 - (Westall, 1983). 2 - (Hansch et al., 1995).

We also measured log K_OW for decabromodiphenyl ether (BDE-209; 8.37 by generator column) and bis-(2-ethylhexyl)-3,4,5,6-tetrabromophthalate (TBPH; 9.21 by slow stir), to provide additional data to help extend the K_BW-K_OW relationship at high log K_OW. From a methodological perspective, measuring the K_OWs of these highly hydrophobic chemicals was very demanding. The challenges for slow stir method included the extraction of large volumes of aqueous phase, long equilibration times, low solubilities of the chemicals in the alcohol phase, insuring that the chemical was dissolved in alcohol phase, and having analytical instrumentation optimized for the lowest detection limits possible. The OECD guidance (OECD, 2006) discusses these issues, and we cannot over-emphasize their importance to making these measurements. The challenges for the generator column method included pumping large volumes of water through the columns, getting the columns to come to equilibrium, dissolving the chemicals to coat the diatomaceous earth, and demonstrating that K_OW was independent of flowrate conditions.

The log K_BWs for 29 chemicals were measured using the OECD slow stir technique (Table 1 and Table 2). Among these chemicals, literature K_BW values were available for benzene and chlorobenzene (Westall, 1983; Hansch et al., 1995; Table 1) and these agreed closely with our measurements (Table 1). The collected log K_BWs from this study only and their available log K_OWs data are shown in Figure 2. The data are plotted with the following regression equation:

$$\log K_{\mathit{OW}}=1.7769(\pm 0.0560)\log K_{\mathit{BW}}-1.5158(\pm 0.2365)r^2=0.978n=25$$ (3)

Chemical classes were categorized to qualitatively show that there is no significant bias relating to compound structure (Figure 2). Additional data within each class would be needed to resolve any bias concerns.

Table 2.

Measured n-butanol/water (K_BW) and estimated n-octanol/water partition coefficients (K_OW).

Chemical	Measured Log KBW	Estimated Log KOWs
	This Studya	Eq. 3 Regressionb	OPERAc	Tetko et al.e	NICEATMc	Upperf	ACDg	Chemaxonh	SPARCi	EPISuitec
BDEthane-209	5.60 (0.14, 2)	8.43 (7.78 – 9.09)	7.72d	7.87	9.68	10.6	11.1	12.4	12.3	13.6
m-TCP	5.78 (0.06; 4)	8.76 (8.09 – 9.45)	8.47d	9.79	9.73	13.2	11.7	12.7	15.1	14.5
p-TCP	5.82 (0.08; 4)	8.82 (8.16 – 9.49)	8.48d	9.80	10.3	13.2	11.2	12.7	15.0	14.5
p-QTCP	6.47 (0.13; 3)	9.98 (9.29 – 10.67)	6.06d	10.4	11.9	16.7	14.3	16.4	19.6	18.9

^aaverage (propagated error, n).

^bThis study, predicted using Equation 3 and (95% prediction interval).

^cUS-EPA via https://comptox.epa.gov/dashboard (accessed August 15, 2018).

^dPrediction unreliable: Outside of global applicability domain and low local applicability domain.

^eALOGPS 2.1 (Tetko et al., 2005).

^fUpper (Admire et al., 2015).

^gACD via http://www.chemspider.com/Chemical-Structure.82781.html (accessed August 15, 2018).

^hChemaxon via http://www.chemspider.com/Chemical-Structure.82781.html (accessed August 15, 2018).

ⁱ(Carreira et al., 1994; Hilal et al., 1995).

Fig. 2.

Log K_OWs vs. log K_BWs measured in this study. Regression coefficients (± standard error).

In Figure 3, our K_BW (Table 1; Supplementary Data: Table S7) are plotted with literature values (shown in Figure 1) and supports/continues the linear trend previously depicted in Figure 1. The regression equation is:

$$\log K_{\mathit{OW}}=1.5020(\pm 0.0221)\log K_{\mathit{BW}}-0.5327(\pm 0.0448)r^2=0.964n=173$$ (4)

Fig. 3.

All Log K_OWs vs. log K_BWs values. Regression coefficients (± standard error).

The residuals for the regressions are plotted in Supplementary Data: Figure S1-A-B. For the all data regression, there is a slight negative bias in residuals for chemicals with log K_BWs ranging from 2 to 4 and a slight positive bias in residuals for chemicals with log K_BWs above 4 (Supplementary Data: Figure S1-A). There is no thermodynamic basis, we believe, for the relationship to be nonlinear between K_OW and K_BW. The fit of the regression for the data collected within this study (Figure 2) is better in terms of the residuals, i.e., more evenly distributed across the range of K_BWs in the measurements (Supplementary Data: Figure S1-B), and regression is based solely upon neutral hydrophobic organic chemicals. Therefore, the developed regression from this study i.e., Equation 3, will be used for estimating the K_OWs for the 4 selected chemicals within this study, and will be beneficial for estimating additional K_OWs for highly hydrophobic chemicals via their measured log K_BWs.

In Table 1, the quantitative structure-activity relationship (QSAR) estimated log K_OWs are relatively consistent to our Eq. 3 regression K_OW estimations for less hydrophobic chemicals. However, as chemicals become more hydrophobic, difference between Eq. 3 and other QSAR predictions increase greatly. This is noticeable around PCB-155 and beyond for QSAR log K_OW estimations. Our Eq. 3 regression predictions are relatively comparable to the measured literature log K_OWs for more hydrophobic chemicals, thus producing a reliable regression for estimating log K_OWs where measuring K_BW is feasible.

To address the fourth objective, we used the K_BW-K_OW regression formed within this study (Figure 2) to predict log K_OWs for highly hydrophobic chemicals in which there are no literature data available. The log K_BWs for tetradecachloro-m-terphenyl (m-TCP), tetradecachloro-p-terphenyl (p-TCP), and octadecachloro-p-quaterphenyl (p-QTCP) were successfully measured (Table 2). To meet the OECD equilibrium criteria of slopes not significantly different from zero with a minimum of four sampling points over time in the slow stir measurements, equilibration times ranging from 2 to 4 weeks were required for these chemicals. Shown in Table 2, log K_OWs estimates using several different QSAR algorithms are provided ranging between 6–19.

The predicted log K_OWs using equation 3 from our collected K_BW data for the 4 highly hydrophobic chemicals are substainally lower than most of the K_OWs forecasted by the selected QSAR models (Table 2). As an example, EPISuite, SPARC, and ACD prediction models are estimated up to 10 orders of magnitude higher compared to this study’s log K_OW predictions for BDEthane-209, m-TCP, p-TCP, and p-QTCP. There is no clear reason for this, though, EPISuite, SPARC, and ACD algorithms use molecular fragmentation methods that are calibrated based on empirical datasets (Hilal and Karickhoff, 2004; Cabalo and Knox, 2014; Stenzel et al., 2014) and substantial extrapolation are needed due to the lack of measured data for highly hydrophobic chemicals. How the experimental data is assigned to a set of sub-structural fragments differ in the extent of correction factors, use of secondary regressions, and applied intermolecular interactions parameters. The majority of QSAR K_OW estimations are primarily dependent on limited empirical parameters and the quality of empirical data available. Therefore, it becomes difficult to pinpoint what chemical parameters may cause log K_OW over-predictions aside from the shortage of empirical data for highly hydrophobic chemicals. Other factors within QSAR models may include but are not limited to overbearing correction factors stimulated by compound symmetry (Cabalo and Knox, 2014), absence of quantum mechanic contributions such as molecular distortion/orbital hybridization, steric effects, etc., or molecular thermodynamic assumptions between similar chemical classes and sub-stuctures (Maggiora, 2006; Scior et al., 2009). OPERA log K_OW predictions are relatively low for the 4 selected hydrophobic chemicals (Table 2), and the OPERA model estimates for log K_OWs are based on a reviewed dataset of over 14,000 chemicals (Mansouri et al., 2018). Similarly, EPISuite predictions also implement a similar number of reviewed chemicals (e.g. 13,700) (US-EPA, 2012). Consequently, the estimations for highly hydrophobic chemicals lack experimental validation and therefore true accuracy is indefinite.

Conclusions

We have successfully expanded the K_BW-K_OW correlation shown in Figure 1 over a wider range of organic chemicals including several having log K_OWs greater than 9. It was proven that K_BWs are measurable for highly hydrophobic chemicals using the OECD slow stir technique for chemicals with estimated log K_OWs greater than 10. With the developed linear correlation between K_BW and K_OW, measured K_BWs can be used to estimate a chemical’s K_OW and provide a more accurate estimate of K_OW for chemicals. To illustrate the importance of having an unbiased estimate for a highly hydrophobic chemical, consider the K_OW estimates ranging from 8.48 to 15.0 for p-TCP (Table 2). Using BCFBAF algorithm of EPISuite (US-EPA, 2012), log BAFs (bioaccumulation factor) for upper trophic level fish predicted for these K_OW estimates range from 6.82 to 0.72 (L/kg (ww)), respectively. Assuming an environmentally low concentration of p-TCP in water of 1 pg/L (part per quadrillion), predicted residues in fish would range from 6,607,000 to 5.24 pg/kg (ww). This example demonstrates the importance of using accurate K_OW measurements with predictive models for assessing a chemical’s environmental behavior and bioaccumulation.

To estimate log K_OWs for highly hydrophobic chemicals, we recommend that one simply use the developed regression given in Eq. 3 with a K_BW measured using the OECD slow stir methodology; i.e., the methodology used within this study. It is highly advised to measure K_BWs for one or more less hydrophobic chemicals with known log K_OWs for method validation. Additionally, predictions of log K_OWs above 10 are extrapolations, and their accuracy are a bit tenuous without further experimental measurements. At this time, we have no basis for determining biases with Eq. 3, if any, related to differences in chemical classes. Additional K_BW measurements and K_OW measurements for additional highly hydrophobic chemicals are needed to address this question, to span Eq. 3 to higher K_OWs, and to further validate the K_OW–K_BW relationship.

HIGHLIGHTS.

• n-Butanol/water partition coefficients are measurable for hydrophobic chemicals

• Plot of log K_OW vs log K_BW is linear, r²=0.98, for K_OWs extending into 10⁹ range

• K_OWs were estimated using K_OW-K_BW relationship and measured K_BWs

• K_OWs estimated for highly hydrophobic chemicals are lower than their QSAR estimates