Registration Dossier

Environmental fate & pathways

Bioaccumulation: aquatic / sediment

Currently viewing:

Administrative data

Link to relevant study record(s)

Reference
Endpoint:
bioaccumulation in aquatic species: fish
Type of information:
(Q)SAR
Adequacy of study:
key study
Reliability:
2 (reliable with restrictions)
Rationale for reliability incl. deficiencies:
results derived from a valid (Q)SAR model and falling into its applicability domain, with adequate and reliable documentation / justification
Justification for type of information:
The software program BCFBAFTMis part of the Estimations Programs Interface for Windows (EPI SuiteTM). It is a Windows®-based suite of physical/chemical property and environmental fate estimation programs developed by the US Environmental Protection Agency (EPA) and Syracuse Research Corp. (SRC). The estimation methods in EPI Suite™have been developed by government, academic, and private sector researchers over many years and represent some of the best techniques currently available.

The intended application domain of EPI Suite is organic chemicals, and inorganic as well as organometallic chemicals generally can be considered outside the domain. Data files are available containing the experimental data sets used to derive and validate program methodologies or test program accuracy.

The BCFBAF Program is an update and expansion of the previous BCFWIN Program that was part of the EPI Suite version 3.20. The update pertains to estimation of Bioconcentration Factor (BCF). The BCFBAF program estimates BCF of an organic compound using the compound's log octanol-water partition coefficient (Kow). For the update, a more recent and better evaluated database of BCF values was used for both training and validation. The BCF data were re-regressed using the same methodology as in the original BCFWIN program..

The US EPA is using this predictive model for assessing chemicals under the Toxic Substance Control Act (TSCA). The tool is also accepted by ECHA and explicitly mentioned in the “Guidance on information requirements and chemical safety assessment, Chapter R.6: QSARs and grouping of chemicals”.

7.1. Bioconcentration Factor (BCF)


7.1.1. Estimation Methodology

The original estimation methodology used by the original BCFWIN program is described in a document prepared for the U.S. Environmental Protection Agency (Meylan et al., 1997). The estimation methodology was then published in journal article (Meylan et al, 1999).

The BCFBAF Program updates the BCF estimation methodology of the BCFWIN program by using an updated and better evaluated BCF database for selecting training and validation datasets. The exact same regression methodology used to derive the original BCFWIN method was used to derive the BCFBAF method for estimating BCF.

Experimental BCF Data

The measured BCF values used in the revised regressions were selected from a quality reviewed BCF database (Arnot and Gobas, 2006); details of the data quality review methods are described in Arnot and Gobas (2006). Single BCF values were selected for each compound (median values were generally selected for compounds with multiple values).

The BCF values selected for the BCFBAF training and validation datasets are available in Appendix G and via Internet download at:
http://esc.syrres.com/interkow/EpiSuiteData.htm ... A substructure searchable version of the data can be downloaded at: http://esc.syrres.com/interkow/EpiSuiteData_ISIS_SDF.htm


Estimation Methodology

The following is a brief summary of the estimation methodology:
The BCFBAF method classifies a compound as either ionic or non-ionic. Ionic compounds include carboxylic acids, sulfonic acids and salts of sulfonic acids, and charged nitrogen compounds (nitrogen with a +5 valence such as quaternary ammonium compounds). All other compounds are classified as non-ionic.

Training Dataset Included:
466 Non-Ionic Compounds
61 Ionic Compounds (carboxylic acids, sulfonic acids, quats)

Methodology for Non-Ionic was to separate compounds into three divisions by Log Kow value as follows:
Log Kow < 1.0
Log Kow 1.0 to 7.0
Log Kow > 7.0

The following graph of the raw data illustrates the divisions and the comparison of the new BCFBAF regression lines to the previous BCFWIN regression lines:


For each division, a "best-fit" straight line was derived by common statistical regression methodology. The graph does not adjust individual data points with correction factors derived for BCFBAF. The regression methodology includes derivation of correction factors based on specific structural features. Appendix E lists all correction factors used by BCFBAF (with a comparison to BCFWIN). Non-ionic compounds are predicted by the following relationships:

For Log Kow 1.0 to 7.0 the derived QSAR estimation equation is:

Log BCF = 0.6598 Log Kow - 0.333 + Σ correction factors
(n = 396, r2 = 0.792, Q2 = 0.78, std dev = 0.511, avg dev = 0.395)

The previous BCFWIN equation:
Log BCF = 0.77 Log Kow - 0.70 + Σ correction factors


For Log Kow > 7.0 the derived QSAR estimation equation is:

Log BCF = -0.49 Log Kow + 7.554 + Σ correction factors
(n = 35, r2 = 0.634, Q2 = 0.57, std dev = 0.538, avg dev = 0.396)

The previous BCFWIN equation:
Log BCF = -1.37 Log Kow + 14.4 + Σ correction factors

Certain super-hydrophobic chemicals (Log Kow >7.0) selected from the empirical database had reported BCF values with measured water concentrations that exceed water solubility limits. These BCF values were corrected based on estimates of water solubility limits (Arnot and Gobas, 2006).

For Log Kow < 1.0 the derived QSAR estimation equation is: All compounds with a log Kow of less than 1.0 are assigned an estimated log BCF of 0.50 (same as in BCFWIN).


Ionic compounds are predicted as follows:
log BCF = 0.50 (log Kow < 5.0)
log BCF = 1.00 (log Kow 5.0 to 6.0)
log BCF = 1.75 (log Kow 6.0 to 8.0)
log BCF = 1.00 (log Kow 8.0 to 9.0)
log BCF = 0.50 (log Kow > 9.0)

The graph of Ionic Compounds versus Log Kow:



Metals (tin and mercury), long chain alkyls and aromatic azo compounds require special treatment.


7.1.2. Estimation Accuracy

Accuracy of the Training Set:



Error Histogram for the Training Set:



Accuracy of the Validation Set:



7.1.3. Estimation Domain

Appendix E gives for each correction factor the maximum number of instances of that factor in any of the 527 training set compounds (the minimum number of instances is of course zero, since not all compounds had every correction factor). The minimum and maximum values for molecular weight and logKow are listed below. Currently there is no universally accepted definition of model domain. However, users may wish to consider the possibility that bioconcentration factor estimates are less accurate for compounds outside the MW and logKow ranges of the training set compounds, and/or that have more instances of a given correction factor than the maximum for all training set compounds. It is also possible that a compound may have a functional group(s) or other structural features not represented in the training set, and for which no fragment coefficient was developed; and that a compound has none of the fragments in the model’s fragment library. In the latter case, predictions are based on molecular weight alone. These points should be taken into consideration when interpreting model results.

Training Set (527 Compounds):

Molecular Weight:
Minimum MW: 68.08 (Furan)
Maximum MW: 991.80 Ionic: (2,7-Naphthalenedisulfonic acid, 4-amino-5-hydroxy-3,6-
bis[[4-[[2-(sulfooxy)ethyl]sulfonyl]phenyl]azo]-, tetrasodium salt)
Maximum MW: 959.17 Non-Ionic: (Benzene, 1,1 -oxybis[2,3,4,5,6-pentabromo-)
Average MW: 244.00

Log Kow:
Minimum LogKow: -6.50 Ionic: (2,7-Naphthalenedisulfonic acid, 4-amino-5-hydroxy-3,6-bis[[4-[[2-(sulfooxy)ethyl]sulfonyl]phenyl]azo]-, tetrasodium salt)
Minimum LogKow: -1.37 Non-Ionic: (1,3,5-Triazine-2,4,6-triamine)
Maximum LogKow: 11.26 (Benzenamine, ar-octyl-N-(octylphenyl)-)

7.2. Biotransformation Rate in Fish (kM)


7.2.1. Estimation Methodology

The whole body primary biotransformation rate constant (kM) model for fish

The model estimates screening level whole body primary biotransformation half-lives (HL; day) and rate constants (kM; /day) for discrete organic chemicals in fish. An evaluated database of kM estimates in fish (Arnot et al., 2008a) was used to develop and evaluate the model. The predictions are calculated based on multiple linear regressions of development data set (n=421) against counts of 57 molecular substructures, the octanol-water partition coefficient and molar mass. The coefficient of determination (r2) for the development set is 0.82, the cross validation (leave-one-out coefficient of determination, q2) is 0.75, and the mean absolute error (MAE) is 0.38 log units (a factor of 2.4). Results for the external validation of the model using an independent test set (n=211) are r2 = 0.73; MAE = 0.45 log units (factor of 2.8).

The model predicts half-life (HLN) and rate constant (kM,N) values “normalized” for a 10 g fish at 15ºC. For comparisons with other estimated values (in vitro, in vivo) and for use in mass balance models it is recommended that the model kM,N predictions be converted to mass and temperature specific kM,X values as

kM,X = kM,N (WX/WN)-0.25 exp(0.01(TX – TN))

where WX is the study-specific mass of the organism (kg), WN is the normalized mass of the organism (0.01 kg), TX is the study-specific water temperature (ºC), and TN is the normalized water temperature (15ºC). The model provides kM,X values (/day) for 0.1 kg, 1 kg and 10 kg fish.

For molecules with fragments that appear to be readily biotransformed (e.g., see regression coefficients for esters, ureas, etc), the model may predict extremely short HLN values (or very fast kM,N values). When the model predicts values that exceed proposed whole body maximum rate constant values (Arnot et al., 2008a), the whole body maximum values are provided and recommended to replace the original model predictions.

A full description of assumptions and limitations of the mass balance method used to develop the kM database is available (Arnot et al., 2008b). Briefly, biotransformation is defined as the change of the parent substance to another molecule or a conjugated form of the parent substance. The model calculates kM as a whole body value, namely the fraction of the mass in the whole body biotransformed per unit of time. The model does not provide predictions for the formation of specific biotransformation products (some of which may be more toxic than the parent compound), nor does it identify specific pathways for the biotransformation process (Phase I oxidations or reductions or Phase II conjugations). If formed metabolites are known, these can be re-introduced to the model as a distinct substance for novel predictions. The model assumes first-order processes and cannot estimate biotransformation rates that may occur under non-first order conditions (e.g., enzyme saturation).

Arnot JA, Mackay D, Parkerton TF, Bonnell M. 2008a. A database of fish biotransformation rates for organic chemicals. Environmental Toxicology and Chemistry 27(11): in press.

Arnot JA, Mackay D, Bonnell M. 2008b. Estimating metabolic biotransformation rates in fish from laboratory data. Environmental Toxicology and Chemistry 27: 341-351.


BCFBAF Model Regression - Derivation for kM

A dataset of 632 experimental kM biotransformation rates in fish (compiled in units of log biotransformation half-lives in days) was divided into a training set of 421 compounds for model derivation and validation set of 211 compounds for model testing. The compounds used for model training and validation are listed in Appendix I and Appendix J, respectively.

Initially, each individual compound in the training set was divided into structural fragments based on the same fragments used by the BIOWIN Program (biodegradation probability) and BioHCwin Program (biodegradation of hydrocarbons). Fragments not occurring in any training set compound were excluded from the model derivation. Initial regressions were used to identify fragments having no statistical significance (with coefficient values having little or no effect on results), and these fragments were excluded from the final regression. Several new fragments were added based on structural similarities of the training set compounds.

The final multiple-linear regression was performed on a matrix containing the number of occurrences of each fragment in each compound plus the logKow and molecular weight of each compound. The solution column of the matrix was the experimental log biotransformation half-life of each compound in days. Appendix F lists the individual fragments. The multiple-linear regression was performed with CoStatTM statistical software (CoHort, 2008).

The final multiple-linear regression-derived equation (which is used by the BCFBAF program to estimate the kM Biotransformation Half-Life) is:

Log kM/Half-Life (in days) = 0.30734215*LogKow - 0.0025643319*MolWt - 1.53706847 + Σ(Fi*ni)

where LogKow is the log octanol-water partition coefficient, MolWt is the Molecular Weight, and Σ(Fi*ni) is the summation of the individual Fragment coefficient values (Fi) as listed in Appendix F times the number of times the individual fragment occurs in the structure ( ni). The -1.53706847 is the equation constant.

A complete description of the methodology has been submitted for publication.


7.2.2. Estimation Accuracy

Method Training Accuracy

The following graph illustrates the estimation accuracy of the training set:



The following histogram illustrates the estimation error of the training set:



Method Validation Accuracy

The following graph illustrates the estimation accuracy of the validation set:





7.2.3. Estimation Domain

Appendix F gives for each fragment the maximum number of instances of that fragment in any of the 421 training set compounds (the minimum number of instances is of course zero, since not all compounds had every fragment). The minimum and maximum values for molecular weight and logKow are listed below. Currently there is no universally accepted definition of model domain. However, users may wish to consider the possibility that biotransformation estimates are less accurate for compounds outside the MW and logKow ranges of the training set compounds, and/or that have more instances of a given fragment than the maximum for all training set compounds. It is also possible that a compound may have a functional group(s) or other structural features not represented in the training set, and for which no fragment coefficient was developed; and that a compound has none of the fragments in the model’s fragment library. These points should be taken into consideration when interpreting model results.

Training Set (421 Compounds):

Molecular Weight:
Minimum MW: 68.08 (Furan)
Maximum MW: 959.17 (Decabromodiphenyl ether)
Average MW: 259.75

Log Kow:
Minimum LogKow: 0.31 (Benzenesulfonamide)
Maximum LogKow: 8.70 (Decabromodiphenyl ether)

Uncertainty in the model predictions must be considered for model applications. The median confidence (uncertainty) factor for the database used to develop the model is about 5.5. A confidence factor of 5.5 suggests that 95% of the expected values for a biotransformation rate constant fall between 5.5 x kM and kM/5.5 assuming a log normal distribution. This degree of uncertainty corresponds to approximately 1.5 orders of magnitude of variance in the distribution. The log MAE from the test set corresponds to a confidence factor of about 7 (~1.7 orders of magnitude variance in the distribution) and could also provide screening level guidance for the expected range of values for application of HLN and kM,N estimates. This level of uncertainty (1.5 – 1.7 orders of magnitude) is also generally consistent with present estimates for intra- and inter-species and route of exposure variability (Arnot et al., 2008a).

The model contains a large set of unique structural fragments so that it can be broadly applicable to diverse chemical structures; however, these fragments do not reflect the entire domain of possible structural fragments for organic chemicals. The model is not expected to provide accurate results for all chemicals in all fish species and it is difficult to define precisely the domain of applicability. The model may not successfully predict biotransformation rates for substances that have molecular components that significantly affect biotransformation processes and were not included in model development. The database used to develop the model did not include many substances that appreciably ionize at physiological pH or larger molecules (molar mass >600); therefore, the model may not accurately predict values for such substances. The data set used to develop the model did not include metals or organometals, pigments or dyes, or perfluorinated substances and the model should not be used for these substances.

7.3. Arnot-Gobas BAF-BCF


7.3.1. Estimation Methodology


The Arnot-Gobas BCF and BAF model (Arnot and Gobas, 2003)

The Arnot-Gobas model estimates steady-state bioconcentration factor (BCF; L/kg) and bioaccumulation factor (BAF; L/kg) values for non-ionic organic chemicals in three general trophic levels of fish (i.e., lower, middle and upper) in temperate environments. The model calculations represent general trophic levels (i.e., not for a particular fish species) and are derived for “representative” environmental conditions (e.g., dissolved and particulate organic carbon content in the water column, water temperature). Thus, it provides general estimates for these conditions in absence of site-specific measurements or estimates. The default temperature for the BCF and BAF calculations is 10oC (temperate regions); therefore, the model predictions are not recommended for arctic, sub-tropical or tropical regions or for comparisons with other vastly different conditions (e.g., laboratory tests at ~25oC). Site-specific food web models, bioaccumulation models and bioconcentration models are available for specific modeling requirements (e.g., http://www.rem.sfu.ca/toxicology/models/models.htm , http://www.trentu.ca/cemc).

The model includes mechanistic processes for bioconcentration and bioaccumulation such as chemical uptake from the water at the gill surface (BCFs and BAFs) and the diet (BAFs only), and chemical elimination at the gill surface, fecal egestion, growth dilution and metabolic biotransformation (Arnot and Gobas 2003). Other processes included in the calculations are bioavailability in the water column (only the freely dissolved fraction can bioconcentrate) and absorption efficiencies at the gill and in the gastrointestinal tract. The model requires the octanol-water partition coefficient (KOW) of the chemical and the normalized whole-body metabolic biotransformation rate constant (kM, N; /day) as input parameters to predict BCF and BAF values. The required kM, N value must be normalized to a fish of 10 g (Arnot et al., 2008). Model predictions may be highly uncertain for chemicals that have estimated log KOW values > 9. The model is not recommended at this time for chemicals that appreciably ionize, for pigments and dyes, or for perfluorinated substances.

The BAF calculations were derived from the parameterization and calibration of the model to a large database of measured BAF values from the Great Lakes (Lake Ontario, Lake Erie and Lake St. Clair). The measured BAFs are for chemicals that are poorly metabolized (e.g., PCBs) and were generally grouped into lower, middle and upper trophic levels of fish species. The overall food web biomagnification factors (β) in the BAF model are calibrated to each trophic level of measured BAF values (Arnot and Gobas, 2003). Therefore, in the absence of metabolic biotransformation the BAF model predictions are in general agreement with measured BAFs in fish of these general trophic positions from the Great Lakes for chemicals that are poorly metabolized.

7.3.2 Other considerations in using Arnot-Gobas BAF-BCF

The model may not adequately capture biotransformation at the first trophic level for chemicals that are readily biotransformed in invertebrates and plankton. Unfortunately, there is not available a kM prediction model for invertebrates and plankton that adequately captures possible biotransformation in these lower trophic-level positions; i.e. diet for low- and mid-trophic level fish. Therefore, the model may be somewhat conservative in that it assumes negligible biotransformation for invertebrates and plankton. However, the model does give users a sense of the range of BAFs that might be observed in the environment, and this is useful because an uncertainty analysis is not included for kM, Kow, trophic position etc. Also, an underlying issue often is whether biomagnification or biodilution is more likely for a given chemical, and including three trophic levels allows at least some insight to be gained.

In the environment there are multiple trophic levels of fish and other aquatic organisms, and both biomagnification and trophic dilution are observed. When chemicals biomagnify the highest BAFs are at higher trophic levels; conversely when chemicals are readily biotransformed, the highest BAFs are at lower trophic levels. The model appears to capture these phenomena.

The Arnot-Gobas model assumes default lipid contents of 10.7%, 6.85% and 5.98% for the upper, middle and lower trophic levels, respectively, as given in Appendix K. Since the lab studies from which most data in the measured BCF database were derived typically used fish with 3-5% lipid content, this may help explain why the regression-based BCF model typically yields estimated BCF values lower than from the Arnot-Gobas model. A reasonable way to compare BCF and BAF values across measured and estimated data is to convert them to 100% lipid basis. This can be done if the % lipid is known or can be estimated, by dividing the wet weight BCF (units of L/kg) by the % lipid expressed as a fraction. For example, a BCF of 5,000 based on wet weight for a fish with 10% lipid is 5,000 L/kg divided by 0.1 = 50,000 L/kg lipid weight.

Arnot JA, Gobas FAPC. 2003. A generic QSAR for assessing the bioaccumulation potential of organic chemicals in aquatic food webs. QSAR and Combinatorial Science 22: 337-345.

Arnot JA, Mackay D, Parkerton TF, Bonnell M. 2008. A database of fish biotransformation rates for organic chemicals. Environmental Toxicology and Chemistry 27(11): 2263-2270.


Equations Used To Estimate BAF

Appendix K contains the program code (and equations) used by the BCFBAF program to estimate the Arnot-Gobas BAF and BCF values.




Qualifier:
equivalent or similar to
Guideline:
other: QSAR using EPI Suite (BCFBAF v3.01 (Sept 2012)
Version / remarks:
Estimation Program EPI Suite (US EPA)
Type:
BCF
Value:
276.2 L/kg
Validity criteria fulfilled:
yes
Conclusions:
BCF was calculated using the US EPA Estimation program BCFBAF v.3.01 (US EPA)
Log BCF Arnot-Gobas method (upper trophic) = 2.441 (BCF = 276.2)
Executive summary:

The software program BCFBAFTMis part of the Estimations Programs Interface for Windows (EPI SuiteTM). It is a Windows®-based suite of physical/chemical property and environmental fate estimation programs developed by the US Environmental Protection Agency (EPA) and Syracuse Research Corp. (SRC). The estimation methods in EPI Suitehave been developed by government, academic, and private sector researchers over many years and represent some of the best techniques currently available.

The intended application domain of EPI Suite is organic chemicals, and inorganic as well as organometallic chemicals generally can be considered outside the domain. Data files are available containing the experimental data sets used to derive and validate program methodologies or test program accuracy.

The BCFBAF Program is an update and expansion of the previous BCFWIN Program that was part of the EPI Suite version 3.20.  The update pertains to estimation of Bioconcentration Factor (BCF).  The BCFBAF program estimates BCF of an organic compound using the compound's log octanol-water partition coefficient (Kow).  For the update, a more recent and better evaluated database of BCF values was used for both training and validation.  The BCF data were re-regressed using the same methodology as in the original BCFWIN program..

The US EPA is using this predictive model for assessing chemicals under the Toxic Substance Control Act (TSCA). The tool is also accepted by ECHA and explicitly mentioned in the “Guidance on information requirements and chemical safety assessment, Chapter R.6: QSARs and grouping of chemicals”.

BCF was calculated using the US EPA Estimation program BCFBAF v.3.01 (US EPA)

Bioaccumulation Estimates (BCFBAF v3.01):

Log BCF from regression-based method = 2.273 (BCF = 187.4 L/kg wet-wt)

Log Biotransformation Half-life (HL) = -0.0954 days (HL = 0.8028 days)

Log BCF Arnot-Gobas method (upper trophic) = 2.441 (BCF = 276.2)

Log BAF Arnot-Gobas method (upper trophic) = 2.457 (BAF = 286.3)

Description of key information

BCF was calculated using the US EPA Estimation program BCFBAF v.3.01 (US EPA)

Bioaccumulation Estimates (BCFBAF v3.01):

Log BCF from regression-based method = 2.273 (BCF = 187.4 L/kg wet-wt)

Log Biotransformation Half-life (HL) = -0.0954 days (HL = 0.8028 days)

Log BCF Arnot-Gobas method (upper trophic) = 2.441 (BCF = 276.2)

Log BAF Arnot-Gobas method (upper trophic) = 2.457 (BAF = 286.3)

Key value for chemical safety assessment

BCF (aquatic species):
276.2 L/kg ww

Additional information

The software program BCFBAFTMis part of the Estimations Programs Interface for Windows (EPI SuiteTM). It is a Windows®-based suite of physical/chemical property and environmental fate estimation programs developed by the US Environmental Protection Agency (EPA) and Syracuse Research Corp. (SRC). The estimation methods in EPI Suitehave been developed by government, academic, and private sector researchers over many years and represent some of the best techniques currently available.

The intended application domain of EPI Suite is organic chemicals, and inorganic as well as organometallic chemicals generally can be considered outside the domain. Data files are available containing the experimental data sets used to derive and validate program methodologies or test program accuracy.

The BCFBAF Program is an update and expansion of the previous BCFWIN Program that was part of the EPI Suite version 3.20.  The update pertains to estimation of Bioconcentration Factor (BCF).  The BCFBAF program estimates BCF of an organic compound using the compound's log octanol-water partition coefficient (Kow).  For the update, a more recent and better evaluated database of BCF values was used for both training and validation.  The BCF data were re-regressed using the same methodology as in the original BCFWIN program..

The US EPA is using this predictive model for assessing chemicals under the Toxic Substance Control Act (TSCA). The tool is also accepted by ECHA and explicitly mentioned in the “Guidance on information requirements and chemical safety assessment, Chapter R.6: QSARs and grouping of chemicals”.