You have free access to this content
Hazard characterization of 3-MCPD using benchmark dose modeling: Factors influencing the outcome
Article first published online: 17 OCT 2012
Copyright © 2012 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
European Journal of Lipid Science and Technology
Special Issue: Euro Fed Lipid Highlights 2012
Volume 114, Issue 10, pages 1225–1226, October 2012
How to Cite
Abraham, K., Mielke, H. and Lampen, A. (2012), Hazard characterization of 3-MCPD using benchmark dose modeling: Factors influencing the outcome. Eur. J. Lipid Sci. Technol., 114: 1225–1226. doi: 10.1002/ejlt.201200250
- Issue published online: 17 OCT 2012
- Article first published online: 17 OCT 2012
- Manuscript Received: 27 AUG 2012
- Manuscript Accepted: 27 AUG 2012
- 3-monochloropropane-1,2-diol (3-MCPD);
- Benchmark dose modeling;
- Hazard characterization;
The article by Rietjens et al. in this issue of European Journal of Lipid Science and Technology [p. 1140–1147] deals with benchmark dose (BMD) calculations using data of the two long-term studies available for 3-monochloropropane-1,2-diol (3-MCPD) in rats in order to calculate benchmark dose lower-bound confidence limit 10% (BMDL10) values as point of departure for the most sensitive endpoint (tubular hyperplasia). Experimental data 1, 2, software (PROAST, US-EPA) and instructions how to use both 3 are available for everyone. The BMD approach is relatively new in toxicology and exhibits some advantages over the conventional approach as the BMDL10 is less dependent on study design than the NOAEL (No Observed Adverse Effect Level) and incorporates information from every single animal, not only from one dose group. However, calculated BMDL10 values very much depend on the user's choices (data selection, model assumptions and interpretation) and may therefore be influenced by a possible interest to get a higher or lower tolerable daily intake (TDI). This is explained in the following commentary.
Experimental data is available for male and female Fischer 344 rats (Sunahara et al. 1, n = 50 each) and for male and female Sprague-Dawley rats (Cho et al. 2, n = 50 each). Obviously, female Sprague-Dawley rats are much less sensitive for tubular hyperplasia. They were excluded, while the remaining three groups were pooled for the BMD calculations. But are these three groups comparable in sensitivity for the critical effect? Rietjens et al. argue that a test (likelihood ratio test) found no significant difference in dose-response. However, such a comparison of different dose-response curves is tricky. How big must a difference be at which part of the curve to provide a statistically different result? The BMDL10 especially depends on the slope at low doses as well as the proportion of animals showing the effect in the control group. A view on the data of Sunahara et al. of males (control group: 3/50, lowest dose of 1.10 mg/kg bw/day: 6/50) and females (control group: 2/50, lowest dose of 1.40 mg/kg bw/day: 4/50) compared to the data of Cho et al. (control group: 1/50, lowest dose of 1.97 mg/kg bw/day: 11/50) already suspects that the latter group (male Sprague-Dawley rats) is more sensitive than the Fischer344 rats. This is confirmed by BMD analysis showing lowest BMDL10 values (all with the LogProbit model) for the animals of Sunahara et al. to be distinctly higher (males: 0.72 mg/kg bw/day, females: 0.66 mg/kg bw/day) than the one for the male animals of Cho et al. (0.27 mg/kg bw/day). Therefore, the new data of Cho et al. provide evidence that male Sprague-Dawley rats are even more sensitive for tubular hyperplasia than Fischer344 rats; accordingly, pooling of the three groups is not a conservative approach.
Regarding the relevant choices for the BMD calculations, the most important are: the benchmark response (commonly accepted: 10% for quantal data, 5% for continuous data), the models used including their constraints, the model fit accepted (usually p>0.05), as well as the selection of BMDL10 results as point of departure (lowest value?, best fit?, model averaging?). Depending on the choices, calculated BMDL10 values are in broad span. In the case of 3-MCPD, the lowest BMDL10 value would be 0.07 mg/kg bw/day, using data of the male Sprague-Dawley rats and the gamma model without constraints (not in line with European Food Safety Authority (EFSA) 3, but accepted by Rietjens et al. pointing to a paper in preparation). Hwang et al. 4 calculated a value of 0.87 mg/kg bw/day, using the same data set, but choosing the result with the best fit (logistic model; no calculations made with the logprobit model, providing a distinctly lower value of 0.27 mg/kg bw/day with an only slightly lower goodness of fit). Rietjens et al. provide a value of 0.72 mg/kg bw/day, using pooled data of the three subgroups and averaging of 7 model results which, however, include 3 (low) model results (for log-logistic, Weibull, and gamma) obtained without restriction (not in line with EFSA 2009, see below); accordingly, the use of the constraints recommended would provide a result distinctly above 1 mg/kg bw/day. Therefore, results of these BMD calculations vary by a factor of more than 10.
All the choices can be justified by plausible arguments but may as well be influenced by interests regarding the outcome. Therefore, it makes sense to use guidelines how to use BMD calculations in risk assessment. For food safety issues, EFSA provided such a guideline in 2009 3, but Rietjens et al. did not follow it although they conveyed this impression throughout the paper. They calculated low BMD/BMDL10 values for the models log-logistic, Weibull, and gamma (Table 3), as they did not use the constraints recommended (c>1) to avoid these models having undesirable properties. For example, the lowest BMDL10 was calculated using the gamma model (0.446 mg/kg bw/day); with the constraints mentioned, the value would be 2.24 mg/kg bw/day. Furthermore, Rietjens et al. performed model averaging (instead of using the lowest BMDL10 as point of departure), referencing to EFSA 3: “…. it should be noted that the EFSA opinion recommended the lowest BMDL10 value as the overall BMDL10, or, if possible, apply model averaging to obtain a less conservative BMDL”. However, this is not the case as EFSA [3, page 32] states: “Selecting the lowest BMDL amongst those for the same endpoint tends to be conservative, but it is recommended to use this until more advanced methods, such as ‘model averaging’ (Wheeler and Bailer 2007, 2008) have been fully developed and validated.”
Taking all together, the derivation of Rietjens et al. looks arbitrary, being conservative in one issue (no model constraints) and not conservative in two other issues (pooling of the three subgroups, model averaging).
With increasing knowledge and experience, the EFSA opinion may be changed in the time to come, but use of the current opinion 3 and the data of the most sensitive animals (male Sprague-Dawley rats) provides a lowest BMDL10 value of 0.27 mg/kg bw/day (log-probit model). Accordingly, the TDI would be 2.7 µg/kg bw/day (using a default uncertainty factor of 100), and therefore practically identical with the current one (2 µg/kg bw/day) derived by the Scientific Committee on Food (SCF) and the Joint FAO/WHO Expert Committee on Food Additives (JECFA) in 2001 using the data of Sunahara et al. 1 and a classic approach with an additional factor of 5 considering that only data on a lowest observed adverse effect level (LOAEL) was available.
The authors have declared no conflict of interest.
- 3European Food Safety Authority (EFSA), Scientific Opinion: Use of the benchmark dose approach in risk assessment. Guidance of the Scientific Committee. EFSA J. 2009, 1150, 1– 72. Available at: http://www.efsa.europa.eu/en/scdocs/doc/1150.pdf