Dose-Response Modeling and Benchmark Calculations from Spontaneous Behavior Data on Mice Neonatally Exposed to 2,2',4,4',5-Pentabromodiphenyl Ether

Salomon Sand*,1, Dietrich von Rosen{dagger}, Per Eriksson{ddagger}, Anders Fredriksson{ddagger}, Henrik Viberg{ddagger}, Katarina Victorin* and Agneta Falk Filipsson*

* Institute of Environmental Medicine, Karolinska Institutet, P.O. Box 210, 17177 Stockholm, Sweden; {dagger} Department of Biometry and Informatics, Swedish University of Agricultural Sciences, P.O. Box 7032, 750 07 Uppsala, Sweden; and {ddagger} Department of Environmental Toxicology, Uppsala University, Norbyvägen 18A, S-75236 Uppsala, Sweden

Received April 27, 2004; accepted July 7, 2004


    ABSTRACT
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIAL AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
In this paper the benchmark dose (BMD) method was introduced for spontaneous behavior data observed in 2-, 5-, and 8-month-old male and female C57Bl mice exposed orally on postnatal day 10 to different doses of 2,2',4,4',5-pentabromodiphenyl ether (PBDE 99). Spontaneous behavior (locomotion, rearing, and total activity) was in the present work quantified in terms of a fractional response defined as the cumulative response after 20 min divided by the cumulative response produced over the whole 1-h test period. The fractional response contains information about the time-response profile (which differs between the treatment groups) and has appropriate statistical characteristics. In the analysis, male and female mice could be characterized by a common dose-response model (i.e., they responded equally to the exposure to PBDE 99). As a primary approach, the BMD was defined as the dose producing a 5 or 10% change in the mean fractional response. According to the Hill model, considering a 10% change the lower bound of the BMD for rearing, locomotion, and total activity was 1.2, 0.85, and 0.31 mg PBDE 99/kg body weight, respectively. A probability-based procedure for BMD modeling was also considered. Using this methodology, the BMD was defined as corresponding to an excess risk of 5 or 10% of falling below cutoff points representing adverse levels of fractional response.

Key Words: spontaneous behavior; dose-response; benchmark dose; health risk assessment.


    INTRODUCTION
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIAL AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Polybrominated diphenyl ethers (PBDEs) belong to a group of environmental contaminants known as brominated flame retardants, which, for example, can be found in building materials, electrical devices, and textiles. PBDEs are persistent compounds that are found in the environment and in human mother's milk (Meironyté et al., 1999Go). Animal experiments have been undertaken to assess the risk for neurotoxicological effects associated with exposure to these compounds during a critical phase of neonatal brain development, referred to as the "brain growth spurt" (Davison and Dobbing, 1968Go). Induction of disturbances in spontaneous behavior after neonatal exposure to 2,2',4,4'-tetrabromodiphenyl ether (PBDE 47), 2,2',4,4',5-pentabromodiphenyl ether (PBDE 99), 2,2',4,4',5,5'-hexabromodiphenyl ether (PBDE 153), and 2,2',3,3',4,4',5,5',6,6'-decabromodiphenyl ether (PBDE 209) has previously been observed in male NMRI mice (Eriksson et al., 2001Go, 2002Go; Viberg et al., 2002Go, 2003aGo, 2003bGo). Similar developmental defects have also been reported after neonatal exposure to polychlorinated biphenyls (Eriksson, 1998Go).

The use of dose-response models in health risk assessment of chemicals has received more attention over the recent years. Due to various limitations associated with the traditional procedure applied in this field of risk assessment (i.e., the NOAEL [no-observed-adverse-effect-level] approach), the benchmark dose method has been suggested as an alternative (Crump, 1984Go; Falk Filipsson et al., 2003Go; USEPA, 1995Go). According to the benchmark dose methodology, a dose-response model is fitted to data, and the model is used for estimating a dose, the benchmark dose (BMD), that corresponds to a predetermined change in response. Using this approach, an approximate lower confidence bound of the benchmark dose is suggested to replace the NOAEL as a point of departure in the determination of acceptable daily intakes (ADIs), or reference values, for environmental toxicants.

Spontaneous behavior data are continuous in nature. For continuous endpoints, different procedures for benchmark dose calculations have been discussed (Barnes et al., 1995Go; Crump, 2002Go; Falk Filipsson et al., 2003Go; Sand et al., 2003Go; Slob, 2002Go). According to one of the procedures the benchmark dose (BMD) is defined as corresponding to a percentage change in the mean response (Edler et al., 2002Go; Slob and Pieters, 1998Go). In this case, the BMD depends on the dose-response model for the mean response, and the variation in response between individual animals is reflected in the lower confidence bound of the BMD. A probability-based approach to modeling continuous data has also been suggested (Crump, 1995Go; Gaylor and Slikker, 1990Go; Kodell and West, 1993Go). According to this methodology a cutoff value is determined, below (or above) which continuous responses are considered adverse (or extreme). The benchmark dose is herein defined as the dose level that corresponds to a certain increase in the probability, or risk, of falling below the cutoff value compared to background. Using the latter approach, the dose-response model for the mean response, as well as the variance, jointly influence the estimation of the benchmark dose (Sand et al., 2003Go). Neurotoxicological effects have been used as a basis in previous illustrations and applications of the probability-based approach to modeling continuous endpoints, including experimental data (Gaylor and Slikker, 1990Go, 1994Go; Kodell et al., 1995Go; Slikker et al., 1998Go) as well as epidemiological data (Budtz-Jorgensen et al., 2000Go; Crump et al., 1998Go, 2000Go; Jacobson et al., 2002Go).

The aim of the present article is to illustrate how benchmark dose calculations may be conducted for spontaneous behavior data from animal experiments. Spontaneous behavior can be interesting from a risk assessment point of view, since it may be an early indicator of neurotoxic disturbances (Eriksson, 1997Go, 1998Go). Previous experience of applications of risk assessment methodologies, involving dose-response modeling, to this type of data is limited (Bogdan et al., 2001Go). In the paper, the benchmark dose concept is introduced for spontaneous behavior data observed in 2-, 5-, and 8-month-old male and female C57Bl mice neonatally exposed to PBDE 99. The procedure where the BMD is defined as corresponding to a percentage change in the mean response is considered, as well as a probability-based approach for BMD estimation.


    MATERIAL AND METHODS
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIAL AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Spontaneous behavior test. 2,2',4,4',5-pentabromodiphenyl ether (PBDE 99) was administered orally, via a metal gastric tube, on postnatal day 10 to male and female C57Bl mice at dose levels of 0.4, 0.8, 4, 8, or 16 mg/kg body weight (as a single dose). Control mice received 10 ml/kg body weight of the 20% fat emulsion vehicle. Each of the six different dosage categories (treatment groups) contained three to five litters. Spontaneous behavior was studied at the age of 2, 5, and 8 months. For each testing occasion, ten animals were randomly picked from each treatment group. The animals were tested between 0800 and 1200 h under the same ambient light and temperature as in their housing conditions. Motor activity was measured during a 60-min period, divided into three 20-min time intervals, in an automated device consisting of cages (40 x 25 x 15 cm) placed within two series of infrared beams (low and high levels) (Rat-O-Matic, ADEA Elektronik AB, Uppsala, Sweden) (Fredriksson, 1994Go). The cages were placed in individual soundproofed boxes with separate ventilation. Twelve devices were run simultaneously. Thus, two animals from each treatment group were tested in parallel (2 x 6 = 12), allowing a balance between treatments. Three different measurements of spontaneous behavior were recorded:

In the study there were no signs of systemic toxicity in the PBDE 99-treated mice during the experimental period. More information about the basis for benchmarks dose modeling study used as can be found in Viberg et al., (2004)Go, where the basic behavior material is presented.

Quantification of spontaneous behavior. In the study under investigation, spontaneous behavior (locomotion, rearing, and total activity) was registered in test subjects during a 60-min period. In Figure 1, experimental observations at 2 months are illustrated for locomotion. The cumulative number of counts produced after 20, 40, and 60 min of testing is here denoted by R20, R40, and R60, respectively. As is observed in Figure 1, the normal habituation profile reflected by the controls seems to correspond to a certain reduction in the rate of activity during the experiment. Habituation is also observed at lower doses of PBDE 99. However, at the higher doses any habituating behavior seems not to be apparent. The spontaneous behavior observed over the 60-min period was in this work quantified in terms of a fractional response that equals the ratio between the response (cumulative number of counts) after 20 min and the response produced over the entire 60-min period (i.e., 100 x R20/R60). Considering the low-exposed animals, most of the counts generated during the experiment are produced in the 0 to 20-min period. However, the percentage counts produced in the 0 to 20-min period decrease at higher levels of exposure to PBDE 99 (Fig. 1). Thus, the fractional response illustrates how the habituating abilities decrease with dosage of toxicant. The time-response profiles for rearing and total activity (data not shown) are similar to that illustrated for locomotion (Fig. 1).



View larger version (18K):
[in this window]
[in a new window]
 
FIG. 1. The cumulative number of counts for locomotion observed in 2-month-old male (part A) and female (part B) C57Bl mice neonatally exposed to 0, 0.4, 0.8, 4, 8, or 16 mg PBDE 99/kg body weight. At each dose level, means (large circles) and individual observations (small circles) after 20, 40, and 60 min of testing (i.e., R20, R40, and R60) are presented. Note that the dashed lines are used for keeping track of means resulting from the same treatment and are not estimates of time-response relationships.

 
Model fitting. The maximum likelihood approach was used in this work. For continuous dose-response data, which are normally distributed with constant variance over the dose levels, di, the log-likelihood function, log L, takes the form:

(1)
where N is the total number of animals, g, the number of dose groups, ni, the number of animals in the i'th dose group, , the unbiased sample variance in the i'th dose group and, , the mean fractional response observed in the i'th dose group. The variance, {sigma}2, and parameters defining the mean response, µ(di), are estimated by the maximization of log L. Two different dose-response models were used in this work, a modified version of the Hill model,

(2)
and an exponential model, Exp,

(3)
In model (2) and (3), A is the background, and k and n are slope (shape) parameters. Several types of dose-response models were considered for this work. The two models used in the manuscript were selected based on their abilities to describe the data.

Likelihood ratio test. Likelihood ratio test statistics may be used to compare model fits and construct confidence intervals. It can be shown that minus twice the difference between the values of the maximized log-likelihood function associated with two models µ1 and µ2, i.e., –2(log Lµ1 – log Lµ2), approximately follows a {chi}2 distribution with degrees of freedom equal to the difference in the number of parameters between the two models. A critical level {alpha} = 0.05 is commonly employed as default in this type of testing (i.e., if p ≤ 0.05 the two models are considered to be significantly different). According to these test statistics, initial analysis indicated that a constant variance assumption could not be discharged considering the response definition employed in the work (i.e., the fractional response).

Simultaneous analysis of data. From the study, spontaneous behavior data were available for both male and female mice. Apart from separate analysis of the data, a certain model may be fitted simultaneously to data from both sexes for a given endpoint. Considering the maximum likelihood approach, Equation 1 needs to be slightly modified so that it also sums over k number of data sets (i.e., k = 2 for simultaneous analysis of both sexes). Simultaneous analysis of different subpopulations is of interest, since that allows a formal assessment of what model parameter/s (if any) may depend on sex. In the work, this was tested by regarding all the model parameters (A, k, n, and {sigma}2) simultaneously as well as individually using likelihood ratio tests statistics.

Benchmark dose calculation. The benchmark dose (BMD) can be defined as corresponding to a certain change in the continuous response. Among several definitions proposed, the continuous benchmark response, here denoted cBMR, may be expressed as a percentage change relative to the background value,

(4)
This definition has also been termed the critical effect size, CES (Slob and Pieters, 1998Go). The BMD can be solved by combining the expression for the cBMR and that for model (2) or (3).

Probability-based approach for benchmark dose calculation. In addition to the procedure of benchmark dose calculation presented in the previous section, a probability-based approach was also considered. According to this methodology a cutoff value is determined, below (or above) which responses are considered adverse. For data that are assumed to be normally distributed with constant variance, the probability of adverse response, p(di), at dose, di, may be expressed as the proportion falling below (or exceeding) the cutoff value, c. Considering a decreasing response with dose, the equation for the probability model, p(di) equals

(5)
where {phi} denotes the standard normal distribution function. Besides the cutoff value, p(di) is dependent on the dose-response model, µ(di), and the standard deviation, {sigma}. The benchmark dose (BMD) may be defined as corresponding to a specified additional increase in the risk of falling below the cutoff point, i.e. the benchmark response, BMR = p(BMD) – p(0). If combining the equation for the BMR with Equation 5, the continuous response level associated with the benchmark dose (BMD) is given as

(6)
The BMD is estimated by solving the equation for the dose-response model, µ(di), at µ(BMD).

Cutoff value for spontaneous behavior variables. Using the probability-based approach for BMD estimation the cutoff value, c, may be determined indirectly as corresponding to a specified tail proportion of the control distribution, p(0) (equivalent to the background response rate). In this case, c is estimated from and and (i.e., Equation 5 is solved for c). It is usually convenient to specify the cutoff in terms of p(0), since it may be hard to make judgment of what level of response is adverse for many toxicological endpoints. In the present work a cutoff, p(0) = 0.05 was selected.

An alternative way of determining a cutoff point for the spontaneous behavior variables analyzed was also considered. In Figure 1, there seems to be a dose level (around 4–8 mg PBDE 99/kg bw) at which the rate of activity on an average basis is constant during the experiment (this was also observed for rearing and total activity, data not shown). At such level of exposure habituation would be "zero." "Zero" habituation is represented theoretically by a linear time-response relationship departing from the origin. For this special case, the fractional response equals 100 x (20/60) = 100/3. Further, we define the range of fractional responses associated with habituation as, µ(d0) – 100/3. Using this definition the cutoff, c, was specified as the fractional response where habituation was considered to be reduced by 50%,

(7)

Software. Mathematical and statistical procedures required for calculations associated with this work were implemented in Matlab 6.5.


    RESULTS
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIAL AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Dose-Response Data Observed in 2-Month-Old Mice
In the present work spontaneous behavior was quantified in terms of a fractional response defined as cumulative the number of counts produced after 20 min divided by the total number of counts generated during the experiment (i.e., after 1 h of testing). In Table 1, dose-response data for this response definition are presented for the three behavior variables analyzed. According to Table 1, for 2-month-old male and female mice (which is the age period of main focus herein) the no observed adverse effect level (NOAEL) is 0.4 mg PBDE 99/kg bw for total activity, and 0.8 mg PBDE 99/kg bw for locomotion and rearing. As is shown in Table 1, the standard deviation is fairly constant over the dose levels, supporting the constant variance assumption that could not be discharged according to likelihood ratio test statistics.


View this table:
[in this window]
[in a new window]
 
TABLE 1 Dose-Response Data Observed in 2-Month-Old Male (M) and Female (F) C57Bl Mice Neonatally Exposed to PBDE 99

 


View larger version (17K):
[in this window]
[in a new window]
 
FIG. 3. The Hill and Exp models fitted simultaneously to male and female data on rearing observed in 2-month-old C57Bl mice neonatally exposed to PBDE 99. The benchmark doses (BMDs) correspond to a cBMR = 0.10. The mean of the fractional response for male (triangles) and female (squares) mice, as well as individual observations (small circles), are illustrated at each dose level.

 
Dose-Response Modeling of Data Observed in 2, 5, and 8-Month-Old Mice
Following the procedure for simultaneous analysis of data presented in Materials and Methods, it was generally concluded as most efficient to use common model parameters for male and female mice. As shown in Table 2, the overall analysis suggests that for each endpoint the two sexes may be characterized by a common dose-response model (p > 0.05). However, the parameter-specific analysis indicates that the variance is significantly different between the sexes in total activity (p ≤ 0.05). Thus, for this endpoint a sex-dependent variance was employed. The resulting parameter estimates and maximum log-likelihood values associated with selected Hill and Exp models are presented in Table 3. Considering total activity, the models from the simultaneous analysis do not describe the data well. However, in separate analysis of the male data the fit was improved according to likelihood ratio test statistics (the models from this analysis may also be applicable to female mice, but with a variance of approximately twice the size). In Figures 2 through 4 the estimated Hill and Exp models are illustrated for each endpoint.


View this table:
[in this window]
[in a new window]
 
TABLE 2 Analysis of Sex Dependence of the Dose-Response Model Parameters

 

View this table:
[in this window]
[in a new window]
 
TABLE 3 Parameter Estimates and Values of the Maximized Log-Likelihood Function Associated with Selected Hill and Exp Models

 


View larger version (17K):
[in this window]
[in a new window]
 
FIG. 2. The Hill and Exp models fitted simultaneously to male and female data on locomotion observed in 2-month-old C57Bl mice neonatally exposed to PBDE 99. The benchmark doses (BMDs) correspond to a cBMR = 0.10. The mean of the fractional response for male (triangles) and female (squares) mice, as well as individual observations (small circles), are illustrated at each dose level.

 


View larger version (15K):
[in this window]
[in a new window]
 
FIG. 4. The Hill and Exp models fitted to data on total activity observed in 2-month-old male C57Bl mice neonatally exposed to PBDE 99. The benchmark doses (BMDs) correspond to a cBMR = 0.10. The mean of the fractional response (triangles) as well as individual observations (small circles) are illustrated at each dose level.

 
In concordance to the data observed at 2 months, data from the additional age periods were most efficiently described using common model parameters for males and females (data not shown). All together, the results at the three age periods indicated that for each endpoint male and female mice responded similarly to the exposure to PBDE 99. To investigate possible reversibility of the effects observed at 2 months, dose-response curves were compared over the different age periods. The Hill model fitted to data on locomotion observed at the age of 2, 5, and 8 months is illustrated in Figure 5. As is depicted, the neurobehavioral effects observed at 2 months are sustained over time. Similar results were also obtained for total activity and rearing (data not shown).



View larger version (19K):
[in this window]
[in a new window]
 
FIG. 5. The Hill model fitted simultaneously to male and female data on locomotion observed in 2-, 5-, and 8-month-old C57Bl mice neonatally exposed to PBDE 99. The mean fractional response for male and female mice is illustrated at each dose level, for each age period.

 
Benchmark Doses for Behavior Effects After Exposure to PBDE 99
Benchmark doses (BMDs) with lower bounds (BMDLs) corresponding to continuous benchmark response levels (cBMRs) of 5 and 10% are given in Table 4. According to the Hill model at a cBMR = 0.10, the BMDLs for rearing, locomotion, and total activity are 1.2, 0.85, and 0.31 mg/kg bw, respectively. For the Exp model the corresponding values are 0.94, 0.72, and 0.35 mg/kg bw, respectively. In general, it is observed that differences between the Hill and the Exp models in terms of BMDLs are close to a factor 1, indicating similarity in model predictions (Table 4). The similarity in estimated dose-response relationships from the two models is also shown in Figures 2 through 4. According to the BMDs/BMDLs (Table 4), total activity is the most sensitive endpoint. In the range of cBMRs selected (5–10%) the BMD (and BMDL) for total activity and locomotion differ by a factor of 2 to 3. In Figures 2 through 4 it is observed that the dose-response is steeper in the low-dose region for total activity compared to the other behavior endpoints.


View this table:
[in this window]
[in a new window]
 
TABLE 4 Benchmark doses (BMDs) with lower bounds (BMDLs) corresponding to different continuous benchmark response levels (cBMRs)

 
The probability-based procedure of benchmark dose calculation was applied to data on locomotion. The Hill model, estimated from simultaneous analysis of male and female data observed at 2 months, was used as a basis (Table 3). Benchmark doses corresponding to a 5 and 10% additional risk of falling below specified cutoff levels are presented in Table 5. The BMDL estimated given a standard cutoff value, p(0) = 0.05, is somewhat lower compared to that estimated under the alternative cutoff formulation (Equation 7) (e.g., 0.63 vs. 0.99 mg PBDE 99/kg bw for a BMR = 0.10). As is noted in Table 5, the estimate of the background response rate, p(0), is in the range of 0.01 for the alternative cutoff point. Using Equation 6, it can be explored that the absolute distance, |µ(BMD) – µ(0)| increases, which results in a higher BMD, when p(0) decreases.


View this table:
[in this window]
[in a new window]
 
TABLE 5 Benchmark doses (BMDs) with lower bounds (BMDLs) for locomotion corresponding to different levels of additional risk (BMRs)

 
The BMDL estimated for locomotion using a cBMR of 5 or 10% (Table 4, Hill model) is within the range of those estimated to correspond to a 5 or 10% additional risk, respectively (Table 5).


    DISCUSSION
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIAL AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
In this paper procedures are suggested for how benchmark dose (BMD) calculations may be conducted for parameters used for testing spontaneous behavior in animals. Previous application of the benchmark dose approach to experimental neurotoxicity data has mostly focused on neurochemical and neurohistological endpoints (Gaylor and Slikker, 1990Go, 1994Go; Kodell et al., 1995Go; Slikker et al., 1998Go). The present paper extends this work to neurobehavioral endpoints.

While the benchmark dose concept is quite straightforward for quantal data, dose-response modeling approaches and BMD calculations from continuous endpoints (which are the focuses of the present work) have rendered more discussion (Barnes et al., 1995Go). The probability-based methodology partly considered herein is typically the procedure that has been suggested for BMD calculations from continuous data (Crump, 1995Go; Gaylor and Slikker, 1990Go; Kodell and West, 1993Go). Using this approach, the BMD is defined as the dose corresponding to an excess risk of falling below (or exceeding) a cutoff for "abnormal" response.

The probability-based approach for BMD estimation is interesting, since it allows introducing the concept of risk for continuous data. However, for continuous endpoints the BMD has also been presented as the dose causing a percentage change in the mean response (Slob and Pieters, 1998Go). This latter methodology was used in this work as a primary approach. While this procedure lacks the possibility to associate the BMD with an explicit level of additional risk for adverse response, it may reward some attention since it is less complex. The non-risk-based procedure primarily focuses on the mean response and not the estimate of variance, which may be imprecise in animal experiments of small sample size (Gaylor and Chen, 1996Go). Also, use of the mean response function has been discussed as a central element when comparing differences in sensitivity between subpopulations for a given exposure (Slob, 2002Go), which was an aspect of this work since data on both male and female mice were available. However, also focusing on comparability with respect to previous applications of the BMD approach to continuous data, the probability-based procedure was partly considered in the work.

Quantification and Modeling of Spontaneous Behavior
Since spontaneous behavior testing is performed over time, different ways to account for the time factor when quantifying the response to treatment may be used. Results associated with spontaneous behavior testing have previously been analyzed in terms of a habituation quotient (Eriksson et al., 2001Go). The habituation ratio equals the response (no. of counts) observed in the 40–60 min period divided by the response in the 0–20 min period (i.e., 100 x (R60 – R40)/R20). An increase in this ratio indicates a diversion from the normal habituating behavior (which is illustrated in terms of a low habituation ratio) when exposed to toxicants. Considering the fractional response employed in the present paper, a similar interpretation may be used. In this case, the decreasing habituation capacity at increasing exposure of toxicant is correlated to a decrease in the fractional response. Further, as pointed out in the sections on the likelihood ratio test and dose-response data observed in 2-month-old mice, using the fractional response the assumption of common variance was appropriate. This is of interest, since it simplifies the dose-response model, particularly considering the probability-based approach in which the variance is an important part of the model.

Benchmark Calculations Using a Continuous Benchmark Response Level
A general problem in toxicology and health risk assessment of chemicals concerns the question of what level of response may be nonadverse/adverse for a certain endpoint. In the context of the benchmark dose approach this issue is closely related to specifying the value of the continuous benchmark response, cBMR (Dekkers et al., 2001Go). Similarly, considering the probability-based procedure, this relates to the specification of the cutoff value (and the BMR). Ideally, knowledge concerning what level of change in an endpoint that is acceptable should be used as a basis for determination of the cBMR (Slob, 2002Go; Slob and Pieters, 1998Go). However, this type of information is generally lacking for most continuous endpoints. In the face of limited biological knowledge, default values may have to be employed. The cBMRs used in this work (i.e., 0.05 and 0.10) should be interpreted in this way, i.e., be seen as general defaults and not in particular established for the endpoints investigated (Edler et al., 2002Go). The suggestion of a biologically based cBMR for the behavior endpoints analyzed is beyond the scope of this paper. However, considering the range of fractional responses associated with habituation (defined in the section on cutoff value for spontaneous behavior variables), at the cBMRs of 5 and 10%, habituation may be considered to be reduced by approximately 10 and 20%, respectively (calculations can be worked out using the background estimates in Table 3, Equation 4, and the definition of the range of fractional responses associated with habituation in the section describing cutoff value for spontaneous behavior variables).

According to the results in Table 4, in the region of responses observed, total activity is more sensitive compared to the other endpoints (according to the BMD and BMDL). It is realized that in general it is not appropriate to compare different continuous endpoints at the same cBMR. The probability-based approach, which takes the variance into account, could rather be more preferable for such an exercise. However, since spontaneous behavior (locomotion, rearing, and total activity) in this work was analyzed using a common measurement (i.e., in terms of the fractional response), a certain percentage change in this standardized unit was considered comparable.

The critical endpoint upon which risk assessment is based has usually been defined as the first adverse effect obtained as the dose level increases (USEPA, 1995Go). From this context, use of total activity as a critical endpoint may also protect for other alterations in behavior caused by neonatal exposure to toxicants. However, it may be noted that, while locomotion and rearing are distinct movements, total activity nonspecifically represents different types of vibrations associated with movements (i.e., locomotion and rearing) and also acute effects like shaking and tremor. Considering the two former endpoints, locomotion is a more specific measure of activity, since the rearing variable to a large extent may indicate exploratory activity and learning rather than undifferentiated motor activity (Archer et al., 1990Go, Fredriksson et al., 1992Go). Thus, even though total activity is interesting from a traditional risk assessment context, the other spontaneous behavior variables (particularly locomotion) may be preferable due to their higher degree of specificity.

With regard to each of the endpoints analyzed, dose-response data on 2-month-old male and female mice could be characterized by a common dose-response model (Table 2), indicating that males and females were equally sensitive to the exposure to PBDE 99. This was also the case with respect to the data observed at 5 and 8 months of age. Further, the analysis of data observed at different age periods indicated that neurobehavioral defects caused by exposure to PBDE 99 might be sustained over time (Fig. 5).

Benchmark Calculations Using Probability-Based Procedures
The probability-based method of benchmark dose calculation was in the present work applied to one of the behavior variables, locomotion. This variable was selected since it represents a common measure of motor activity, and also for reasons discussed in the previous section. In addition, the dose-response models estimated for this endpoint are based on a quite large body of data—120 animals (60 males and 60 females). Since the probability-based procedure for BMD estimation is more complex, it is of interest to use a large study group as a basis.

A crucial aspect of the probability-based approach concerns the determination of the cutoff value. In previous applications, the cutoff point has been specified as corresponding to some tail proportion of the control distribution, p(0), or similarly as a number of standard deviations from the control mean. The value of p(0) has typically been suggested to be in the range of 0.01–0.05 (Kodell et al., 1995Go). In epidemiological studies, p(0) = 0.05, seems to have evolved as standard (Budtz-Jorgensen et al., 2000Go; Crump et al., 2000Go; Jacobson et al., 2002Go). This value of the cutoff has been referred to be consistent with the definition of the normal range in clinical data (Crump 2002Go; Crump et al., 2000Go). Since a selection of p(0) = 0.05 may serve as a future policy, as one case we decided to perform calculations using this cutoff.

An alternative approach for defining the cutoff value for spontaneous behavior data was also considered. The cutoff was in this context specified as the fractional response where habituation was considered to be reduced by 50% (Equation 7). In general, there may be an interest in defining a cutoff not solely based on statistics. Using the default procedure, the cutoff value may be highly dependent on the model: it depends on the distributional assumption, the variance, and the mean response function estimated. The alternative cutoff used in the paper (Equation 7) is independent in the two former respects. It should be noted that the influence of variance on the BMD is different depending on the cutoff formulation. In Sand et al. (2003)Go it was discussed that, given a cutoff in terms of p(0), the BMD increases with the variance, while the opposite is the case when the cutoff is fixed (not dependent on the variance). Use of a fixed cutoff may be more relevant when comparing different populations (e.g., a population with large variance presumably includes a high number of sensitive individuals which would be reflected by a high estimate of p(0) and a low BMD [Equation 6], and vise versa) (Sand et al., 2003Go; Slob and Pieters, 1998Go).

The selection of the benchmark response level (BMR) concerns another important aspect of decision making. Originally, for quantal data the BMR was proposed as an additional (or extra) risk of 1–10% (Crump, 1984Go; USEPA, 1995Go). Studies investigating the effect of model dependence for quantal data have shown that that the choice of model may not have high impact on the BMDL for a BMR down to 5% (Allen et al., 1994Go; Sand et al., 2002Go). However for continuous data, model dependence can be more pronounced. Continuous models are more diverse in their structures, and the range of continuous responses may not be constrained (as opposed to quantal data, where the response always ranges from 0 to 1). Thus, the use of a BMR = 0.10 may be more appropriate. For determination of guidance values, the BMDL represents a point of departure, and before arriving at a final reference dose, uncertainty factors are applied to the BMDL. It has been discussed that the uncertainty factors could reflect the choice of the BMR (Crump, 2002Go).

Summary and Conclusion
This work, using PBDE 99 as model substance, illustrates that proposed methods of quantitative health risk assessment can be implemented for spontaneous behavior variables. In the context of the work, a response definition was developed which was used as a basis for benchmark dose modeling. The benchmark dose concept represents an improvement of the traditional practice in risk assessment of chemicals (i.e., the NOAEL approach). In the present work NOAELs were in the same range as BMDLs corresponding to a continuous benchmark response level of 10% (cBMR = 0.10). Compared to the NOAEL approach, the BMD methodology makes more use of the available data (e.g., the BMDs/BMDLs estimated for locomotion and rearing are based on data observed in 120 individual subjects). Also, an explicit response level is associated with the benchmark dose, suggesting the point of departure for human health risk assessment to be based on more information.


    ACKNOWLEDGMENTS
 
This work was supported by grants from MISTRA/NewS (A New Strategy for the Risk Management of Chemicals, Dnr 98003), and also performed within the EU funded CASCADE Network of Excellence (Contract Nr: FOOD-CT-2003–506319).


    NOTES
 

1 To whom correspondence should be addressed at Institute of Environmental Medicine, Karolinska Institutet, P.O. Box 210, 17177 Stockholm, Sweden. Fax: +468343849. E-mail: Salomon.Sand{at}imm.ki.se.


    REFERENCES
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIAL AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Allen, B. C., Kavlock, R. J., Kimmel, C. A., and Faustman, E. M. (1994). Dose-response assessment for developmental toxicity: III. Statistical models. Fundam. Appl. Toxicol. 23, 496–509.[CrossRef][ISI][Medline]

Archer, T., Beninger, R. J., Järbe, T. U. C., and Seiden, L. S. (1990). Latent learning in a radial arm maze following neonatal dopamine depletion. Behav. Pharmacol. 1, 191–199.

Barnes, D., Datson, P., Evans, J., Jarabek, A., Kavlock, R., Kimmel, C., Park, C., and Spitzer, H. (1995). Benchmark dose workshop: Criteria for use of a benchmark dose to estimate a reference dose. Regul. Toxicol. Pharmacol. 21, 296–306.[CrossRef][ISI][Medline]

Bogdan, M., MacPhail, R., and Glowa, J. (2001). A randomization test-based method for risk assessment in neurotoxicology. Risk. Anal. 21, 107–116.[CrossRef][ISI][Medline]

Budtz-Jorgensen, E., Grandjean, P., Kieding, N., White, R. F., and Weihe, P. (2000). Benchmark dose calculations of methylmercury-associated neurobehavioral deficits. Toxicol. Lett. 112–113, 193–199.[CrossRef][ISI]

Crump, K. (1984). A new method for determining allowable daily intakes. Fundam. Appl. Toxicol. 4, 854–871.[ISI][Medline]

Crump, K. (1995). Calculation of benchmark doses from continuous data. Risk Anal. 15, 79–89.[ISI]

Crump, K., Kjellstrom, T., Shipp, A., Silvers, A., and Stewart, A. (1998). Influence of prenatal mercury exposure upon scholastic and psychological test performance: Benchmark analysis of a New Zeeland cohort. Risk Anal. 18, 701–713.[CrossRef][ISI][Medline]

Crump, K., Van Landingham, C., Shamlaye, C., Cox, C., Davidson, P., Myers, G., and Clarkson, W. (2000). Benchmark concentrations for methylmercury obtained from the Seychelles child development study. Environ. Health Perspect. 108, 257–263.[ISI][Medline]

Crump, K. (2002). Critical issues in benchmark calculations from continuous data. Crit. Rew. Toxicol. 32, 133–153.

Davison, A. N., and Dobbing, J. (1968). Applied Neurochemistry, Blackwell, Oxford.

Dekkers, S., de Heer, C., and Rennen, M. (2001). Critical effect sizes in toxicological risk assessment: A comprehensive and critical evaluation. Environ. Toxicol. Pharmacol. 10, 33–52.[CrossRef][ISI][Medline]

Edler, L., Poirier, K., Dourson, M., Kleiner, J., Mileson, B., Nordmann, H., Renwick, A., Slob, W., Walton, K., and Wurtzen, G. (2002). Mathematical modeling and quantitative methods. Food Chem. Toxicol. 40, 283–326.[CrossRef][ISI][Medline]

Eriksson, P. (1997). Developmental neurotoxicity of environmental agents in the neonate. Neurotoxicology 18, 719–726.[ISI][Medline]

Eriksson, P. (1998). Perinatal developmental neurotoxicity of PCBs. Report 4897. Stockholm, Sweden: Swedish Environmental Protection Agency, 1–56.

Eriksson, P., Jakobsson, E., and Fredriksson, A. (2001). Brominated flame retardants: A novel class of developmental neurotoxicants in our environment? Environ. Health Perspect. 109, 903–908.[ISI][Medline]

Eriksson, P., Viberg, H., Jakobsson, E., Örn, U., and Fredriksson, A. (2002). A brominated flame retardant, 2,2',4,4',5-pentabromodiphenyl ether: Uptake, retention, and induction of neurobehavioral alterations in mice during a critical phase of neonatal brain development. Toxicol. Sci. 67, 98–103.[Abstract/Free Full Text]

Falk Filipsson, A., Sand, S., Nilsson, J., and Victorin, K. (2003). The benchmark dose method—review of available models, and recommendations for application in heath risk assessment. Crit. Rew. Toxicol. 33, 505–542.

Fredriksson, A. (1994). MPTP-induced behavioral deficits in mice: Validity and utility of a model of parkinsonism. Acta Universitatis Uppsaliensis, Comprehensive Summaries of Uppsala Dissertations from the Faculty of Medicine.

Fredriksson, A., Dahlgren, L., Danielsson, B., Eriksson, P., Dencker, L., and Archer, T. (1992). Behavioral effects of neonatal metallic mercury exposure in rats. Toxicology 74, 151–160.[CrossRef][ISI][Medline]

Gaylor, D., and Chen, J. (1996). Precision of benchmark dose estimates for continuous (nonquantal) measurements of toxic effects. Regul. Toxicol. Pharmacol. 24, 19–23.[CrossRef][ISI][Medline]

Gaylor, D., and Slikker, W. (1990). Risk assessment for neurotoxic effects. Neurotoxicology 11, 211–218.[ISI][Medline]

Gaylor, D., and Slikker, W. (1994). Modeling for risk assessment of neurotoxic effects. Risk Anal. 14, 333–338.[ISI][Medline]

Jacobson, J. L., Janisse, J., Banerjee, M., Jester, J., Jacobson, S. W., and Ager, J. (2002). A benchmark dose analysis of prenatal exposure to polychlorinated biphenyls. Environ. Health Perspect. 110, 393–398.

Kodell, R., and West, R. (1993). Upper confidence limits on excess risk for quantitative responses. Risk Anal. 13, 177–182.[ISI][Medline]

Kodell, R. L., Chen, J. J., and Gaylor, D. W. (1995). Neurotoxicity modeling for risk assessment. Regul. Toxicol. Pharmacol. 22, 24–29.[CrossRef][ISI][Medline]

Meironyté, D., Norén, K., and Bergman, A. (1999). Analysis of polybrominated diphenyl ethers in Swedish human milk: A time-related trend study, 1972–1997. J. Toxicol. Environ. Health A. 58, 329–341.[CrossRef][ISI][Medline]

Sand, S., Falk Filipsson, A., and Victorin, K. (2002). Evaluation of the benchmark dose method for dichotomous data: Model dependence and model selection. Regul. Toxicol. Pharmacol. 36, 184–197.[CrossRef][ISI][Medline]

Sand, S., von Rosen, D., and Falk Filipsson, A. (2003). Benchmark calculations in risk assessment using continuous dose response information: The influence of variance and the determination of a cutoff value. Risk Anal. 23, 1059–1068.[CrossRef][ISI][Medline]

Slikker, W., Scallet, A., and Gaylor, D. (1998). Biologically-based dose-response model for neurotoxicity risk assessment. Toxicol. Lett. 102–103, 429–433.[CrossRef]

Slob, W., and Pieters, M. N. (1998). A probabilistic approach for deriving acceptable human intake limits and human health risks from toxicological studies: General framework. Risk Anal. 18, 787–798.[CrossRef][ISI][Medline]

Slob, W. (2002). Dose-response modeling of continuous endpoints. Toxicol. Sci. 66, 298–312.[Abstract/Free Full Text]

U.S. Environmental Protection Agency (USEPA). (1995). The Use of the Benchmark Dose (BMD) Approach in Health Risk Assessment. Final report. EPA/630/R-94/007. Risk Assessment Forum, U.S. Environmental Protection Agency, Washington, DC.

Viberg, H., Fredriksson, A., and Eriksson, P. (2002). Neonatal exposure to the brominated flame retardant 2,2',4,4',5-pentabromodiphenyl ether causes altered susceptibility in the cholinergic transmitter system in the adult mouse. Toxicol. Sci. 67, 104–107.[Abstract/Free Full Text]

Viberg, H., Fredriksson, A., and Eriksson, P. (2003a). Neonatal exposure to polybrominated diphenyl ether (PBDE 153) disrupts spontaneous behavior, impairs learning and memory, and decreases hippocampal cholinergic receptors in adult mice. Toxicol. Appl. Pharmacol. 192, 95–106.[CrossRef][ISI][Medline]

Viberg, H., Fredriksson, A., Jakobsson, E., Örn, U., and Eriksson, P. (2003b). Nerobehavioral derangements in adult mice receiving decabrominated diphenyl ether (PBDE 209) during a defined period of neonatal brain development. Toxicol. Sci. 76, 112–120.[Abstract/Free Full Text]

Viberg, H., Fredriksson, A., and Eriksson, P. (2004). Investigations of strain and/or gender differences in developmental neurotoxic effects of polybrominated diphenyl ethers in mice. Toxicol. Sci. 81, 344–353.[Abstract/Free Full Text]