       Using Internal Dose-Response Comparisons to Identify Dose Metrics
                      Which Best Correlate With Toxicity
                              (Version 09/16/13)
                                       
An internal dose metric such as a measure of toxicant concentration in the blood is expected to be a better predictor of response than the applied dose (e.g., concentration in air) since it is closer to the site of damage and chemical activity which causes that damage.  Further, a good internal dose metric should correlate with or be predictive of toxicity irrespective of the route of exposure by which it occurs.  However this is only true if the metric is in fact a measure of the likelihood of a toxic response, or intensity of a toxic effect.  For example, if the parent compound is the toxic moiety while a metabolite is relatively inert, then the concentration of the metabolite is a measure of exposure, but one that would likely be a worse predictor than the applied dose.
For NMP the existing toxicity data identify the parent (NMP) rather than the primary metabolite (5HNMP) as the proximate toxicant.  But toxicity could depend on peak concentration (Cmax, irrespective of the number of days or hours over which it occurs), the area-under-the curve (AUC) of concentration during the vulnerable period of fetal development (lasting a week or more in rats, likely longer in humans), or some intermediate measure.  Here we consider the highest 24-hour AUC (Max AUC24) in addition to the two metrics mentioned above for developmental effects; i.e., the highest AUC for a 24-hour period, which would reflect a need for more than momentary exposure for a toxic effect, but that a significant developmental effect could result from a single day exposure during  a critical period of vulnerability.
Below we first analyze dose-metrics related to reductions in fetal body weight (BW) as a toxic endpoint and then skeletal aberrations.  By comparing these metrics and observed responses for two (or more) routes of exposure, we can determine if a metric is at least consistent in its prediction of a toxic response across routes.  For example, if a much higher Cmax occurs for oral exposures for which the toxic endpoint is not triggered than for inhalation exposures for which the response is triggered, that would be an inconsistent metric relative to the response, indicating that some other metric would be better.  Specifically, we compare dose-response relationships based on the following internal doses estimated using the EPA-revised PBPK model: a) Max AUC24, the maximum 24-hr AUC as described above; b) Average AUC24, the average daily AUC over the entire study period (total AUC divided by the number of days), GD6-20 for Becci et al. (1982) and GD5-21 for Saillenfait et al. (2003); and c) Cmax.
Fetal BW 
Becci et al. (1982) observed a reduction in fetal BW at the highest dose administered by dermal exposure (Table 1), while Saillenfait et al. (2003) observed a significant reduction in fetal BW at their highest inhalation exposure (Table 2), though with a clearer trend across intermediate doses.  Becci et al. (1982) applied their dermal doses on GD 6 to GD 15 and evaluated the impact on the developing pups at GD 20.  Saillenfait exposed by inhalation from GD 6to GD 20 (6 h/d) and evaluated the pups on GD 21. 


Table 1.  Becci et al. (1982) Fetal BW Dose-Response
                               Exposure (mg/kg)
                              Max AUC24 (mg-h/L)
                            Average AUC24 (mg-h/L)
                                  Cmax (mg/L)
                                 Infant BW (g)
                                       0
                                       0
                                       0
                                       0
                                      3.5
                                      75
                                      536
                                      359
                                      202
                                      3.5
                                      237
                                     2070
                                     1385
                                      641
                                      3.5
                                      750
                                     8964
                                     6000
                                     2050
                                     2.8*
*Significantly different from water treated group (negative control), p < 0.05

Table 2.  Saillenfait et al. (2003) Fetal BW Dose-Response
                              Exposure (mg/m[3])
                              Max AUC24 (mg-h/L)
                            Average AUC24 (mg-h/L)
                                  Cmax (mg/L)
                                 Infant BW (g)
                                       0
                                       0
                                       0
                                       0
                                 5.67 +- 0.37
                                      122
                                      96
                                      94
                                      12
                                 5.62 +- 0.36
                                      243
                                      194
                                      191
                                      24
                                 5.47 +- 0.25
                                      487
                                      404
                                      397
                                      50
                                 5.39 +- 0.45*
                                    BMCL05
                                      255
                                      246
                                     30.7
                                       - 
*Significant difference from the control (air), P<0.05.

The first thing to be noted is that all of the internal doses for the Saillenfait et al. (2003) study are well below the corresponding doses for the Becci et al. (1982).  Further, the Benchmark Concentration Lower (95%) confidence limit for a 5% response (BMCL05) values obtained using each of the internal metrics from Saillenfait et al. (2003) with their observed BW response are all below the lowest corresponding internal doses for Becci et al. (1982), where there is no indication of any toxicity.
The internal doses estimated with the PBPK model indicate that the lack of responses in fetal BW at 75 and 237 mg/kg by dermal application did not occur because a smaller amount of NMP penetrated to the maternal circulation than occurred from the inhalation exposures of Saillenfait et al. (2003) where an effect occurred.  Other factors must be considered for the dose-response comparison.
How is it possible that the Saillenfait et al. (2003) animals appear to have responded at much lower internal doses than the animals dosed by Becci et al. (1982)?  A key difference in the two studies is the days when dosing vs. observation occurred:
   Saillenfait exposed GD 6-20 and sacrificed on GD 21;
   Becci exposed GD6-15 and sacrificed on GD 20.
A possible explanation is that the 5 days between administration of the last dose and animal sacrifice in the Becci et al. study allowed for a partial recovery with regard to this endpoint.  NMP's pharmacokinetics are such that it is almost entirely cleared within 24 hours.  So the animals in the Becci et al. study would have had 5 days of development with essentially zero internal dose.  On the other hand, given the very high internal doses associated with the Becci et al. (1982), it seems unlikely that a larger effect on fetal BW would not have occurred, if they had continued dosing until GD 19 (one day before sacrifice in that study) or GD 20 (the last day of exposure in the Saillenfait et al., 2003 study).  
Thus it appears that the fetal BW decrement depends on continued exposure, rather than intermittent exposure.  Thus, a higher exposure occurring earlier in gestation is not the principal determinant for this adverse effect on BW.  Therefore, the average AUC24 is likely the best internal dose metric for this endpoint, in particular when evaluating the risks of workplace exposures which can occur repeatedly over longer periods of gestation.
If this explanation is valid, then human internal doses from dermal as well as inhalation exposure should be evaluated against internal dose BMCL05 values obtained from the Saillenfait et al. (2003) study, particularly for workplace scenarios.
Note that BMD calculations for the fetal BW endpoint were not conducted for Becci et al., (1982) because the mean fetus weight values were reported without standard deviations.  Consequently, BMDL05 comparisons for skeletal effects, but not the fetal BW endpoint, were conducted for the dermal and inhalation internal doses reported in Becci et al., (1982) and Saillenfait et al. (2003).
Skeletal Effects
Becci et al. (1982) observed a number of significant skeletal abnormalities at their highest dose.  Here the incomplete ossification of vertebrae (Table 3) is specifically considered because the dose-response is amenable to benchmark analysis, and Saillenfait et al. (2003) also reported this endpoint (Table 4).  As seen in Tables 3 and 4 (same internal dose metrics as those for fetal BW), the results of Saillenfait et al. (2003) are consistent with those of Becci et al. (1982), irrespective of dose metric, for skeletal effects.  In particular Saillenfait et al. (2003) did not see a significant effect at the lower internal dose levels from their inhalation exposures, compared to the effect seen by Becci et al. (1982).  Thus all of the internal metrics appear to be consistent between the Becci et al. (1982) dermal exposures and Saillenfait et al. (2003) studies.  Note that the highest dose (750 mg/kg) in Becci et al., (1982) showed a significant decrease in fetal BW and significant increased incidence for vertebrae incomplete ossification.  But changes in fetal BW appear to be reversible and depend on continued exposure that would be best expressed as the average AUC24.  On the other hand, the type of fetal skeletal effects analyzed here (i.e., vertebrae/ thoracic vertebral centra, incomplete ossification) may be less reversible and may result from acute -dose effects.  For vertebrae incomplete ossification, consistency of the two dose-response data sets does not preclude Cmax or Max AUC24 as metrics.






Table 3.  Becci et al. (1982) "Vertebrae, incomplete ossification
                               Exposure (mg/kg)
                              Max AUC24 (mg-h/L)
                            Average AUC24 (mg-h/L)
                                  Cmax (mg/L)
                                 Litters/total
                                       0
                                       0
                                       0
                                       0
                                     12/24
                                      75
                                      536
                                      359
                                      202
                                     11/22
                                      237
                                     2070
                                     1385
                                      641
                                     14/23
                                      750
                                     8964
                                     6000
                                     2050
                                    18/22*
                                  BMCL05[†]
                                      110
                                      74
                                     31.2
                                       - 
*Significantly different from water treated group (negative control), p < 0.05
[†]Only LogLogistic model tested for AUC metrics; was selected for a proportional average AUC metric


Table 4.  Saillenfait et al. (2003) "Thoracic vertebral centra, incomplete ossification"
                              Exposure (mg/m[3])
                              Max AUC24 (mg-h/L)
                            Average AUC24 (mg-h/L)
                                  Cmax (mg/L)
                                 Litters/total
                                       0
                                       0
                                       0
                                       0
                                     10/24
                                      122
                                      96
                                      94
                                      12
                                     5/20
                                      243
                                      194
                                      191
                                      24
                                     8/19
                                      487
                                      404
                                      397
                                      50
                                     5/25


Further indication of which internal metric is best may come from also analyzing the oral toxicity study of Saillenfait et al. (2002), for which the dose-response is given in Table 5.  Unfortunately incomplete ossification of the vertebrae was not reported in this study, but the total incidence of skeletal abnormalities was provided.

Table 5.  Saillenfait et al. (2002) "Litters with skeletal malformations"
                               Exposure (mg/kg)
                              Max AUC24 (mg-h/L)
                            Average AUC24 (mg-h/L)
                                  Cmax (mg/L)
                                 Litters/total
                                       0
                                       0
                                       0
                                       0
                                     0/20
                                      125
                                      659
                                      574
                                      86
                                     0/21
                                      250
                                     1478
                                     1281
                                      179
                                     0/24
                                      500
                                     3498
                                     3024
                                      380
                                    12/25**
                                      750
                                     5911
                                     5110
                                      591
                                     3/5**
**Significant difference from vehicle control, p < 0.01

The Becci et al. (1982) dermal results appear roughly consistent with the Saillenfait et al. (2002) oral studies; the skeletal responses become significant in the same range of doses.  Hence examining these more closely may give insight as to which is the best dose metric.
However, the incomplete ossification characterized by Becci et al. (1982) had a 50% background rate, while the skeletal malformations scored by Saillenfait et al. (2002) had a zero background rate.  In order to put these on a more similar scale, the extra risk was calculated: 
	Extra risk = (response  -  background)/(100%  -  background) .
For Saillenfait et al. (2002), this is just the fraction of litters effected since the background is zero, but for Becci et al. (1982) it is the increase over background relative to the fraction of unaffected litters in the control group.  Plots of the extra risk for each study vs. the three metrics follow (Figure 1).  
1)	The average AUC metric (first plot) appears slightly inconsistent in that there is an increase in the Becci et al. (1982) response at 1385 mg-hr/L, but no response in the Saillenfait et al. (2002) study at 1281 mg-hr/L.  Some caution is warranted, though, since the change from control seen by Becci et al. (1982) is only 2 additional litters affected and by itself is not statistically significant. 
2)	Both the max AUC24 and Cmax metrics are consistent in where the dose response changes from control.  It appears more so for the Cmax metric, but this may be a result of the dose spacing.
3)	Given that total skeletal abnormalities (Saillenfait et al., 2002) should be inclusive of the specific vertebrae delayed ossification (Becci et al., 1982), that the response at higher doses is greater for the Saillenfait et al. (2002) study when based on the Cmax metric (last plot) does not necessarily indicate an inconsistency for this metric.
Hence both the Cmax and the max AUC24 metrics appear reasonable for further evaluation.

Figure 1.  Plots of "extra risk" vs. dose for Becci (1982) dermal exposures vs. Saillenfait et al. (2002) oral studies
                                       
                                       
                                       
                                       
Combined inhalation and oral dose-response analysis
Since the Saillenfait et al. (2003) inhalation data and the Becci et al. (1982) dermal data for vertebrae delayed ossification appear consistent relative to the internal dose metrics, the two studies are analyzed together for both the Cmax metric and the max AUC24 metric, since both of those also appear consistent with the oral response of Saillenfait et al. (2002), as shown just above (Tables 6-9; Figures 2 and 3).  Combining the data sets should provide additional statistical power for identifying the BMD(L) and provide a more robust dose-response (low to high).
Table 6.  Combined Becci et al., (1982) dermal and Saillenfait et al. (2003) inhalation data for vertebrae incomplete ossification, Cmax metric 
Dose -- Cmax
N
Effect
notes
                                       0
                                      48
                                      22
                                   combined
                                     11.9
                                      20
                                       5
                                  Saillenfait
                                     24.5
                                      19
                                       8
                                  Saillenfait
                                     49.57
                                      25
                                       5
                                  Saillenfait
                                     202.4
                                      22
                                      11
                                     Becci
                                     641.1
                                      23
                                      14
                                     Becci
                                    2049.6
                                      22
                                      18
                                     Becci

Table 7.  Benchmark dose analysis of combined Becci et al. (1982) dermal and Saillenfait et al. (2003) inhalation data for vertebrae incomplete ossification, Cmax metric
                            BMD Estimation Results
Model
P-Value
AIC
BMD 
(BMR=10%)
BMDL
(BMR=10%) 
BMD
(BMR=5%)
BMDL 
(BMR=5%)
Gamma
0.125
235.31
180.45
102.61
91.49
49.95
Logistic
0.179
233.77
244.03
169.56
123.48
85.78
LogLogistic
0.140
235
174.78
57.27
98.26
27.13
LogProbit
0.203
233.329
277.44
169.82
192.93
118.09
Multistage
0.205
233.327
162.95
102.49
79.33
49.90
Probit
0.175
233.84
251.66
181.67
127.01
91.67
Weibull
0.125
235.33
168.05
102.5
82.63
49.90
Quantal-Linear
0.205
233.327
162.95
102.49
79.33
49.90


Note: Multistage model and Quantal-Linear model have the lowest AIC. But the BMDLs exceed 3-fold range, so the BMDL from the Log logistic model should be selected.  The Log logistic and the best fitted model Quantal-linear is shown below.


The BMDL05 is essentially identical to that obtained from the analysis of the BW data using the same metric (30.9 mg/L) and for the analysis of the Becci et al. data alone (31.2 mg/L).
Figure 2.  Plots for log-logistic model and quantal linear model, BMDL05 (Cmax) for vertebrae incomplete ossification
                                       
                                       
Table 8.  Combined Becci et al., (1982) dermal and Saillenfait et al. (2003) inhalation data for vertebrae incomplete ossification, max AUC24 metric
                                Dose  -  AUC24
                                       N
                                    Effect
                                     notes
                                       0
                                      48
                                      22
                                   combined
                                     95.8
                                      20
                                       5
                                  Saillenfait
                                     194.4
                                      19
                                       8
                                  Saillenfait
                                     404.2
                                      25
                                       5
                                  Saillenfait
                                     536.5
                                      22
                                      11
                                     Becci
                                    2070.3
                                      23
                                      14
                                     Becci
                                    8963.7
                                      22
                                      18
                                     Becci

Table 9.  Benchmark dose analysis of combined Becci et al. (1982) dermal and Saillenfait et al. (2003) inhalation data for vertebrae incomplete ossification, max AUC24 metric
                            BMD Estimation Results
Model
P-Value
AIC
BMD 
(BMR=10%)
BMDL
(BMR=10%) 
BMD
(BMR=5%)
BMDL 
(BMR=5%)
Gamma
0.07
236.82
853.56
439.84
444.50
214.13
Logistic
0.11
235.22
1079.48
733.36
545.60
370.63
LogLogistic
0.08
236.39
783.86
247.37
447.77
117.18
LogProbit
0.13
234.57
1187.74
698.74
825.93
485.89
Multistage
0.12
234.86
718.01
438.26
349.56
213.36
Probit
0.11
235.28
1114.7
790.95
562.06
298.77
Weibull
0.07
236.85
788.37
438.71
395.28
213.58
Quantal-Linear
0.12
234.86
718.01
438.26
349.56
213.36
Note: Models that cannot fit the data properly are not considered. LogProbit model, which has the lowest AIC value is selected. Figure for LogProbit model is shown below (next page).
                                       
                                       
                                       
                                       




Figure 3.  Plots for log-logistic model and quantal linear model, BMDL05 (max AUC24) for vertebrae incomplete ossification
                                       

References
Becci, P. J., Knickerbocker, M. J., Reagan, E. L., Parent, R. A., and Burnette, L. W. (1982). Teratogenicity study of N-methylpyrrolidone after dermal application to Sprague-Dawley rats. Fundam. Appl. Toxicol. 2(2), 73-76. 
Saillenfait, A. M., Gallissot, F., and Morel, G. (2003). Developmental toxicity of N-methyl-2-pyrrolidone in rats following inhalation exposure. Food Chem. Toxicol. 41(4), 583-588. 

