Quantification of Toxic Effects
     for Water Concentration-Based Aquatic Life Criteria

                               PartB
Section 4: Pentachloroethane Lethality to Juvenile Fathead Minnows
                           Russell J. Erickson
                      Mid-Continent Ecology Division
              National Health and Environmental Effects Laboratory
                     Office of Research and Development
                    U.S. Environmental Protection Agency
                           Duluth, Minnesota
                                Final
                             July 20, 2011

-------

-------
Section 4.  Pentachloroethane Lethality To Juvenile Fathead Minnows






4.1 Overview



       Erickson et al. (1991) conducted a series of experiments on the toxicity of



pentachloroethane (PCE) to juvenile fathead minnows. These experiments included evaluations



of bioaccumulation kinetics, the time-course of mortality under both constant and time-variable



exposures, the response offish growth rate to constant and time-variable exposures, and the



relationship of toxic effects to PCE accumulation. This section will examine mortality in these



experiments and evaluate the applicability of the toxicity models discussed in Section 2 of Part A



of this report, first considering the "implicit" models already applied to copper toxicity in Section



3, and then the "explicit" models using information gathered in these experiments regarding PCE



accumulation kinetics and the relationship of effects to PCE accumulation. Both deterministic



and stochastic toxicity models will be evaluated, but only in their simplest forms (Equations 2.1-



2.7, Equations 2.22-2.28), because the data in these  experiments are insufficient to consider more



complicated models with multiple processes and/or  compartments.  Subsequent work will



evaluate growth effects in these experiments.



4.2 Study Description



       The study of Erickson et al. (1991) consisted of the following experiments:



(1) A bioconcentration experiment (Experiment Bl) in which ca. 4-week-old fathead minnows were



exposed to five levels of PCE (ranging from 1 mg/L, a no effect concentration for growth and



survival, to 10 mg/L, lethal within 12 h) for 48 h, followed by an elimination period of 24 h in



uncontaminated water. Accumulation of PCE in the fish was monitored at 1, 2, 4, 8, 24, and 48 h



during the exposure period and at 1, 2, 4, 8, and 24 h during the elimination period.



(2) 4-d tests of survival (Experiments Al, A2) with ca. 4-week-old fish. In Experiment Al,



continuous exposures at five concentrations (Test Ale) were evaluated simultaneously with daily 6-



h pulses of five intensities (Test Alp). In Experiment A2, continuous exposure (Test A2c) was



contrasted both to daily 6-h pulses (Test A2p) and to an incrementally increasing ("stepped")

-------
exposure (Test A2s), in which the five treatments started at the five same concentrations as the



continuous exposure and each was increased daily to the next higher concentration (terminating



each treatment after it reached the highest concentration for one day).  Accumulation of PCE was



measured in samples offish that died and in fish surviving at the end of the test.



(3) 28-d tests of survival and growth, starting with 2- to 3-week-old fish. In Experiment Cl,



continuous exposure (Test Clc) and daily 6-h pulses (Test Clp6h) were evaluated. In Experiment



C2, continuous exposure (Test C2c) was contrasted to exposures only during the first 2 weeks (Test



C2b2w) or the final 2 weeks (Test C2f2w).  In Experiment C3, continuous exposure (Test C3c),



daily 3-h pulses (Test C3p3h), and daily 8-h pulses (Test C3p8h) were evaluated. In Experiment



C4, continuous exposure (Test C4c) was contrasted with exposures of 1 d per week (Test C4pld)



and 3 d per week (Test C4p3d).  The highest exposure concentrations  in these tests was set to be



near acutely lethal levels. Growth rates were determined by weighing random samples of test



organisms at weekly intervals.  Accumulation of PCE was measured in fish  collected for growth



determinations and, for experiments C3 and C4, in subsamples offish that died.



(4) An experiment (Experiment Gl) to determine the time dependence of growth effects, both



during and after exposure. Fish were exposed to five levels of PCE for 7 d and then to clean water



for 11 days. Growth rates were determined by weighing random samples offish at 2-3 d intervals.



Accumulation of PCE was measured in samples offish collected for growth determination.



Accumulation of PCE was also determined in fish dying at the higher  exposure concentrations and



in selected live fish collected throughout the period in which mortality occurred.



4.3 Mortality Observations



       Figure 4.1 shows cumulative percentage mortality for five continuous exposure



treatments from Experiments A2, Gl, and C2  (left side) and five pulsed and stepped exposure



treatments from Experiments Al and A2 (right side).  For Experiment Gl, there was substantial



sampling of surviving organisms throughout periods with high mortality rates. For such a



situation, deaths  during different observation intervals will have different  effects on cumulative



mortality depending on the number of live samples preceding each interval. Thus, cumulative

-------
Figure 4. 1
selected cc
and A2 (rij
100-
80-
60-
40-
20-
0-
C
100-
80-
60-
40-
>? 0-
o 100-
^ 80-
o5 60'
L£ 40-
(D
Q_ 20-
0) Q.
1 C
g 100-
13 80-
° 60-
40-
20-
0 •
C
100-1
80-
60-
40-
20-
n •
Observed mortality (bold lines) as a fun
ntinuous exposures from Tests A2, Gl, a
^ht side).
r
/
/ G1-1
1
1
) 24 48 72 96 120 144 1£
f
1 A2c-1
f
ction of watei
nd C2 (left sic
•10 100i
•8 80 -H
6 60 Jl
4 40 Jl
•2 20 I
• o o l_
58 C
•10 100-
•8 80 •
•6 60 •
-4 40 •
.2 20-
. n 0 •
)
"
/
) 24 48 72 96 120 144 168 0
r 10 100 -|
S

C2c-1

) 24 48 72 96 120 144 1C
G1-2

^— — -
•8 80 •
• 6 60 •
.4 40 •
-2 20 •
- n 0 •
58 C
•10 100-
•8 80 •
• 6 60 •
.4 40-
-2 20-
^ n •


^
)
1
) 24 48 72 96 120 144 168 0
r 10 100 -I
A2c-2
. 	

•8 80 •
• 6 60 •
.4 40 •
-2 20 •
n .


0 24 48 72 96 120 144 168 0
Time (h)
concerj
ie) and
^^^^^^H
2
~
2

tra
tin
•*•
4
/
4

tion tim
le-varia
^^^^^^H
4
^^^^^^H
4

e-
Dle
/
series (narrow line
exposures from 1
r*^^—
A2p-1

8 72 9
/
8


24 48
/
/


A1p-1

72 9
A2p-2


72 9
A2s-1

s) for
ests Al
10
8
6
4
2
n
6
10
8
6
4
2 ^!
6 Q_
poo,
o ^^^
C
• 6 O
•4 ^
-i— >
•0 °
6 0
O
MO ,_
(D
"8 "CO
• 6 ^
• 4
• 2
24 48 72 96
/" 10



A2s-4




/
/
• 8
• 6
• 4
• 2
24 48 72 96

-------
percentage mortality at any time cannot be calculated simply as the sum of the mortalities



observed up to that time divided by the starting number of organisms, even if this starting



number is reduced by the live samples up to that time. Rather, for the "ith" observation period, an



incremental fraction mortality (f\ii) is calculated as the deaths during that period divided by the



actual number of organisms present at the start of the period. Then, the cumulative fraction



mortality is calculated as l-n(l-fm~) (i.e., the Kaplan-Meier estimator).  Consider a test



starting with 20 organisms in which 4 died during each of the first two observation intervals and



8 survivors were sampled at the end of the first observation interval. For the first period the



incremental fraction mortality would be 0.20, based on the initial number of organisms. For the



second period, the incremental fraction mortality would be 0.50, based on this period starting



with 8 organisms after both the first period deaths and the sampling. The cumulative fraction



mortality would be 0.60 after the second period, in contrast to 0.67 based simply dividing the



cumulative deaths (8) by the starting number minus the live  samples (20-8=12).



       Figure 4.1  illustrates how mortality due to PCE has steep relationships to exposure



concentration and time. For the continuous exposures (left panels of Figure 4.1), the highest



concentration (Panel Gl-1) resulted in complete mortality within 14 h, but just a 15% drop in the



exposure concentration (A2c-l) resulted in complete mortality being delayed to  28 h, and an



additional 15% drop (C2c-l) caused mortality to  not reach 90% until 4 d and to not reach 100%



within 7 d. Further reduction to  about 5.0 mg PCE/L resulted in virtually no mortality over 96 h



(Panel A2c-2) and 48 h (Panel Gl-2), and for the latter exposure just slight increases in



concentration after 48 h was enough to cause 30% mortality to be reached at 7 d. Although not



shown in the figure, for continuous exposures in other experiments, mortality was never



appreciable at exposures concentrations less than 5 mg PCE/L and was generally 100% within



several days for exposure concentrations near and above 7 mg PCE/L.



       The time variable exposures (right panels of Figure 4.1) also reveal steep relationships of



mortality to time and concentration.  For a 6-h pulse to previously unexposed fish, mortality



exceeded 80% at 9.2 mg PCE/L (A2p-l), but was only 40% at 8.4 mg PCE/L (Alp-1),  12% at

-------
7.0 mg PCE/L (A2p-2), and absent at 5 mg PCE/L (not shown). After the first 6-h pulse, there



was no additional mortality for pulses of 6-7 mg PCE/L (Panel A2p-2), suggesting rapid



reduction in chemical accumulation and/or stress between pulses, such that later pulses could not



add enough stress to kill additional fish. Higher pulses (Panels Alp-1, A2p-l) did show some



incremental mortality after the first pulse, suggesting that such recovery between pulses is not



complete; however, this incremental mortality is also probably partly attributable to higher



concentrations in the later pulses. Rapid recovery from toxic stress is also indicated by the



absence of mortality during the intervals between the pulses.



       For the stepped exposures, mortality was nonexistent when exposure was 5 mg PCE/L or



less, and was  100% within 24 h once exposure was near or over 8 mg PCE/L (Panels A2s-l,



A2s-4 in Figure 4-1). These stepped exposures also suggest prior exposure does not increase



how rapidly mortality occurs once lethal exposures are imposed; rather, the opposite seems to be



true. For A2s-4, despite the fact that the prior exposure should create body burdens more than



halfway to lethal levels, mortality when the exposure is stepped up to over 9.5 mg PCE/L only



reached 25% in  6 h,  much lower than the 80% mortality for a 6-h pulse of similar concentration



in Treatment A2p-l, and the 50% mortality for the first 6-h of a lower concentration (7.8 mg



PCE/L) in Treatment A2s-l.



       However, although the relationship of mortality to exposure concentration and time is



obviously very steep, care should be taken not to infer too much about the exact relationships,



because of uncertainties in exposure concentrations (due to limited sampling, measurement error,



and/or time variability)  and variability in organism susceptibility. Because even a 10-20%



change in concentration appears  to have substantial effects on mortality in some cases, analytical



error of just the  same amount can confound results and make inferences about differences



between similar exposures uncertain. One example of possible uncertainties in exposure or



variability among experiments is that in exposure A2p-l, mortality reaches 80% in the first 6-h



pulse of 9.2 mg  PCE/L, whereas only 50% mortality is reached in the first 6 h in exposure Gl-1,



where the measured  concentration is actually slightly higher.






                                           7

-------
4.4 Kinetics of PCE Accumulation



       Figure 4.2 shows the results for four treatments in the bioconcentration experiment



(Experiment Bl). For each treatment, the dashed line shows the water concentration, which rises



within 2 h to at least 90% of its eventual values, and remains roughly constant until PCE input is



terminated, after which the concentration shows an exponential decline, dropping about 90% in



the subsequent 2 h. The PCE accumulations in sampled fish are shown by the filled circles. For



the lower three concentrations, there is a rather rapid rise in the first 4 h to about half of the



eventual value, and attainment of approximate steady state within 24 h. After the termination of



exposure, accumulation declines similarly - by about half in 4 h and by more than 95% after 24



h.  For the highest treatment concentration, fish exhibited toxic responses (lethargy,



disequilibria) starting around 4 h and were all dead by 24 h.  Because of this toxicity,



accumulation rate relative to the treatment concentration was slower than in the lower, nonlethal



treatment concentrations.



       The simple, single-compartment, first-order accumulation model (Equations 2.1, 2.2) was



parameterized based on the data from the lower three treatment concentrations in Figure 4.2.



Uptake and elimination rate constants were both assumed to vary among individual fish with a



log triangular distribution. Model parameters thus consisted of a mean and standard deviation



for both \og(ku) and log(fe). The accumulation model was integrated across the measured



exposure times-series (using linear interpolation between data points) to provide, for each time at



which accumulation was measured, a model-estimated accumulation as a function of model



parameter values. Mathematical search routines (see Section 3) were then used to determine the



parameter values that maximized the likelihood of the observed accumulations, resulting in



estimated means and standard deviations of 1.619 and 0.226 for \og(ku) and -0.560 and 0.010 for



log(fe).  This analysis thus inferred that the variability of the data was almost entirely due to



variation in the uptake constant among individuals. These parameter values correspond to a



median ku of 41.6 ml/g/h and median ks of 0.275/h.



       Figure 4.2 also shows the fit of the model to the data, using the parameter estimates to

-------
Figure 4.2.  Measured accumulation (solid circles) versus time for a 48 h exposure and 24 h depuration at four
concentration levels (dashed lines) in Test B1.  Solid line denotes median prediction of single compartment, first
order toxicokinetics model (median ku=4l.6 ml/g/h, median fe=0.275/h) fitted to data from lower three
concentration levels (highest level not used because of toxicity). Dashed lines denote 25th and 75th percentiles of
predicted variation among individual organisms.
                                         24        36         48
                                            Time (hours)
                                                                        60
                                                                                  72

-------
randomly generate 1000 pairs of ku and fe values and calculating the model-estimated



accumulation for each pair of parameters. The bold lines denote the median of these



accumulation estimates and the dotted lines denote the 25th and 75th percentiles.



       For the lower three treatment concentrations, the median line is consistently near the



middle of the data spread, except at 8 h into both the accumulation and elimination phases. This



suggests the toxicokinetics are more complicated than the single-compartment, first-order



behavior assumed in the model; for example, two-compartment kinetics would show slower



uptake after several hours due to the outer compartment approaching equilibrium, subsequent



uptake reflecting slower accumulation into an inner compartment (see Section 2.1.4).  However,



this deviation is relatively minor - at 8 h into accumulation, the deviation of the median model



line from the median data averages less than 15% - and will not be further considered here.



       For the highest treatment concentration, the model overestimates accumulation,



presumably due to the toxic effects on metabolic processes leading to reduced uptake. This



illustrates a potential problem in incorporating such accumulation information into time-



dependent effects models - exposures that are of interest because they cause effects will have



different kinetics of accumulation than lower exposures which are often used for  accumulation



studies. To account for this aspect of toxicity, a reduction in the accumulation model  parameters



would be needed as a function of accumulation.  Such possible refinements will not be made



here but rather be a subject for possible later work as deemed appropriate.



4.5 Relationship of Mortality to PCE Accumulation



       Figure 4.3 summarizes accumulation in both dead and surviving test organisms as a



function of time and treatment concentration for two tests (A2c, Gl) in which accumulation was



comprehensively documented. It is assumed that accumulation in dead organisms did not change



significantly between the times of death and sampling.



       For Test A2c, at the highest concentration (Treatment A2c-l), mortality was complete in



the first day and accumulation was measured in all the dead individuals (Figure 4.3).  On



average, accumulation increased with time-to-death, and at 24 h was especially variable. This






                                           10

-------
Figure 4.3 Measured PCE accumulation in dead (filled circles) and surviving (open circles) fish for two
concentrations levels in Tests A2c and Gl, contrasted with cumulative mortality (solid line).
100 7 --
80 •
H 6° •
jji
0 40 •
C 20-
CD
O n •
s_ u
CD
/I

_ v ^w
• ^/v •
^*^ •

1 III
Q_ 2 5 10 20
CD
> 100 7 •
•§ 80 •
£
D 60,
0 [
40 c
20 •
0 •
•t#
i • IMQ
If1
'To^B
/
2 5 10 20

- 9nnn -inn - , . - 9nnn
A2c-1





• i
• 1 000 80 •
• 500 60 •

• 200 40 '
•100 20-
cr\ 0 -
h 50 u |
A2c-2 Q
n
u



iii ii
•1000 ^
• 500 >
0)
• 200 Q
•100 ^1
O)
r 50 £
50 100 200 2 5 10 20 50 100 200 C
•;— ;
_onnn inn _ _ onnn m
G1-1





• 1 000 80 -
• 500 60 -
(
•200 40 I
•100 20*
™ n -
G1-2 fi fe a B

DQ-, O * QD ^^*
o o jT
• 1000 E
• 500 £3

-------
whole-body accumulation. Given the distribution of accumulation observed for dead individuals



in Treatment A2c-l, substantial mortality should have been observed in Treatment A2c-2.



       Although not presented here, the lipid content of each fish was measured and the same



problem - of lower PCE accumulation in dead organisms at a high exposure concentration than



in surviving organisms exposed to a lower concentration - is observed for lipid-normalized



whole-body accumulation values. This does not contradict the fundamental importance of



accumulation to toxicity, but rather  suggests that whole-body accumulation is not a good



measure of the accumulation relevant to toxicity. A likely contributing factor to this problem is



that toxicity should be related to specific compartments in the individual and  accumulation in



such compartments might have a different time course than whole-body accumulation.  For



example, PCE in the circulating blood and well-perfused tissues would respond more quickly to



water concentrations, whereas total  body accumulation, especially after substantial time, might



mainly reflect chemical partitioned into lexicologically less important tissues. A more



complicated toxicokinetic model and a toxicodynamics model focused on a specific



compartment would thus be necessary to better relate mortality, accumulation, and exposure



(e.g., Section 2.1.4), but such a model cannot be adequately supported with the data obtained in



these experiments.



       The results in Test Gl (Figure 4.3) reinforce these difficulties in relating accumulation to



death in the toxicity models being considered. Although it must be acknowledged that  some of



the variability in this data is analytical uncertainty, there is clearly a large amount of variability



among individuals, which is inconsistent with the stochastic model. But if just average



accumulations are considered, the fact that the survivors in the later stages of Treatment Gl-2



have accumulation greater than the individuals that died in this exposure also is incompatible



with the tenets of the model. For the deterministic model, this is also a problem, because it is



statistically improbable for the accumulation in the dead organisms to be so much lower than that



of the randomly selected surviving organisms.  Again, this does not belie the importance of



accumulation, but rather only the utility of the whole-body accumulation monitored in these






                                           12

-------
experiments.



       Despite these problems regarding the application of these accumulation measurements to



the "explicit" forms of the toxicity models, these data can still be used to estimate model



parameters, and thus be used to evaluate the implications of these problems to model predictions.



       For the deterministic model, the desired parameters are the mean and variability of the



distribution of lethal accumulation thresholds.  This cannot simply be the mean and standard



deviation of accumulation in dead organisms (unless the data set consists of exposures in which



all the individuals died), because surviving organisms also have useful information in that they



establish minimum values for their lethal accumulations.  The data in Figure 4.3 were therefore



subjected to maximum likelihood analyses to estimate the distribution of lethal accumulations, in



which (a) the likelihood of the accumulation measured in a dead individual is the frequency of



that accumulation value within the distribution and (b) the likelihood of the accumulation



measured in a live individual is 1.0 minus the cumulative probability of that accumulation value



within the distribution. For a pooled analysis of all the data in Figure 4.3, this resulted in a mean



and standard deviation for the log(LA) of 2.79 and 0.39. Analyses on the separate tests produced



similar values - 2.68 and 0.34 for Test Ale and 2.86 and 0.42 for Test Gl.  Analyses on just the



treatments in which mortality was complete produced slightly lower means and standard



deviations - 2.56 and 0.30 for Treatment Alc-1 and 2.65 and 0.22 for Treatment Gl-1.



       For the stochastic model, the toxicodynamic parameters to estimate from the



accumulation data include the lethal accumulation threshold AO and the killing rate d (Equation



2.26). To estimate these parameters, it is necessary to first estimate, for the accumulation



measured in each individual, the time-series of the accumulation leading up to that measurement,



because this model uses the whole times-series to determine the integrated probability of death



(Equations 2.22-2.26).  As already noted, the fact that accumulation varies so much among



individuals is inconsistent with the tenets of the stochastic model, but the desired times-series



just depends on the value which was inferred (Section 4.4 above) to not be the source of this



variability, so that the median value for ke of 0.275/h could be used to generate these times-series






                                            13

-------
for illustrative purposes. Using such accumulation times-series for each individual, the



probability of death can be determined as a function of AO and d, and a likelihood can thus be



computed for the observed combination of dead and surviving organisms with their respective



times and accumulations. However, the problems already noted in the accumulation  data



resulted in such an analysis inferring that this likelihood was maximized for a value for AO of 0



(no threshold for effects) with a value for d of 0.000027/h/(mg PCE/g wwt).  Constraining AO to



be 100 mg PCE/g wwt (lower than all but a few of the measured accumulations in dead



organisms, Figure 4.3) and estimating just d resulted in a slightly higher value for d of



0.000032/h/(mg PCE/g wwt), which will be used in model calculations below.



4.6 Application of Implicit Deterministic Mortality Model



4.6.1 Model Parameterization



       One consequence of both (a) the steep relationships of mortality to time and



concentration and (b) the impact of exposure concentration uncertainties on interpreting these



changes is that this data set does not have the information to support consideration of models of



lethality more complex than the simplest ones in Section 2. In fact, this data set provides a good



test of whether even the simplest model can be parameterized with limited information.



       Mortality data from continuous exposures in Tests A2, C2, C3, C4, and Gl (in which the



highest concentrations caused sufficient mortality to support model parameterization), were used



to parameterize model Dl as described in Sections 2 and 3 of part A of this report, with no



consideration being needed for how to treat delayed mortality, which was absent in these PCE



tests.  Parameters were estimated based on data from each individual test and on the pooled data



from tests Al, C4, and Gl  (in which mortality was monitored multiple times during the 12 h of



exposure, whereas in  tests C2 and C3 mortality was not monitored until 24 h). Table 4.1



summarizes the estimated parameter values from all six parameterizations.



       The parameterization using the individual tests involves a small amount of information -



often just two concentrations with mortality being complete in a short time at the higher



concentration and being slight or absent at the lower concentration. Nevertheless, parameter






                                           14

-------
Table 4. 1 . Maximum likelihood estimates for deterministic model Dl parameters (implicit version).
Parentheses denote standard error of estimate.
Tests Used
for Parameter
Estimation
A2, G1,C4
A2c
Gl
C2c
C3c
C4c
Median
Parameter Value
LCoo
(mg PCE/L)
6.15
5.65
5.84
6.67
6.65
6.05
k
(1/hr)
0.160
0.155
0.141
0.136
0.104
0.167
Mean
logio Parameter Value
LCoo
0.789
(0.006)
0.752
0.766
0.824
0.823
0.782
k
-0.796
(0.027)
-0.810
-0.852
-0.862
-0.984
-0.776
Standard Deviation
logioParameter Value
LCoo
0.058
(0.003)
0.020*
0.023
0.020*
0.022
0.062
k
0.322
(0.021)
0.295
0.224
0.020*
0.020*
0.437
* minimum allowed log standard deviation = 0.02
estimates were successfully derived from each set, although for some sets a minimum variability



among individuals was imposed on LC^ or k (this being a minimum standard deviation of the log



parameter = 0.02, so that the maximum parameter value was at least 25% higher than the



minimum). Despite the limited data, the median parameter estimates varied across the sets only



from 5.65 to 6.05 mg PCE/L for LC^ and 0.141 to 0.167/h for k (Table 4.1) for the three data sets



with early mortality information. For the tests without early mortality observations (C2c and



C3c) parameter estimates were also close to these ranges, but were higher for the LC^ and lower



for &£, which raises the possibility that predicting mortality at short times will be uncertain based



on such data sets.



4.6.2 Model Performance Using Parameterizatiom Based on Tests A2c, Gl, and C4c.



       For parameterizations based on tests in which early mortality observations  were made



(A2, Gl,  and C4), there were good fits to the mortality observed in the continuous exposures



(left panels of Figure 4.4). For the high exposure concentrations (A2c-l and Gl-1), the model



predicts rapid toxicity, including substantially more rapid toxicity when the exposure



concentration is only slightly higher in Gl-1 compared to A2c-l. The observed mortality falls
                                           15

-------
Figure 4. 4. C
time-series (n
exposures fro
dashed lines <
std. dev. for s
model param
100-
80 •
60-
40 •
20
0 j
C
100 •
80'
60 •
40-
20 •
>? 0-
o 100.
^ 80.
0 60'
L^ 40.
0
Q_ 20'
•^ I
^ 100.
13 80 •
60.
40 •
20.
0 •
C
100 •
80'
60 •
40'
20 •
)bserved mortality (bold solid lines) an
arrow solid lines) for continuous expos
m Tests Al and A2 (right side). All pi
ienote pooled model parameterizations
eparate model parameterizations using
sterizations using 2 other tests without
/ G1-1
I
24 48 72 96 120 144 16
-j£t
1 A2c-1

d predicted n
sures from Tt
edictions are
using tests P
these three t
early mortali
10 100i
8 80 -H
6 60 |l
4 40^
2 20*/
n 0 1
>8 C
- 1 0 1 00 •
-8 80-
- 6 60 •
.4 40-
.2 20-
n n •
I
) 24 48 72 96 120 144 168 0
i- 10 1 00 -|

IT 	
// C2c-1
r •
•8 80-
• 6 60 •
.4 40 •
-2 20-
n f) .


A
r.
) 24 48 72 96 120 144 168 0
.-10 1 00 ,
G1-2

^,r-*=~~ -
• 8 80 •
• 6 60-
.4 40-
-2 20 •
n n •


) 24 48 72 96 120 144 168 0
i-10 1 00 -I
A2c-2
	

-8 80-
- 6 60 •
.4 40 •
-2 20 •
n .

0 24 48 72 96 120 144 168 0
Time (h)
lortality as a function of water conce
;sts A2, Gl, and C2 (left side) and tii
for the implicit deterministic model
^2c, Gl, C4c and gray band denotes i
ssts. Dotted lines denote mean of ad
ty observations (C2c, C3c).
....
2
2

••••••HI!
i •
4 4
i~.
4 4


" A2p-1

.
8 72 9
/ 	
— A1p-1-»

8 72 9
A2p-2


ntration
ne-variable
Dl. Bold
nean ± 1
ditional
• 10
• 8
•6
• 4
•2
• 0
6
• 10
•8
• 6
• 4
2 ^
og
6 Q_
[1°E
0 -^^^ff
C
• 6 O
2 0
n O
u c
24 48 72 96 O

r :

A2s-1

"8 |
•6 >
• 4
•2
24 48 72 96


A2s-4


'/

J/
•8
• 6
• 4
• 2
24 48 72 96
16

-------
within 1.0 standard deviation of the mean prediction based on the individual parameterizations.



On average, the model predicts the attainment of 90-100% mortality to be slower than observed,



but uncertainties in such high mortality would be of little importance to risk assessments.



       When exposure is reduced to about 7 mg PCE/L (C2c-l), the model predicts well the



degree to which mortality slows down, with the observed mortality again bracketed by the



uncertainty of predictions based on the individual parameterizations. For the pooled



parameterization, the eventual degree of predicted mortality is 15% low, further illustrating the



underestimation of high mortality already noted. This underestimation of mortality for the more



tolerant individuals can occur due to overestimation of the variability of Co and/or ks among



individuals, resulting in a subset of organisms estimated to be more tolerant than appropriate.



For the pooled parameterization, this can occur due to variation across experiments being



interpreted by the model as variability across individuals.



       Consistent with the observed mortality, model-estimated mortality for exposure Gl-2



shows little mortality at early times when exposures are near 5 mg PCE/L and also shows



moderate mortality as exposure lengthens and concentrations increase to about 5.5  mg PCE/L.



For the parameterizations based on the individual data sets, substantial variability exists, but this



is to be expected because the steepness of the response relationships causes small uncertainties in



exposure concentrations and model parameters to have large effects on partial mortality



predictions. For exposure A2c-2, the model predicts little or no mortality when exposure



concentrations remain near 5 mg PCE/L, consistent with what was observed.



       Large uncertainties of effects at concentrations in a steep section of the effects versus



concentration curve are to be expected and are not generally  of concern for risk assessments, for



which the uncertainty of the concentration causing a particular effect is of more interest. With



regard to this, the estimated effect concentration (EC) ranges across the different



parameterizations are very small - 5.7 to 6.2 mg PCE/L for the 96-h ECso, 5.0-5.4 mg PCE/L for



the 96-h ECio, and 5.9-6.6 for the 24-h EC5o - despite the limited data.



       Although these good fits to the continuous exposures indicate that the model captures the






                                           17

-------
important features of the relationship of mortality to PCE concentration and exposure duration,



model predictions for the pulsed and stepped exposures in the right panels of Figure 4.4 are also



important for establishing the merits of the model.  The model underestimates mortality for the



most intense pulse exposure (A2p-l), even in the first pulse, by nearly a factor of 2.0.  This is



partly due to the underestimation already noted regarding the quickness of mortality for the more



tolerant organisms, but also probably reflects the uncertainty issues also noted regarding steep



toxicity relationships.  For exposure Alp-1, the prediction for the first pulse is good, but the



model fails to predict the observed incremental mortality in subsequent pulses. One possible



reason for this is that the single kinetic constant in the model can cause an overestimation of the



degree of recovery between pulses for this intense exposure (e.g., >90% reduction in



accumulation and/or damage).  In contrast, for the less intense pulse of A2p-2,  the model does



predict well both the degree of mortality in the first pulse and the absence of mortality in



subsequent pulses.



       Again, data presentations such as Figure 4.4 highlight uncertainties in effects at a



particular exposure concentration, which are informative, but typically of less interest to risk



assessments than the uncertainties in effect concentrations. For a single 6-h pulse, the average



predicted EC50 is 9.8 mg PCE/L, whereas the observed EC50 is 8.5 mg PCE/L, just 13% lower.



For four daily pulses, the observed EC50 is 7.5 mg PCE/L based on average pulse



concentrations, 20% less than the average predicted EC50 of 9.4 mg PCE/L. Thus, the model



underestimation of the risk of pulsed exposures is relatively minor.



       For the stepped exposures, the good predictions for A2s-l are  expected because this is no



different than the first day  of the continuous exposures already discussed. The model does



predict well that there will  be total mortality when exposures are stepped up to lethal levels on



later days after previous nonlethal exposures (A2s-4, and also A2s-2 and A2s-3, not shown, in



which lethal exposures were on the second and third day). However,  the slower rate of mortality



in such incrementally increasing exposures, that appears to come from the prior nonlethal



exposure, is not predicted.  This suggests some importance of toxicokinetic or toxicodynamic






                                           18

-------
processes that are not included in the model. However, again, these have relatively little



importance to the exposure concentrations predicted to be lethal for any particular exposure



times-series shape.



4.6.3 Model Performance Using Parameterizations Based on Tests C2c and C3c.



       When the model is parameterized based on tests in which early mortality is not



monitored, predictions are worse. For continuous exposures with high concentrations, the



average predicted mortality for parameterizations based on Tests C2 and C3 is delayed



substantially from that observed (Gl-1, A2c-l, C2c-l, A2s-l in Figure 4.4). At the lower



exposures of 5.0-5.6 mg PCE/L, the moderate mortality observed in Treatment Gl-2 is not



predicted at all, despite this occurring over a timeframe consistent with the observations in C2



and C3 used to parameterize the model.  Little or no mortality is predicted for the pulsed



exposures (A2p-l, Alp-1, A2p-2 in Figure 4.4). This emphasizes the need for at least some



mortality information over a broad range of timeframes if these models are to be effective.



4.7 Application of Implicit Stochastic Mortality Model



4.7.1 Model Parameterization



       As for the deterministic model, mortality data from continuous exposures in Tests A2c,



C2c, C3c, C4c, and Gl were used to parameterize the implicit version of stochastic model SI, as



previously described in Sections 2 and 3 in Part A of this report.  Again,  parameters were



estimated based on data from each individual test and on  the pooled data from Tests Al, C4, and



Gl. Table 4.2 summarizes the estimated parameter values from all six parameterizations.



Although parameter estimates were successfully obtained, the estimates for fe in four of the



parameterizations were at a maximum allowed value (1.0/h, beyond which this parameter has no



practical consequences), about 4-fold greater than the value estimated directly from



bioaccumulation data in Section 4.4, and were more than 2-fold higher or lower in the other two



parameterizations. This difficulty in the model parameterization reflects the fact that both fe and



d represent aspects of the kinetics of the toxicity response (fe regards accumulation of chemical



or damage and d addresses the mortality rate for a given accumulation), and the data must






                                           19

-------
Table 4.2. Maximum likelihood estimates for stochastic model SI parameters (implicit version).
Tests Used
for Parameter
Estimation
A2, G1,C4
A2c
Gl
C2c
C3c
C4c
Parameter Value
C0
5.25
5.14
5.28
6.35
6.58
4.68
d
0.030
0.040
0.040
0.054
0.079
0.014
fe
1.0*
1.0*
0.604
1.0*
0.3*
1.0*
logioParameter Value
CO
0.718
0.711
0.723
0.803
0.719
0.670
d
-1.528
-1.401
-1.399
-1.270
-1.596
-1.848
fe
0.000*
0.000*
-0.219
0.000*
-0.898*
0.000*
* log mean k constrained to be <0.0
contain at least some indication of two-phase kinetics for mortality for the model to partition the



kinetics between these two processes. That ks is so much greater than the known kinetics of



accumulation is indicative of the model parameters not actually describing what they purport to.



4.7.2 Model Performance



       When parameterized using Tests Gl, A2c, and/or C4c, stochastic model mortality



estimates show both limitations and merits of the model. As for the deterministic model, the



rapid, high mortality in Treatments Gl-1  and A2c-l is predicted well (Figure 4.5), although with



some underestimation of the death rate as mortality nears 100%.  The model also predicts the



slower and approximately complete toxicity for Treatment C2c-l. However, it greatly



overestimates the mortality in Treatment  Gl-2.  This is due to this model assuming no



differences in sensitivity among individuals, so that once one individual dies, all the others will



eventually die if exposure does not decrease below the threshold. Regarding this, the apparent



leveling off of mortality for Treatment A2c-2 at the upper range of the predictions is due to such



declining exposure.  That this model lacks consideration of differences among organisms and



treats partial mortality only as an issue of insufficient time to reach complete mortality is a



limitation, and results in vanishingly small differences in ECs for different effect levels (an "all
                                           20

-------
Figure 4. 5. C
time-series (n
exposures fro
dashed lines <
std. dev. for s
model param
100 •
80.
60 •
40-
20 •
0-
C
100 •
80'
60'
40 •
20'
>> Q.
o 100.
^ 80 •
-!-•
0 60.
i^ 40 •
0
Q_ 20'
'•^
^ 100.
13 80 •
60'
40.
20 •
0 •
(
100'
80.
60 •
40.
20 •
Q.
(
)bserved mortality (bold solid lines) an
arrow solid lines) for continuous expos
m Tests Al and A2 (right side). All pi
ienote pooled model parameterizations
eparate model parameterizations using
sterizations using two other tests witho
/ G1-1
f
) 24 48 72 96 120 144 16
/ A2c-1
i
3 24 48 72 96 120 144 1f
^.. 	
/ .-'
// C2c-1
(••
3 24 48 72 96 120 144 1«
G1-2
,»••
^^^
d predictec
sures from
edictions £
using test
these thres
ut early m(
10 100-
8 80-
6 60 •
4 40-
2 20-
n 0 '
>8 C
•10 100-
•8 80-
• 6 60 •
.4 40 •
-2 20-
• n 0 •
58 C
•10 100-
• 8 80 •
•6 60-
.4 40 •
-2 20-
• 0 0 •
58 C
•10 100-
• 8 80 •
• 6 60 •
.4 40-
.2 20-
n .
. mortality
Tests A2
ire for the
3 A2c, Gl
3 tests. D
jrtality ob
F
) 2
^^
£z
r as a function of water conce
Gl, and C2 (left side) and th
implicit stochastic model SI
, C4c and gray band denotes
otted lines denote mean of ad
servations (C2c, C3c).
*» ^^^~
i
t
4 4

t
A2p-1

8 72 9
t r -
-•"A1p-1 	

) 24 48 72 9


ft
»• 	



A2p-2
^— -*'""'


) 24 48 72 9
^^
A
/I:
V

A2s-1

ntration
ne-variable
Bold
mean ± 1
ditional
• 10
•8
• 6
• 4
•2
• 0
6
• 10
•8
•6
• 4
2 ^
6 Q_
• 10 g>
O ^.^fi'
C
-6 O
. 4 (0
2 0
• n ^
6 0
O
r 10 j_
JU
•6 >
• 4
• 2
i= 	 1 	 1 	 1 	 1 	 1 	 1 	 r u - i 	 1 	 1 	 1 	 r u
3 24 48 72 96 120 144 168 0 24 48 72 96
i-10 100 i _jf 10
A2c-2
	

• 8 80 •
• 6 60 •
.4 40-
-2 20 •


A2s-4


V
if
I
•8
• 6
• 4
• 2
3 24 48 72 96 120 144 168 0 24 48 72 96
Time (h)
21

-------
or nothing" response) for prolonged exposures with constant, or regularly fluctuating,



concentrations. However, for the PCE toxicity of concern here, this results in little error because



the steep toxicity relationships of PCE already result in a narrow EC range. The ECs estimated



using the stochastic model are similar to those for the deterministic model and do not vary much



among the different parameterizations, being 5.3-5.5 mg PCE/L for the 96-h ECso, 4.8-5.3 mg



PCE/L for the 96-h ECio, and 6.0-6.9 for the 24-h EC50



       Regarding the pulsed exposures, like the deterministic model, the stochastic model



underestimates mortality in the first pulse at the high exposure in Treatment A2p-l.  Unlike the



deterministic model, the stochastic model does predict substantial incremental mortality across



multiple pulses, which results in good predictions for Treatment Alp-1, but incorrectly predicts



substantial mortality after the first pulse for Treatment A2p-2, where none was observed.



Overall, this incremental mortality predicted by the stochastic model is of questionable merit



because it reflects again that this model assumes all organisms have the same susceptibility and



that the partial mortality in an initial pulse just reflects the finite probability within a finite time



of any individual dying; thus, those individuals that survive the first pulse will eventually all



succumb to subsequent pulses of sufficient magnitude.  It thus does not allow for partial



mortality that persists across  pulses, as is evident in Treatment A2p-2 in Figure 4.5 and was also



evident for copper in Section 3.  However, this does not necessarily mean that this model  does



not provide mortality estimates that would still be useful in risk assessments.  For a single 6-h



pulse, the stochastic model estimates an ECso (11.0 mg PCE/L) only 30% greater than what was



observed (8.5 mg PCE/L), and, for four daily pulses, an ECso of 6.9 mg PCE/L, within 10% of



what was observed (7.5 mg PCE/L).



       For the stepped exposures, the stochastic model successfully predicts that mortality will



be heavy when exposure increases from about 5 mg PCE/L to about 9 mg PCE/L, and it better



predicts the rate of mortality  during the lethal step because the parameter d makes this rate less



sensitive to prior exposure than for the deterministic model. However, like the deterministic



model, it does not address how prior exposure might actually slow the mortality rate, as was






                                           22

-------
apparent in the data.



       As for the deterministic model, in most cases, predictions are worse when the stochastic



model is parameterized with tests lacking mortality observations at early times (Figure 4.5). This



again emphasizes that these toxicity models require information on mortality across a reasonable



broad range of time scale.  Although the required data would require only modest additional



effort in standard tests and already is often collected, this does create some problems in



exploiting reported test results, which do not typically provide results across multiple exposure



times, even when these data are collected.



4.8 Application of Explicit Deterministic Mortality Model



       The parameter estimates needed for application of the version of deterministic mortality



model Dl  that explicitly addresses accumulation were already reported in Sections 4.4 and 4.5.



For toxicokinetics, the estimated means and standard deviations were 1.619 and 0.226 for \og(ku)



and -0.560 and 0.010 for log(fe), where the units of ku are ml/g/h and those of ks 1/h. For



toxicodynamics, log(,L4) was estimated based on the pooled data from Treatments A2c-l, A2c-2,



Gl-1, and Gl-2 to have a mean 2.79 and a standard deviation of 0.39, where the units of LA are



jig PCE/g wwt. With these parameter estimates, model predictions were made based on



generating 1000 sets of values for ku, fe and LA and calculating fraction mortality as a function



of time and exposure concentration based on  1000 model simulations using these sets.



       The performance for this explicit implementation of the deterministic model is poor



(Figure 4.6). For the continuous exposures, mortality is underestimated at high concentrations



and overestimated at low concentrations. This poor performance is also apparent in the time-



variable exposures on the right side of Figure 4.6, with mortality from pulsed Treatments Alp-1



and A2p-2 being overestimated. Problems with model estimates are particularly evident for



Treatment A2s-4, where substantial mortality is predicted in the earlier steps, where no mortality



was observed.



       This poor performance reflects the problems noted earlier in Sections 4.4 and 4.5:



specifically, the wide variability of model parameters resulting in predictions of a wide range of






                                           23

-------
Figure 4.6. Observed mortality (bold solid lines) and predicted mortality (bold dashed line) as a function of
water concentration time-series (narrow solid lines) for continuous exposures from Tests A2, Gl, and C2 (left
side) and time-variable exposures from Tests Al and A2 (right side).  All predictions are for the explicit version
of deterministic model Dl and use toxicokinetics parameters from Experiment Bl and lethal accumulations
estimated from the pooled data of Tests A2c and Gl.
r 10  100
      80
      60
      40
      20
       0

r 10  100

1
1
f















A2


P


-2





- 8
- 6
- 4

                                                                                                 O)
                                                                                                 -—
                                                                                                 §
                                                                24
         48
         72
              24   48   72   96   120  144  168
     100
      80
      60
      40
      20
       0


/—
I

A2c-2






r 10 100
- 8 80
- 6 60

- 4 40
. 2 20
.n 0
         0    24   48   72   96   120  144  168
24
48
                                    72
96
                                              Time (h)
                                                 24

-------
sensitivity of individuals to specific exposures, so that some individuals survive the high



exposures and some succumb to lower exposures, contrary to what was observed. This does not



belie the importance of accumulation to toxicity, but rather that these relationships are more



complex than expressed in these models, and that this oversimplification can result in erroneous



predictions, especially in the face of various uncertainties in parameterization data (e.g.,



accumulation measurements).



4.9 Application of Explicit Stochastic Mortality Model



       The explicit form of the stochastic mortality model can be implemented based on the



median ku (41.6 ml/g/h) and  fe (0.275/h) estimated in Section 4.4, and the values specified in



Section 4.5 of 100 |ig PCE/g for the lethal accumulation threshold and 0.000032/h/(mg PCE/g



wwt) for the killing rate d. Figure 4.7 provides the resultant model estimates of mortality time-



series.  Although the mortality patterns are much different than for the explicit deterministic



model, there is again  consistent underestimation of mortality for the higher exposures and



overestimation for lower exposures.



       The poor model performance again is due to problems already noted regarding the



relationship of mortality to accumulation and the large variability among individuals  in the



measured accumulation and in the relationship of mortality to accumulation. As already noted,



such variability is inherently inconsistent with the stochastic model.  Using average parameter



estimates based on variable data results in this model estimating mortality to occur at much



lower concentrations  than it actually does and for the mortality at high concentrations to occur



more slowly than  observed.  Again, this does not refute the basic concepts of the model, but



rather reflects errors that can occur when an overly simple model formulation is combined with



uncertain data used for parameterization.
                                           25

-------
Figure 4.7.  Observed mortality (bold solid lines) and predicted mortality (bold dashed lines) as a function of
water concentration time-series (narrow solid lines) for continuous exposures from Tests A2, Gl, and C2 (left
side) and time-variable exposures from Tests Al and A2 (right side).  All predictions are for the explicit version
of stochastic model SI and use median toxicokinetics parameters from Experiment Bl and a killing rate
estimated from pooled data of Tests A2c and Gl, using a lethal accumulation threshold of 100 |j,g PCE/g wwt.
 (D
 (D
Q_
 0)
^
 ^
 E
 ^
O
         0   24    48    72   96   120  144  168
         0    24   48   72   96   120  144  168

k

?^^m
i— i



.*

,«•

A2p-2
.__

.*'

	

• 10 0)
o s,^^
c
-6 g
• 4 2
-1— >
2 0)
.n 0
         0    24   48   72   96   120  144  168
24
48
72
96
         0    24   48   72   96   120  144  168
     100
      80
      60
      40
      20
       0
         0    24   48   72   96   120  144  168
                                              Time (h)
                                                 26

-------
4.10 Summary and Implications to Aquatic Life Criteria



       This analysis of mortality of juvenile fathead minnows exposed to pentachloroethane has



further demonstrated that the toxicity models discussed in Section 2, when used in their



"implicit" form and parameterized directly on the observed relationships between the time-series



of mortality and continuous water exposure, can effectively describe these relationships and



make useful predictions regarding time-variable exposures. These models are simple depictions



of the toxicity processes and thus do not fully describe the mortality relationships. However,



when direct data are lacking, they provide estimates for the effects of exposures that are accurate



enough to be of use in aquatic life criteria applications and other aquatic risk assessments where



information on the relationship of magnitude of effects to different exposure time-series is



needed.



       However, although these models rely on theoretical relationships of effects to



accumulation, using versions of these models explicitly based on relationships of accumulation



to exposure and mortality to accumulation did not provide good predictions for this case study.



This was likely due to an oversimplification of the relationship of mortality to accumulation and



to uncertainties and variability of data used in model parameterization.  This does not argue



against the importance of accumulation in toxicity relationships or in the utility of accumulation



for more simple risk assessments.  However, for predicting magnitude of toxicity across various



exposure times-series, better models and data are needed.



       It should finally be emphasized that the data and analyses of this section only relate to



mortality in an acute timeframe  and to toxicity mechanisms that operate with this timeframe.



Actual criteria applications regarding even just mortality would need to also  consider longer-



term survival data, including the possibility of different mechanisms operating in different



timeframes. Furthermore, the analyses here would be most relevant to highly variable exposure



situations in which transient high exposures would make acute mortality relevant.  For other



exposures scenarios, chronic survival and other endpoints would be more of a concern.



Subsequent reports in this series will be addressing longer exposures.






                                           27

-------
4.11 References
Erickson R, Kleiner C, Fiandt J, Highland T. 1991. Use of toxicity models to reduce uncertainty
    in aquatic hazard assessments: Effects of exposure conditions on pentachloroethane toxicity
    to fathead minnows. Internal report, U.S. Environmental Protection Agency, Mid-Continent
    Ecology Division, Duluth, MN, USA. 37 p.

Kaplan, E.L., Meier, P. 1958. Nonparameteric estimation from incomplete observations. J Am
    StatAssoc 53:457-481.
                                           28

-------