United States
Environmental Protection
Agency
Office of Pesticides and
Toxic Substances
Washington DC 20460
EPA 560 681 003
January 1981
Toxic Substances
Comparison of Static-
Replacement and Flow
Through Bioassays
Using Duckweed,
Lemna gibba G-3
-------
This document is available to the public through the National Technical
Information Service, Springfield, Virginia 22161.
-------
EPA 560/6-81-003
January 1981
COMPARISON OF STATIC-REPLACEMENT
AND FLOW-THROUGH BIOASSAYS
USING DUCKWEED, LEMNA GIBBA G-3
by
John A. Davis
Breedlove Associates, Inc.
Gainesville, Florida 32601
Project Officer
David A. Mauriello
Office of Pesticides and Toxic Substances
Washington, D.C. 20460
U.S. Environmental Protection Agency
Office of Pesticides and Toxic Substances
Washington, D.C. 20460
-------
DISCLAIMER
This report has been reviewed by the Office of Environ-
mental Processes and Effects Research, U.S. Environmental
Protection Agency, Washington, D.C., and approved for
publication. Approval does not signify that the contents
necessarily reflect the views and policies of the U.S.
Environmental Protection Agency, nor does mention of
trade names or commercial products constitute endorse-
ment or recommendation for use.
11
-------
ABSTRACT
Static-replacement and flow-through tests were con-
ducted using CuSO^-SHaO, 2,4,6-trichlorophenol, and o-cresol
to determine if they gave similar LCSO's and EC50's for
duckweed, Lemna gibba G-3. Static-replacement tests also
were conducted using ethylene glycol and di(2-ethylhexyl)
phthalate. Mortality, reproduction, dry weight, and root
length were used to measure effect levels of the toxicants.
LCSO's and ECSO's were calculated using quadratic regression
with log transformation of the independent variable (concen-
tration) and with the following transformations for the de-
pendent variables: arc sin square root of the proportion
(p) of dead to total fronds on Day 7 (mortality), a log
function yielding a growth rate constant K (reproduction),
log!o dry weight and arc sin square root of the ratio of
dry weight to control weight (dry weight), and logio of
root length. ANOVA's were used to test for differences
between the two types of tests, tests within types, and
replicates within tests. A procedure also was provided
for estimating the number of tests and replicates neces-
sary to obtain confidence limits within a given percentage
of the mean.
Of the four effect parameters, mortality and reproduc-
tion produced the best results. The results generally in-
dicated that the highest variation occurred among tests,
regardless of type, and that the smallest variation was
generally within tests (i.e. among replicates). Therefore,
the conclusion was that the best allocation of resources
would be to replicate static-replacement tests in time with
the number of replicates dependent on the toxicant. Gener-
ally, four replicates should be used if no information is
available on the expected variation within tests. The in-
formation gained can then be used to statistically deter-
mine the number of tests and replicates necessary to obtain
given confidence limits and probability levels.
Key words: bioassay, duckweed, Lemna gibba G-3, aquatic
toxicology, copper sulfate, 2,4,6-trichloro-
phenol, o-cresol, ethylene glycol, di(2-ethyl-
hexyl) phthalate
111
-------
IV
-------
TABLE OF CONTENTS
Page
ABSTRACT iii
LIST OF TABLES vii
LIST OF FIGURES xi
1.0 INTRODUCTION 1
2. 0 SUMMARY 3
3. 0 CONCLUSIONS 7
4. 0 RECOMMENDATIONS 9
5. 0 LABORATORY MATERIALS AND METHODS H
5.1 Selection of Test Species 11
5. 2 Selection of Toxicants 11
5. 3 Culture Methods 11
5. 4 Flow-through Tests 15
5.5 Static-Replacement Tests 23
5. 6 Laboratory Conditions 24
5. 7 Scheduling of Tests 24
6. 0 STATISTICAL METHODS 27
6.1 Mortality Data Analysis 27
6.2 Reproduction Data Analysis 33
6.3 Dry Weight Data Analysis 34
6. 4 Root Length Data Analysis 34
6. 5 Comparative Methods 34
7. 0 MORTALITY DATA ANALYSIS 37
7.1 CuSO^ 41
7.2 2,4,6-TCP 48
7.3 0-Cresol 51
7. 4 Ethylene Glycol 53
7.5 Di(2-ethylhexyl)phthalate 53
8. 0 REPRODUCTION DATA ANALYSIS 57
8.1 CuSOt* 57
8.2 2,4,6-TCP 61
8.3 0-Cresol 64
8.4 Ethy lene Glycol 68
8.5 Di(2-ethylhexyl)phthalate 68
9.0 DRY WEIGHT DATA ANALYSIS 73
9.1 CuSO^ 73
9.2 2,4,6-TCP 74
9.3 0-Cresol 74
9. 4 Ethylene Glycol 76
9.5 Di(2-ethylhexyl)phthalate 76
v
-------
TABLE OF CONTENTS (Continued)
10 . 0 ROOT LENGTH DATA ANALYSIS 79
11.0 COST ANALYSIS 81
11.1 Culture Costs 81
11.2 Static-Replacement Test Costs 81
11.3 Flow-through Test Costs 82
11.4 Cost Comparisons 82
12.0 COMPARISON OF MORTALITY, REPRODUCTION,
DRY WEIGHT, AND ROOT LENGTH DATA 87
13.0 LITERATURE CITED 91
VI
-------
LIST OF FIGURES
Page
Figure 5-1 EPA-type flow-through dilutor system 16
Figure 7-1 CuSOi* LCSO's and confidence limits using
inverse, probit maximum likelihood (MLH),
and quadratic regression with arc sin
square root transformation of dependent
variables 40
xi
-------
Xll
-------
1.0 INTRODUCTION
The Office of Pesticides and Toxic Substances of the
U.S. Environmental Protection Agency (EPA) is contemplat-
ing the development of a test standard, using duckweed as
a test organism, for assessing the toxicity of certain
wastes. To facilitate development of the protocol, infor
mation on the precision and design of static and flow-
through bioassay tests was needed. Lemna gibba G-3 was
chosen as the test organism because of its taxonomic sta-
bility and previous work on its physiology. Moreover, it
is small, planktonic, easily cultured, and reproduces veg
etatively, allowing development of a homogeneous clone.
The objectives of this study were
(1) to compare static-replacement bioassays with
flow- through bioassays in terms of cost, pre-
cision, and comparability of LCSO's and ECSO's;
(2) to ascertain the optimum distribution of
replicates and tests for achieving certain
levels of significance; and
(3) to determine the LCSO's and ECSO's (based on
dry weight, reproduction, and root length)
for five compounds :
a) copper sulfate
b) 2,4,6-trichlorophenol (2,4,6-TCP)
c) o-cresol
d) ethylene glycol
e) di(2-ethylhexyl)phthalate.
Several stages of testing were performed (Section
5.0). To compare static-replacement and flow-through
test results, paired tests were run simultaneously using
the toxicants CuSOi* , 2,4,6-TCP, and o-cresol. Additional
concentrations were used in the static-replacement tests
to obtain better estimates of the ECSO's for growth and
reproduction. Static- replacement tests were also per-
formed using ethylene glycol and di (2-ethylhexyl) phthalate.
These toxicants were not used in flow-through tests due
to their extremely high threshold effect levels.
Due to time constraints, data for each of the four
parameters for which the LCSO's and ECSO's were calculated
had to be collected on the same set of tests. When the
-------
LCSO's and ECSO's occurred at widely separated concentra-
tions, both could not be calculated. When this occurred
the LC50 was considered most important and the test con-
centrations were adjusted accordingly. Additional con-
centrations were added to some of the later static-replace-
ment tests to allow estimation of both LCSO's and ECSO's
from the same tests. This was not feasible with the
flow-through tests because of costs and dilutor design
constraints.
An appropriate statistical model and transformation
for calculating LCSO's and ECSO's were selected based on
the coefficients of determination, variances, and statis-
tical probabilities. The various combinations of models
and transformations used are outlined in Section 6.0.
Selection of the best model and transformation was ex-
tremely important as all comparisons and design calcula-
tions were based on the estimated LCSO's and ECSO's and
the variances associated with them. Once the data were
analyzed, the results were used to.predict the number of
tests and replicates necessary to obtain confidence limits
within a given percentage of the mean. The results of the
analyses using each of the four effect parameters were
evaluated to determine which of the four were most appro-
priate for measuring the effect levels of the toxicants.
-------
2.0 SUMMARY
Of the four parameters measured - mortality, repro-
duction, dry weight, and root length - mortality and re-
production data gave the best results. All of the effect
parameters were collected from the same tests due to time
and budget constraints. Therefore, concentration ranges
were not optimal for each of the effect parameters. Es-
timation of the LC50's was deemed most important, and the
concentration ranges were adjusted to bracket the expected
LC50.
The mortality data indicated a significant difference
only between the two types of tests for CuSO^ (P = 0.03).
The two types of tests for 2,4,6-TCP and o-cresol gave
almost identical ECSO's (Table 2-1). This was unusual
because 2,4,6-TCP and o-cresol are much more volatile
than CuSOij. Static-replacement bioassays on volatile
compounds typically underestimate the LC50; flow-through
tests theoretically give a better estimate because of the
continual replacement of the toxicant. The LCSO's gener-
ated were used to determine the distribution of variances
among types, tests, and replicates and to predict the num-
ber of tests and replicates necessary to yield certain
confidence limits and probability levels (Table 7-6).
No mortality data were generated for di(2-ethylhexyl)
phthalate because range-finding tests indicated that con-
centrations near 200,000 mg/L would be necessary to cal-
culate the LC50. Flow-through tests were not conducted
for ethylene glycol because the concentrations needed to
produce LCSO's were in the 10,000 to 50,000 mg/L range
and therefore not cost-effective.
Reproduction data gave the best results when analyzed
using a growth rate constant (Equation 6-13). This trans-
formation produced the highest R2 values and thus the best
estimate of the EC50. The difference in reproductive
stages of duckweed fronds in each replicate (Section 8.1)
caused more variation, and thus less precision, in the
ECSO's than in the LCSO's. Reproduction rates were gener-
ally higher in flow-through tests than in static-replace-
ment tests, possibly due to the continuous addition of
nutrients or the agitation caused by periodic additions
of the nutrients. No indication of crowding or nutrient
depletion was observed in the static-replacement tests.
ANOVA's were used to test for reproduction rate dif-
ferences between static-replacement and flow-through tests,
-------
Table 2-1.
Summary LC50 Data from Quadratic Regression
Using Arc Sin Square Root Transformation for
Paired Flow-through and Static-Replacement
Tests.
Toxicant
CuSO.,
2,4,6-TCP
o-cresol
Test
Type
F & S
F
S
F & S
F
S
F & S
F
S
NO. Of
Tests x
8
4
4
14
7
7
8
4
4
8.90
6.76
11.71
2.04
2.04
2.03
500. 8
464.3
540.3
Range
Min.
3.67
3.67
7.03
0.87
0.87
0.93
382.6
382.6
436.9
Max.
15.00
10.46
15.00
3.12
3.12
3.11
691.7
577.7
691.7
95% Confidence
Limits
5.96
3.27
6.59
1.63
1.45
1.35
426.6
346.7
398.1
13.30
13.93
20.80
2.55
2.87
3.05
588.8
616.6
741.3
P*
0.03
0.64
0.27
ethylene glycol S
25760
17000
40140
16360
40560
*P = probability of difference between types
-------
Results indicated that the two types of tests yielded sig-
nificantly different ECSO's at or below the 0.11 probabil-
ity level (Table 2-2). Predictions of the number of tests
and replicates needed to obtain confidence limits within a
given percentage of the mean EC50 (Table 8-4) were gener-
ally higher than those necessary to obtain LC50 confidence
limits within the same percentage of the mean (Table 7-6).
The root length data could not be analyzed adequately
because dead roots fell off the fronds and were too fragile
to measure; thus data sets for individual replicates were
incomplete, resulting in poor regressions and inaccurate
estimations of ECSO's based on root length. In many cases
ECSO's could not be estimated.
Toxicant concentration ranges were too high to yield
meaningful EC50 data based on dry weight. The concentra-
tion ranges were established to produce the best LC50 es-
timates , and limited funding and time constraints prevented
conducting tests designed to yield valid dry weight data.
0-cresol tests yielded the only data which were complete
enough to compare static-replacement and flow-through
tests. The two types of tests were significantly differ-
ent at the 0.11 probability level (Table 2-3).
-------
Table 2-2.
Summary EC50 Data from Quadratic Regression Using
Growth Rate Constant Transformation for Paired
Flow-through and Static-Replacement Tests.
Toxicant
CuSO.,
2,4,6-TCP
o-cresol
ethylene glycol
di(2-ethylhexyl)
phthalate
Test
Type
F & S
F
S
F & S
F
S
F & S
F
S
S
S
No. Of
Tests
10
5
5
12
6
6
8
4
4
5
7
Range
X
2.82
2.21
3.51
0.07
0.03
0.13
192.7
151.5
245.8
17159
2060
Min.
1.20
1.20
2.92
0.01
0.03
0.03
139.9
139.9
190.5
12796
397
Max.
5.95
2.65
5.95
0.95
0.09
0.95
371.7
154.6
371.7
27179
7582
95% Confidence
Limits
2.16
1.76
2.16
0.02
0.01
0.02
147.3
140.9
155.7
12130
706
3.68
2.79
5.70
0.19
0.05
0.90
252.3
161.9
388.0
24274
6008
P*
0.11
0.10
0.05
*P = probability of difference between types
Table 2-3.
Summary EC50 Data from Quadratic Regression Using
Arc Sin Square Root of the Ratio Dry Weight to
Control Weight Transformation for Paired Flow-
through and Static-Replacement Tests.
Toxicant
o-cresol
Test
Type
F & S
F
S
No. of
Tests
8
4
4
Range
X
23.9
32.0
17.0
Min.
0.17
6.93
0.17
Max.
127.6
127.6
77.9
95% Confidence
Limits
7.73
4.71
0.88
73.79
218.3
328.9
P*
0.11
*P = probability of difference between types
-------
3.0 CONCLUSIONS
Static-replacement and flow-through tests yielded
similar results for mortality data on 2,4,6-TCP and
o-cresol. The mean LCSO's for the two types of CuSOi*
tests were statistically different (P = 0.03). The ma-
jority of the variation within each toxicant occurred
between tests. For all toxicants and groupings of static-
replacement, flow-through, and combined static-replace-
ment and flow-through tests, the between tests variation
averaged 4X the within test variation. Thus, in order
to obtain the best estimate of the true LC50, resources
should be allocated to conduct several tests at different
times. This should be done even at the expense of within
test replication if necessary.
The variation between tests also was greater than
between types of tests (static-replacement and flow-
through) . Therefore, if resources are limited, several
replicated static-replacement tests should be conducted
rather than one or a few flow-through tests.
Static-replacement and flow-through tests produced
similar results for reproduction data on CuSOi* and
2,4,6-TCP (P = 0.11 and P = 0.10, respectively). The
ECSO's generated by the two types of o-cresol tests were
statistically different (P = 0.05). Variance components
were not as evenly distributed in the reproduction data
as in the mortality data. In static-replacement tests
an average of 65% of the variation was between tests,
whereas in flow-through tests the between test variation
averaged 9% (Table 8-4). This indicates that the best
allocation of resources for static-replacement tests would
be to conduct several tests with only a few replicates
(Table 8-4). On the other hand, flow-through tests
should be highly replicated with fewer tests allocated
across time; however, from a practical standpoint, this
would be difficult to implement. Additionally, there is
more variation among tests than between types, so that
conducting several replicated static-replacement tests is
the preferred alternative.
The dry weight data were poor for all toxicants ex-
cept o-cresol. The o-cresol data were complete enough to
allow comparison of static-replacement and flow-through
tests. The two types of tests were not significantly dif-
ferent (P = 0.11). Partial correlation coefficients in-
dicated that dry weight data did not contain a significant
-------
amount of information beyond that contained in the mor-
tality and reproduction data. Given these results, the
variability of the data, and the high cost of collecting
it, dry weight data collection is not recommended unless
funds allow additional tests to be designed specifically
for dry weight data collection. Dry weight did appear to
be affected by much lower toxicant concentrations than
did mortality or reproduction. Therefore, in appropri-
ately designed tests, meaningful data probably could
be obtained.
The root length data did not produce meaningful
ECSO's (Section 10.0). Fragmentation of the roots after
death of the frond precluded their measurement.
-------
4.0 RECOMMENDATIONS
Based on results of this study, the best allocation
of resources would be to conduct static-replacement tests
instead of flow-through tests. Mortality and reproduction
(measured by the growth rate constant) should be assessed
at the conclusion of seven-day tests. The tests should be
replicated in time, with four replicates of each concen-
tration and the control. A minimum of four tests should
be conducted to allow sufficient estimation of the popu-
lation LC50 or EC50 and calculation of test statistics.
Test chambers should be 250 ml beakers or larger and
should contain at least 150 ml, or approximately 2 in. in
a larger chamber, of a suitable growth medium such as Hill-
man's M-medium. Duckweed should be added as complete colo-
nies; to prevent damage, fronds should not be separated.
Three 4-frond colonies should be used when possible, al-
though the data indicated no differences in LCSO's and
ECSO's using only one 4-frond colony. The medium in each
test chamber should be replaced at least every third day
during the testing. Lighting should be continuous at
500 + 50 fc.
The data should be analyzed by quadratic regression
with log transformation of concentration, arc sin square
root transformation of the proportion (p) of dead to total
fronds, and log transformation of the reproduction data
(Equation 6-13) . An LC50 or EC50 should be calculated
for each replicate and analyzed by an ANOVA to test for
statistical differences between replicates. If differ-
ences are significant, the test data and experimental
technique should be examined before continuing. Assuming
no differences, a mean LC50 or EC50, standard error, con-
fidence limits, and range should be calculated for each
test.
The replicate LC50 or EC50 data should also be ana-
lyzed using a nested ANOVA to partition out the variance
components between and within tests (Section 6.1.4).
Variance components should be examined to determine the
distribution of the variation. Normally, the largest
variation occurs between tests, and a single mean LC50
or EC50, ranges, confidence limits, and standard errors
should be calculated for the toxicant. The mean square
and degrees of freedom for between tests should be used
to calculate the confidence limits. If confidence limits
are larger than desired, the mean square from the nested
-------
ANOVA and Equations 6-11 and 6-12 should be used to deter-
mine how many additional tests should be conducted to nar-
row the confidence limits to within the desired range.
10
-------
5.0 LABORATORY PROCEDURES AND METHODS
5.1. Selection of Test Species
Lemna gibba G-3 was chosen as the test species because
of its taxonomic stability and previous research on its phys-
iology. L. minor was not chosen because it cannot be defini-
tively identified as a laboratory organism due to the absence
of flowering (Hillman pers. comm. 1980). Furthermore, the
literature does not deal with a specific strain of L. minor.
L. gibba G-3 does flower and has been defined as a laboratory
organism. It has been used extensively in research in the
United States, Japan, and Europe. Axenic cultures of L. gibba
G-3 were obtained from Dr. W.S. Hillman at the Brookhaven
National Laboratory in New York.*
5.2. Selection of Toxicants
The EPA Office of Toxic Substances selected toxicants
which are representative of a wide variety of compounds
likely to be assayed in duckweed bioassays. The compounds
varied from a soluble inorganic compound to insoluble organic
compounds (Table 5-1). A log of tests conducted is presented
in Table 5-2. Carriers were not used in the testing as they
are not present in natural situations.
Range-finding tests were conducted to determine the
"100%" concentration for each toxicant to be used in the
definitive tests. Ten concentrations of sufficient range
were used to determine the lowest concentration at which
there was 100% mortality. Range-finding tests were con-
ducted in 250 ml glass beakers for seven days. Equipment,
chemicals, and procedures were the same as those used in
the static-replacement tests (Section 5.5).
5.3. Culture Methods
Stock cultures of Lemna gibba G-3 were maintained in
three to six 6 L glass aquaria covered with Nytex screen.
*Biology Dept., Brookhaven National Laboratory, Upton,
New York.
11
-------
Table 5-1. Test Toxicants and Their Physical Characteristics.
Toxicant Formula
Solubility in H2O
copper sulfate CuSOn-5H2O decomposes
(18.2g/100g H2O at 25°C)
OH
2, 4, 6-trichlorophenol vs.
/ (0.09g/100g H2O)
Use
algicide,
fungicide,
herbicide,
bactericide
fungicide,
bactericide ,
preservative
di (2-ethylhexyl)
phthalate
[bis(2-ethylhexyl)
phthalate]
ethylene glycol
o-cresol
(o-hydroxytoluene)
COOCH2CH(C2Hs)(CH2)3CH3
. COOCH2CH(C2Hs)(CHa)3CH3
HOCH2CH2OH
CK3
not soluble in water
miscible with water
soluble in 40 parts
water
(2.5g/100g)
vacuum pumps
antifreeze,
in hydraulic
fluids, solvent
in industrial
processes
disinfectant,
solvent
-------
Table 5-2. Test Log.
Test #
1-14
15
16
17
18
19
20
21
22
23
24
25
26
27-28
29-31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Toxicant
ethylene glycol
di ( 2-ethylhexyl) phthalate
2,4,6-TCP
2,4,6-TCP
2,4,6-TCP
di ( 2-ethylhexyl) phthalate
ethylene glycol
CuSOi,
ethylene glycol
CuSOi*
o-cresol
2,4,6-TCP
(test numbers omitted)
CuSCK
CuSCH
di ( 2-ethylhexyl) phthalate
2,4,6-TCP
2,4,6-TCP
ethylene glycol
o-cresol
ethylene glycol
di ( 2-ethylhexyl) phthalate
di ( 2-ethylhexyl) phthalate
o-cresol
o-cresol
di (2-ethylhexyl) phthalate
ethylene glycol
CuSO^
CuSCH
CuSCH
CuSOu
ethylene glycol
o-cresol
o-cresol
di (2-ethylhexyl) phthalate
2,4,6-TCP
2,4,6-TCP
2,4,6-TCP
CuSOi,
CuSOi*
di (2-ethylhexyl) phthalate
o-cresol
Test Type1
Trials, Range
Range
Range
Range
Flow
Static
Range
Range
Range
Range
Range
Range
Range
Range, Growth
Flow
Static
Range
Flow
Static
Static
Range
Static
Static
Static
Flow
Static
Static
Static
Flow
Static
Flow
Static
Static
Flow
Static
Static
Flow
Static
Static2
Static
Static2
Range
Flow
Date
6/25-8/23/79
8/23/79
8/23/79
8/23/79
8/23/79
8/23/79
9/19/79
9/19/79
9/19/79
9/19/79
9/19/79
9/19/79
9/19/79
9/19/79
9/27/79
9/27/79
9/28/79
10/11/79
10/11/79
10/12/79
10/12/79
10/23/79
10/23/79
10/23/79
11/01/79
11/01/79
11/13/79
11/13/79
11/29/79
11/29/79
12/13/79
12/13/79
12/13/79
1/03/80
1/03/80
1/03/80
1/17/80
1/17/80
1/17/80
1/17/80
1/17/80
1/25/80
1/31/80
13
-------
Table 5-2 (Continued).
Test #
Toxicant
Test Type1
Date
61
62
63
64
65
66
67
68-69
70
71
72
73
74
75
76
77
78
79
80
81
82-84
85
86
87
88
89
90
91
92
93
94
95
o-cresol
o-cresol
di ( 2-ethy Ihexy 1) phthalate
2,4,6-TCP
2,4,6-TCP
ethylene glycol
ethylene glycol
o-cresol
o-cresol
CuSCK
CuSO4
2,4,6-TCP
2,4,6-TCP
2,4,6-TCP
2,4,6-TCP
CuSCK
CuSC>
2,4,6-TCP
CuSO4
toxicant carrier
CuSOit
CuSOit
toxicant carrier
2,4,6-TCP
2,4,6-TCP
di ( 2-ethy Ihexyl) phthalate
2,4,6-TCP
2,4,6-TCP
di ( 2-ethy Ihexy 1) phthalate
2,4,6-TCP
2,4,6-TCP
Static
Static2
Static
Flow
Static
Static
Static2
Reproduction
Flow
Static
Static
Static2
Flow
Static
Static
Static2
Static
Static2
Static
Static
Range
Flow
Static
Range
Flow
Static
Static
Flow3
Static
Static
Flow
Static
1/31/80
1/31/80
1/31/80
2/14/80
2/14/80
2/14/80
2/14/80
2/28/80
3/06/80
3/06/80
3/11/80
3/11/80
3/20/80
3/20/80
4/03/80
4/03/80
4/03/80
4/03/80
4/17/80
4/17/80
5/01-5/08/80
5/22/80
5/22/80
5/22/80
6/05/80
6/05/80
6/05/80
6/19/80
6/19/80
6/19/80
6/26/80
6/26/80
1 Flow = Flow-through
Static = Static-replacement
Range = Range-finding
2 Modified — only one 4-frond colony per beaker instead of three
4-frond colonies per beaker.
3 Aborted
14
-------
Lighting was adjusted to 500 + 50 foot candles (fc) measured
at the water surface. Cultures were thinned three times a
week so that they covered approximately half the surface area
of each aquarium. After thinning, the remaining organisms
were agitated twice with Hillman's medium in a 500 ml Erlen-
meyer flask to remove algae and poured over Nytex screen to
drain off the suspended algae.
An axenic culture of duckweed was maintained at all
times on nutrient medium and agar. Working stock cultures
were established from this axenic reserve culture; however,
working cultures were not maintained axenically. If axenic
stock cultures were used then 1)the tests themselves should
be conducted axenically, or 2)an intermediate acclimation
step should be incorporated between the axenic cultures and
the definitive tests. The first alternative, axenic flow-
through tests, would be very costly to conduct because large
volumes of nutrient solution and toxicants would require
sterilization. Moreover, the entire dilutor system would
have to be sterilized prior to each test. If contamination
occurred during any stage of the test, all data would have
to be discarded and a new test initiated after sterilization.
Continuous lighting and the rapid reproductive rate would
deplete the carbon dioxide in the container and consequently
affect growth rates. Filtered carbon dioxide would have to
be pumped into the chambers to assure an adequate supply.
Also, nutrient solutions might be subtly altered by milli-
pore filtering or autoclaving.
If organisms from axenic working stock cultures were
used in definitive tests which were not conducted axenically,
culture organisms would not be acclimated to test conditions.
For example, the glass containers used to enclose axenic
cultures would remove portions of the light spectrum which
would be present during testing. A minimum of an additional
3 to 5 days would be needed for acclimation. Ten days would
be better to allow the cultures to cycle through several gen-
erations and thereby eliminate effects of the axenic condi-
tions. In summary, it is generally recommended that working
stock cultures be maintained under conditions identical to
those used for definitive tests.
5.4. Flow-through Tests
5.4.1. Equipment and Chemicals
Diluent and toxicant panels — design. The dilutor
systeirPused was a modified EPA design (Peltier 1978) . It
consisted of a panel of diluent proportioning chambers and
a panel of toxicant proportioning chambers, operated by a
cycle control panel (Figure 5-1). Measured amounts of dil-
uent and toxicant were drained synchronously into final
15
-------
Flow control valves
A
A
Overflow
standpipe
V
1—
/
i— Toxicant holding chamber
Liquid level switch — i
ft
C D
y— Pre-mixing chamber
Diluent proportioning
chambers 1-6
• Tygon tube extension
I
1
2
3
4
5
6
Li /^Toxicant proportioning!..
I chambers 1-6
j '.
linn ml
Final mixing chambers
1 1 1 1
I
Figure 5-1.
Test chambers
EPA-type flow-through dilutor system. A and B,
normally open solenoid valves; C, D, and E,
normally closed solenoid valves.
16
-------
mixing chambers which emptied into test chambers containing
the test organisms.
Two siphon-connected 240 L aquaria served as diluent
reservoirs; one 240 L aquarium served as the toxicant reser-
voir. A submersible pump fed both diluent and toxicant to
the control panel via Tygon lines (formula 3603, 3/8 in.).
A submersible pump in the toxicant reservoir agitated the
toxicant to keep it evenly mixed.
There were eight diluent proportioning chambers —
seven overflow-connected and one isolated chamber. Each
of the seven connected chambers received a diluent (nutrient)
solution from the reservoir via a line controlled by solenoid
and variable flow valves. The solution level was controlled
by a liquid level switch. Each chamber drained into a spe-
cific mixing chamber via a #316 stainless steel line (1/8 in.
inside diameter) controlled by an in-line solenoid valve
(Peter Paul model #7lP9ZGV).
The isolated chamber (toxicant holding chamber) in the
diluent proportioning chamber panel received toxicant solu-
tion from the toxicant reservoir via a line also controlled
by solenoid and variable flow valves. The solution level
in the chamber was controlled by an overflow standpipe con-
nected to the toxicant reservoir. The working solution
(2216 ml) in this isolated chamber drained into the toxi-
cant pre-mixing chamber via a stainless steel line controlled
by a solenoid valve.
The toxicant pre-mixing chamber, mounted on a magnetic
stirrer, emptied into six toxicant proportioning chambers
via a Tygon tubing line controlled by a solenoid valve. The
six chambers were overflow-connected; each drained into a
specific final mixing chamber, along with a specified amount
of diluent solution. One of the final mixing chambers re-
ceived only diluent solution, and a second received only
toxicant solution.
The volumes of solution delivered by the panels were
controlled by Tygon tubing extensions on the drain lines in
each chamber to insure precise dilution of the toxicant
(Table 5-3).
Each final mixing chamber was drained by four Tygon
lines (formula 3603, 1/4 in.); each of the four lines emp-
tied into a 6 L glass aquarium serving as a test chamber
for the test organisms. Therefore, each of the seven final
mixing chambers drained into four aquaria. A Nalgene twist
valve at the end of each Tygon line equalized flow in the
four lines from each final mixing chamber. These valves
corrected for the unequal flow rates in the lines created
by different line lengths and surface tension. An elevated,
sloped centerboard, painted with white epoxy paint, sup-
ported the Tygon lines feeding the aquaria.
17
-------
Table 5-3. Volumes of Diluent and Toxicant Delivered by the
Proportioning Chambers.*
Volume (ml)
Chamber Diluent Toxicant
1 440 560
2 680 320
3 820 180
4 900 100
5 944 56
6 1000 1000
*Like-number diluent and toxicant chambers mixed and emptied
into a final 1000 ml mixing chamber (i.e., DI + Ei = 1000 ml)
with the exception of diluent and toxicant chambers 6, which
emptied into separate final mixing chambers.
18
-------
Each 6 L glass receiving aquarium (test chamber) had
an overflow hole drilled at the 2 L level. A Nytex screen,
cemented to the aquarium with silicone rubber, covered the
hole. The aquaria were arranged in four rows, seven aquaria
per row. Two galvanized tin trays, painted with white epoxy
paint, each held two rows of seven aquaria. The sides of
the holding trays were lined with aluminum foil for addi-
tional light scattering. Overflow from the aquaria drained
into the trays and then into a galvanized tub containing a
float-switched submersible pump (Little Giant model 1-M).
Four-foot fluorescent light fixtures (Lakewood Engineering
and Manufacturing Co.) with two light bulbs (Vita Lights)
were suspended on chains above the aquaria to permit adjust-
able light settings.
Diluent and toxicant panels — operating sequence.
The flow apparatus was adjusted at the time control panel
to cycle approximately once every 30 minutes. The final
mixing chambers each received a 1000 ml solution of toxicant
and diluent and subsequently delivered 250 ml to each of
four aquaria. Thus the 250 ml of diluted toxicant emptying
into each aquarium every half hour resulted in a complete
volume change in each aquarium every four hours.
The time control panel activated the pumps in the dil-
uent and toxicant reservoirs, opened solenoid valves A and
B, and closed solenoid valves C, D, and E. Thus the pumps
in the reservoirs filled the diluent proportioning chambers
and the toxicant holding chamber, while the toxicant solution
in the pre-mixing chamber drained into the toxicant propor-
tioning chambers. Flow rates in the reservoir lines were
adjusted to allow the toxicant holding chamber to fill be-
fore the diluent proportioning chambers were full. When
the diluent solution reached the diluent liquid level
switch, solenoid valves A and B were closed, the reservoir
pumps were shut off, and solenoid valves C, D, and E were
opened. This set of synchronous events allowed the diluent
and toxicant proportioning chambers to drain into their re-
spective final mixing chambers while the toxicant holding
chamber refilled the pre-mixing chamber. The 1000 ml in
each final mixing chamber subsequently drained into the
designated test chambers.
Mixing of Hillman's medium. Hillman's M-medium was
used as the nutrient solution for culturing the duckweed
and as a diluent in the tests (Hillman 1961b). The solu-
tion was mixed in a 720 L stainless steel (#316) tank using
analytical reagent grade chemicals. Calibrated glass bot-
tles were used to measure deionized water into the tank.
A specified volume of 678.53 L was standardly mixed to
minimize error. Stock chemicals for adding to the tank
were maintained in high but stable concentrations (Table
5-4). Micronutrients were premixed in a standard solution
to minimize measurements of stock chemicals.
19
-------
Table 5-4. Concentrations of Stock Chemicals Used to Mix
Hillman's M-Medium.
Chemical Concentration(g/L)
Potassium monobasic phosphate (KH2POiJ 200
Potassium nitrate (KNO3) 250
Calcium nitrate [Ca(N03)2•4H20] 400
Magnesium sulfate (MgSOi* • 7H20) 400
Ferric chloride (FeCl3) 21.6
Tartaric acid [HOOC(CHOH)2COOH] 12.0
Micronutrients
Boric acid (H3B03) 2.86
Zinc sulfate (ZnSCM 0.22
Sodium molybdate (NaMoCK) 0.12
Cupric sulfate (CuSOiJ 0.08
Manganese chloride (MnCl2) 3.62
20
-------
After adding deionized water and stock chemicals to
the tank, 10 N NaOH (sodium hydroxide) was added to raise
the pH to a range of 4.8 to 5.2. A submersible pump (Little
Giant model 1) circulated the solution in the tank to insure
uniform concentration. After pH adjustment, the Hillman's
solution (diluent) was pumped into the diluent reservoir
via a Tygon line (formula 3603, 1/2 in.) and was measured
into the toxicant reservoir with a calibrated bottle.
Toxicants. Each flow-through test was conducted with
analytical reagent grade toxicants CuSCK-SHzO, o-cresol,
and 2,4,6-TCP. For each test, the toxicant was diluted by
the following log scale percentages of the highest concen-
tration selected: 100, 56, 32, 18, 10, 5.6, and 0.
5.4.2. Operating Procedures
Duration of tests. Before initiating flow-through
tests, equipment was operated approximately 48 cycles with-
out duckweed in the test chambers. These preliminary cycles
allowed the test chambers to be filled with the proper toxi-
cant dilutions and to equilibrate prior to introduction of
the test organisms. Flow-through tests, with duckweed in
the test chambers, were conducted for seven days (168 h =
336 cycles) for each toxicant at seven different dilutions.
Randomization of test chamber deliveries. The parti-
cular toxicant dilution received by each of the 28 test
chambers was determined with a random numbers table. Re-
randomization was performed prior to each new test.
Calibration of equipment. On Days 0, 4, and 7 equip-
ment was checked for proper operation and recalibrated as
necessary- Volume delivery of the diluent and toxicant
panel chambers was adjusted with the use of Tygon tubing
(1/4 in.) extended at variable lengths above the top of
the standpipe within each chamber. The volume delivery
to test chambers was calibrated by the addition of 1000 ml
deionized water to each chamber and adjustment of Nalgene
twist valves at the ends of the test chamber delivery tubes.
Lights were adjusted to deliver 500 + 50 fc measured at
water surface level with a Protomatic photometer beneath
the middle and ends of each light.
Hillman's medium and toxicant mixing schedules.
Hillman's medium was mixed in the tank approximately every
other day during the seven-day run. Usually enough medium
was immediately pumped out to fill the one toxicant and two
diluent reservoirs. The toxicants, mixed in Hillman's medium,
were then added at the proper concentration to the toxicant
reservoir.
21
-------
Assessment of effects. Upon initiation of an experi-
ment (Day 0), three approximately equal-sized, 4-frond
duckweed colonies from the same stock culture aquarium
were placed in each test chamber. Duckweed fronds and
colonies were inspected on Days 1 (approximately 24 hours
after test initiation), 4, 5, 6, and 7. Fronds were
assessed for the following characteristics:
(1) Total number of fronds - number of fronds in each
test chamber, without regard to possession of any of the
following characteristics. New fronds budding from the
brood pouch were counted as whole fronds.
(2) Necrotic fronds - fronds possessing localized
regions of dead or decaying tissue, usually surrounded by
healthy tissue. These regions may appear as gray, yellow,
reddish, black and/or watery areas of obvious death and
subsequent decay on an otherwise healthy frond.
(3) Chlorotic fronds - fronds possessing areas of
progressive bleaching in color from green to yellow.
(4) Alive fronds - fronds that are either totally
green or possessing necrotic and/or chlorotic characteristics,
(5) Dead fronds - fronds possessing no yellow or green
tissue, usually all brown or white in color.
Colonies were inspected for the following charac-
teristics :
(1) Total number of colonies - number of discrete
groups of fronds without regard to characteristics of mem-
ber fronds. Any solitary fronds were counted as one colony.
(2) Alive colonies - any colonies possessing one or
more live fronds.
(3) Dead colonies - colonies possessing no live fronds.
Water quality measurement. Temperature, conductivity,
and pH were measured in each test chamber on Days 0, 4, 5,
6, and 7 with a YSI 33 SCT conductivity/temperature meter
(Yellow Springs Instruments Co.) and an lonalyser model
407A pH meter (Orion Research). In addition, continuous
measurements of temperature and conductivity were recorded
in a randomly selected flow test chamber for the duration
of each experiment.
Root length and weight measurement. Root length and
weight of five 4-frond colonies from the original stock
culture were measured on Day 0 before initiation of each
test. On Day 7 after completion of an experiment, root
length and weight of all duckweed in each test chamber
were measured. Root lengths were measured with a ruler
22
-------
to the nearest 1 ram. Weights were determined to the nearest
0.0001 g by drying the duckweed for 90 minutes in bottles
over silica gel at room temperature at a negative pressure
of 1 cm mercury (Blackman and Robertson-Cuningham 1953).
Equipment cleaning. All equipment was triple-washed
with soap, hydrogen chloride (5% HC1), and acetone before
reuse, as outlined in Peltier (1978). All Tygon tubing was
replaced upon completion of each experiment.
5.5. Static-Replacement Tests
5.5.1. Equipment and Chemicals
Glass beakers (250 ml) served as test chambers for the
static-replacement tests. Beakers were arranged in rows and
placed on a table covered with a white vinyl cloth. Fluores-
cent light fixtures were suspended on chains above the beakers
to permit adjustable light settings.
Static replacement tests were conducted with analytical
reagent grade toxicants di(2-ethylhexyl)phthalate and ethylene
glycol in addition to CuSOit-SHaO, o-cresol, and 2,4,6-TCP.
The Hillman's nutrient solution and the toxicant concentra-
tions were the same as those used for flow-through tests
(Section 5.4.1) .
5.5.2. Operating Procedures
Duration of tests. Static-replacement tests were con-
ducted for a total of seven days. All flow-through tests
were run concurrently with a static-replacement test of the
same toxicant.
Randomization of beakers. Arrangement of beakers was
determined with a random numbers table. Rerandomization was
performed prior to each new test.
Calibration of lights. Lights were adjusted to deliver
500 + 50 fc measured at water surface level beneath the mid-
dle and ends of each light. On Days 4 and 7 light levels
were checked and adjusted as necessary.
Toxicant mixing procedure. For static-replacement
tests that were conducted concurrently with a flow-through
test, the 100% toxicant concentration was obtained from the
toxicant reservoir in the flow-through apparatus. The other
six toxicant concentrations were obtained by sequentially
diluting the solution from the toxicant reservoir with
23
-------
Hillman's medium. The final toxicant concentrations obtained
were the same as the seven toxicant concentrations delivered
by the flow-through apparatus. Three additional concentra-
tions were generally used to enhance the accuracy of EC50
predictions for reproduction and dry weight.
For static-replacement tests not conducted concurrently
with a flow-through test, the sequence of toxicant concen-
trations was made by adding the toxicant to Hillman's medium
obtained from the diluent reservoir in the flow-through appa-
ratus. Once the maximum desired concentration ("100%") was
obtained, this solution was then diluted as described above
to yield the seven toxicant concentrations.
All beakers were filled to the 150 ml level with toxi-
cant solution. Toxicant solutions were changed on Days 4
and 6 to control algal growth, and to prevent concentration
of toxicants due to evaporative loss or dilution due to vola-
tility- Duckweed colonies were lifted from each beaker with
a small Nytex screen, cleaned of algae, and placed in a beaker
with fresh solution.
Procedures for assessment of effects, water quality
measurement, root length and weight measurement, and equip-
ment cleaning were conducted as described for flow-through
tests (Section 5.4.2).
5.6. Laboratory Conditions
Tests were conducted in a temperature-regulated room
maintained at 25 + 3°C.
5.7. Scheduling of Tests
Tests were scheduled with a 14-day (10 working days)
interval between initiation of consecutive tests (Table 5-5),
Each test was conducted for 7 days (Days 3-9, or 5 work-
ing days) followed by a 7-day period (5 working days) for
breakdown and equipment cleaning, data compilation, and set-
ting up the next test.
A 7-day test period, as opposed to a shorter period,
was chosen for several reasons. Several days are needed
for the fronds to adjust to the new experimental environ-
ment, less crowded conditions, flowing water, and handling
(Hillman 1961a, Walbridge 1977) . Toxic effects are more
apparent over a longer period, particularly at low concen-
tration levels. Growth rate differences are more apparent
over a 7-day period because of the doubling rate of duck-
weed. Thus, the longer test period accentuates differences
in growth and mortality.
24
-------
Table 5-5.
Outline of Tasks Performed During 14-Day
Interval Between Tests.
Day #
1
2
3
4
5
6
7
Day of
Week
Tues.
Wed.
Thurs.
Fri.
Sat.
Sun.
Mon.
Test
Day #
0
1
2
3
4
Tasks
Set up test; calibrate; mix chemi-
cals; mix flow tanks.
Set up test; mix chemicals; start
flow apparatus.
Initiate test; measure water qual-
ity; select 4-frond colonies;
obtain initial weights and root
lengths; mix flow tanks.
Make assessments; mix chemicals;
fill flow tanks to last through
weekend.
None
None
Make assessments; measure water
8
9
Tues.
Wed.
5
6
10
11
12
13
14
Thurs,
Fri.
Sat.
Sun.
Mon.
quality; mix chemicals; replace
toxicants in static tests.
Make assessments; measure water
quality; mix chemicals.
Make assessments; measure water
quality; mix chemicals; replace
toxicants in static tests.
Make assessments; measure water
quality; measure dry weights and
root lengths; break down test.
Clean equipment.
None
None
Clean equipment; set up for next
test.
25
-------
Scheduling tests to begin on Thursday (Table 5-5)
leaves the weekends free of work, provided there is suit-
able storage capacity for the flow operation. A 5-day test
period would still require a 14-day interval between tests
to avoid any weekend work, i.e., 5 days of testing and 5
days of preparation and clean-up. The 7-day test period
allows a more rigorous examination of toxic effects on
duckweed without a real extension of the time period actu-
ally required for the tests.
26
-------
6.0 STATISTICAL METHODS
6.1. Mortality Data Analysis
LCSO's were calculated using several models and
transformations to determine which combination of models
and transformations gave the best estimate. The concen-
tration of the toxicant was transformed using base ten
logarithms. The proportion of dead fronds to the total
number of fronds (p) was transformed using the following
transformations:
angular: y = sin"1 /p~ (6-1)
logit: y = G~1(p), where G = logistic cdf (6-2)
normit: y = F l(p), where F = normal cdf (6-3)
probit: y = F~*(p) + 5, where F = normal cdf. (6-4)
Four types of models were used in the analysis:
linear: y = 30 + 3i x + e (6-5)
quadratic: y = 80 + 61 x + Bax2 + e (6-6)
inverse: x = oto + a-i y + e (6-7)
probit maximum likelihood: E(p) = $(3o + Six) (6-8)
where 3o/ Si, and $2, and ao and ai are regression coeffi-
cients, e is the experimental error term, x is the logio
of concentration, y is the transformed value of p (p =
proportion of dead fronds to total fronds), E(p) is the
probability of killing a frond, and $ is the normal dis-
tribution function.
6,1.1. Weighting Factors
Weighting coefficients were used in all calculations
to account for dependence of the variance on the response
rate. The general form of the weighting coefficients is
given by Finney (1978). For the normit (or probit), logit,
and angular (arc sin square root) transformations, the
27
-------
weighting coefficients are
- -v2
normit (2Tr) 1e , where y = normit of p
(or probit) : p(l-p)
logit: 4p(l-p)
angular: 4 .
The weighting coefficients were incorporated by using them
in a WEIGHT statement in SAS-GLM (SAS 1979). The combi-
nations of models and transformations used are shown in
Table 6-1.
6.1.2. Characteristics of Transformations
Angular. The angular transformation (6-1) is used to
normalize percentages and proportions. It essentially
makes the variance free of p which makes the basic assump-
tions of standard regression analysis more appropriate.
Finney (1964) noted that, for all practical purposes, the
angular transformation is a linear function of the probit
transformation. One disadvantage is that the range over
which the transformation is valid is finite. The model
equation is invalid for very high and very low concentra-
tions which produce near 0 or 100% mortality.
Logit. The logit transformation is y = (1/2)In[p/(1-p)]
(Finney 1978). The normit and logistic functions are very
similar except in the extreme tails, and the resulting es-
timators will be similar. Berkson (1944, 1949) showed that
the logit transformation could be essentially analogous to
the probit. However, the logit transformation is much
easier to compute.
Normit and probit. The normit transformation is the
same as the probit transformation except that 5 is not
added. Specifically, the normit is
Y = F"1 (p) , (6-3)
where F (x) = J (2Tr)~1/2 exp (-t2/2)dt
— CO
is the normal distribution function (Finney 1978). Both
the probit and normit models assume a normal distribution
of the lethal concentrations. The major disadvantage of
these two transformations is that they are difficult to
compute without the use of a computer and packaged rou-
tines such as SAS or BMD (Dixon 1974).
28
-------
Table 6-1. Combinations of Models and Transformations Used
to Analyze Mortality Data.
Model
Transformation
Procedure
Probit maximum likelihood
(Finney 1964)
Probit
PROC PROBIT
(SAS 1979)
Linear regression Arc sin square
(Snedecor and Cochran 1967) root
Normit
Logit
PROC GLM
(SAS 1979)
Quadratic regression Arc sin square
(Snedecor and Cochran 1967) root
Normit
Logit
PROC GLM
(SAS 1979)
Inverse regression (Shuster
and Dietrich 1976)
Arc sin square
root
PROC MATRIX
(SAS 1979)
29
-------
6.1.3. Models
Linear and quadratic regression models are commonly
used statistical tools and do not require explanation
(Snedecor and Cochran 1967). The inverse regression
method is relatively new and requires a brief explanation.
Its application to quantal response assays is discussed
in detail by Shuster and Dietrich (1976). The procedure
applies least squares estimation but reverses roles of
independent and dependent variables. The model is
x = ao + aiy + e (6-7)
_ i
where x = logi 0 (concentration) and y = sin /p~. The pro-
cedure uses weighted least squares, of the horizontal de-
viation from the regression line, to estimate model param-
eters. There are two major advantages of this model:
1) computations are straightforward and use standard
algorithms, and 2) a confidence interval is always given
and the interval is more "direct" because the parameter
is estimated in the same scale as the deviation is mini-
mized. Further, as stated by Shuster and Dietrich (1976),
if the true model is not linear, the asymptotic variance
is less than for other procedures. However, if the true
model is not linear, then procedures that assume a linear
model will produce biased LC50 estimates, i.e. the ex-
pected value of the LC50 estimate will not equal the true
LC50 value. Thus, the estimates based on linear models
will consistently either underestimate or overestimate
the true LC50. For the duckweed experiments analyzed,
the LC50 estimates obtained from the inverse regression
procedure were lower than LC50 estimates based on other
procedures, indicating a larger bias in the inverse re-
gression estimates than in the other estimates. This
bias negates the benefit of the smaller variance of the
inverse regression procedure, and suggests a major dis-
advantage of the inverse regression procedure.
Maximum likelihood estimation is also a well-known
procedure which has long been used in bioassay work and
requires no explanation. A primary disadvantage is that
the estimates often require an iterative solution. Finney
(1977) gives a detailed discussion of maximum likelihood
estimation using probit.
6.1.4. Variation of Mortality Data
ANOVA tests using SAS PROC GLM procedures were
used to partition out sources of variation in the esti-
mates (SAS 1979). LC50 estimates were computed by quad-
ratic regression for each replicate of each test. The
30
-------
transformed p values were then subjected to an ANOVA to
test for replicate effects, goodness of fit of the linear
regression models, and interaction between replicates and
model terms. A table was computed for each of the three
transformations. Table 6-2 is provided as an example.
To estimate the distribution of the variability of
LC50 values split among tests and within tests, a nested
analysis of variance was used (PROC NESTED, SAS 1979).
The partitioned variance was then used to compute the
number of tests and number of replicates per test to be
used for future studies using similar compounds. Using
the nested analysis, estimates were obtained for a2 and
x2, where a2 is the variance of a log LC50 estimate based
on one test with one replicate, and x2 is the variance
component between tests. The variance (v) of the average
log LC50 estimate with k tests and r replicates (Snedecor
and Cochran 1967) is
v = g2 + rx2 . (6-9)
kr
To determine the optimum design, k and r must be selected
to minimize v over the feasible values of k and r. The
experimenter must decide within what percentage he must
know the mean log LC50. Once these factors are known,
the equation
I
s2 + rT2 „ (6-10)
can be solved to give an estimate of the number of tests
and replicates within tests which are needed to reach a
certain level of precision. For Equation 6-10,
t = tabulated t value from standard tables with
k-1 degrees of freedom
g = percentage within which the LC50 must be known
m = estimate of y, the average log LC50
s2 = estimate of a2, the within test variance
T2 = estimate of x2, the among tests variance com-
ponent
k = number of tests
r = number of replicates per test.
The values of m, s2, and T2 are obtained by conducting
a nested analysis of variance of the form
Source of Variation Expected Mean Square
Between Tests a2 + ki T2
Reps within Tests a2
31
-------
Table 6-2.
Sample ANOVA to Determine Replicate Effects,
Goodness of Fit of the Linear Regression Models,
and Interaction between Replicates and Model
Terms.
Dependent Variable:" arc sin square root of total fronds Day 7
Source
Model
Error
Corrected Total
Source
Rep. No.
logj o cone
logic cone • logic cone
logic cone • rep. no.
log 10 cone • logic cone •
df
11
12
23
df
3
1
1
3
rep. no. 3
Sum of Squares
159.28222214
24.49222807
183.77445021
Type I SS
2.57144157
130. 85570857
21.96578577
2. 34475001
1. 54453622
Mean Square
14.48020201
2.04101901
F Value PR
0.42 0.
64.11 0.
10.76 0.
0.38 0.
0.25 0.
> F
7420
0001
0066
7672
8582
"weighted: 4 x total fronds Day 7
32
-------
The quantity s2 is the mean square for replicates within
tests. The quantity T2 is computed as T2 =
(between tests mean square) - (reps within tests mean square)
k
i
The value of ki is equal to a weighted average of the num-
ber of replicates in the tests and should not be confused
with k or r in Equation 6-10.
Equation 6-10 can be rearranged algebraically to cal-
culate the number of tests (k) and the number of replicates
(r) within tests that are necessary to compress the con-
fidence limits to within a given percentage (g) of the
mean LC50. The minimum number of tests is calculated by
k ^ T2 (6-11)
t2 (,gm)z
The right-hand side of the equation is solved using esti-
mates of T2, g, and m from previous experiments using
similar compounds or, if available, from tests using the
same compound. Because k and t2 are interrelated, a table
relating them must be generated from standard t-tables.
Once the minimum number of tests is known, the minimum
number of replicates is computed using the formula
r > ^ . (6-12)
(SE, 2k-T2
6.2. Reproduction Data Analysis
Reproduction, expressed as total number of fronds,
was evaluated using two methods. The first method used
quadratic regression (6-6) with a square root transfor-
mation of total frond number and a log:0 transformation
of the concentration. The square root transformation
makes the variances independent of the means and generally
is used to transform count data. A 50% reduction in total
mean number of control fronds was chosen as the EC50.
The second method also used quadratic regression, but
was based on a growth rate constant K:
K = logio(Fd) - logio(Fo) (6
d
33
-------
where F, = number of fronds on Day d
Fo = number of fronds on Day 0
d = day of assessment.
The EC50 was defined as the concentration which causes the
growth rate constant to equal half the control (0 concen-
tration) growth rate.
One drawback of using a quadratic regression is that
confidence limits about a single EC50 estimate cannot be
directly computed. However, they can be generated by hav-
ing the computer calculate the confidence band about the
regression line and projecting down from the points on the
bands whose ordinates are the square root of half the av-
erage number of total fronds in the four replicates.
6.3. Dry Weight Data Analysis
Dry weight was analyzed by two methods. In the first,
quadratic regression with a log!0 transformation of concen-
tration and dry weight was used to determine the EC50.
Half the mean control (0 concentration) dry weight was
selected as the effect level. This level was substituted
into the equation for the regression line, and the log con-
centration was extracted by solving the equation. In the
second method, quadratic regression was also used with a
log 10 transformation of the concentration, but the depen-
dent variable, dry weight, was transformed using the arc
sin square root of the proportion of dry weight to the
mean control dry weight. The EC50 was calculated by sub-
stituting the arc sin square root of 0.5 (50%) into the
regression equation and solving for the log concentration.
6.4. Root Length Data Analysis
The root length data were analyzed using linear and
quadratic models with a logic transformation of concentra-
tions and root lengths. The coefficient of determination
values were examined to determine if linear or curvilinear
regression was appropriate for data analysis.
6.5. Comparative Methods
Partial correlation coefficients were used to deter-
mine if a dependent variable contained information about
the effects of the toxicant in addition to the information
contained in another variable. PROC GLM with MANOVA was
34
-------
used to calculate the partial correlation coefficients
(SAS 1979) . The coefficients were computed according to
the formula
y, ^/sJs (6-14)
where y^^ and y2 are the variables for which the correlation
is being made while y3 is held constant.
If ry is near 0, then y, contains no informa-
1^2* "3
tion about the effects of y1 in addition to the information
contained in y . Conversely, as rv v approaches 1,
3 y xy 2 • y 3
y increases in information about the effects of y in
2 i
addition to the information provided by y3 about y1. Thus
rv v v measures the effects of y on y that are unre-
•* 1 •* 2 * •* 3 1 2
lated to the effects of y on y . A value of rv ,, ,7
i 3 y\y 2• y 3
near 1 would indicate that collection of data relating to
y would provide additional information beyond that con-
tained in y .
35
-------
36
-------
7.0 MORTALITY DATA ANALYSIS
Coefficients of determination were used to determine
if transformation of the mortality data could linearize
the data. Examination of data such as that in Table 7-1
indicated linearization was not possible in the majority
of the cases. ANOVA's were also computed to determine if
the quadratic term in the regression model was significant.
ANOVA tables similar to Table 7-2 were computed for each
toxicant. The data indicated that the quadratic model
with an arc sin square root transformation of the propor-
tion of dead to total fronds was the most appropriate com-
bination for calculation of the LCSO's. The ANOVA pro-
cedure was also used to ascertain if significant differ-
ences occurred among replicates. Examination of the data
indicated no significant interaction even at modest prob-
ability levels for the majority of tests (Table 7-2). Thus
the replicates could be pooled for some data comparisons.
Although the quadratic model gave the best fit for
the data, one drawback is that confidence limits cannot
be calculated directly. However, the computer can be used
to generate predicted values for the upper and lower con-
fidence limits for selected concentrations. The confidence
limits are then scanned for the value which would give the
LC50. Because some experimenters might not have the facil-
ities or the desire to make the necessary calculations,
probit maximum likelihood computations were also made. As
demonstrated for CuSOi* (Figure 7-1) , the probit method
generally agreed with the quadratic method in estimation
of the LC50. However, the confidence limits for the pro-
cedures showed less agreement. The inverse procedure
generally gave much narrower confidence limits, but the
procedure is biased because of incorrect assumptions
(Section 6.1.3.).
In 89% of the tests analyzed by probit maximum likeli-
hood, heterogeneity factors (h) were used to adjust the
variances in calculation of the confidence limits. The
heterogeneity factor (h) (Finney 1977) is designed to ad-
just for heterogeneous responses in the data and is calcu-
lated as
h = X2/df (7-1)
where x2 = the Chi-square value and df = degrees of freedom
calculated as (k-2) where k is the number of concentrations
in the test. The h factor is used when a significant x2
value occurs for the test. A large x2 value can result
from heterogeneity in the organisms or because the mathe-
matical model used to fit the data is incorrect. The
37
-------
Table 7-1. Coefficients of Determination from Linear and Quadratic Regressions
on Mortality Data Using Arc Sin Square Root, Probit, and Logit
Transformations for 2,4,6-TCP.
u>
oo
Test No.
18
19
35
36
54
55
64
65
74
75
76
80
88
89
92
94
95
Linear Regression
* * *
Arc Sin / Probit Logit
0.505
0.768
0.301
0.230
0.532
0.564
0.607
0.520
0.704
0.376
0.344
0.277
0.776
0.349
0.478
0.616
0.526
0.307
0.656
0.133
0.129
0.299
0.361
0.538
0.406
0.523
0.178
0.219
0.096
0.702
0.161
0.269
0.495
0.392
0.213
0.551
0.090
0.093
0.188
0.270
0.500
0.334
0.415
0.105
0.170
0.042
0.648
0.106
0.188
0.452
0.326
Quadratic Regression
. * * *
Arc Sin / Probit Logit
0.649
0.775
0.640
0.550
0.708
0.790
0.769
0.734
0.785
0.648
0.629
0.470
0.835
0.721
0.762
0.821
0.759
0.561
0.742
0.392
0.416
0.544
0.618
0.709
0.692
0.657
0.444
0.517
0.181
0.795
0.482
0.642
0.707
0.642
0.509
0.712
0.291
0.356
0.403
0.496
0.663
0.649
0.557
0.333
0.456
0.070
0.765
0.350
0.549
0.643
0.568
*transformations
-------
Table 7-2. ANOVA Results Using Arc Sin Square Root
Transformation of CuSOi, Mortality Data.
Test No.
13
14
32
33
46
47
48
49
57
58
72
73
78
79
81
MSE
MSE
2.041
1.682
1.366
1.285
1.660
2.201
2.318
1.511
0.874
0.664
0.606
0.403
1.450
0.567
1.419
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
R2
867
868
965
961
949
902
919
939
976
926
968
915
,953
.928
.948
mean standard error
2 , residual
PF-R
PF-L
sum of
PF-R
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.7420
.8309
.0015
.4208
.0069
.3273
.9550
.8008
.8063
.0922
.2022
.0725
.8199
.3554
.6681
(within test
squares from
corrected total
probability of obtaining
probability of obtaining
PF-L
0.0001
0.0001
0.0001
0.0001
0.0001
0.0001
0.0001
0.0001
0.0001
0.0001
0.0001
0.0001
0.0001
0.0001
0.0001
error)
fitted
sum of squares
a greater F if
a greater F if
PF-LL
0.0066
0.0033
0.9351
0.0006
0.6896
0.0067
0.0106
0.0004
0.0534
0.0001
0.0004
0.0046
0.0007
0.0125
0.8390
model
replicates
no linear
PF-LR
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.7672
.3722
.2870
.5958
.5059
.4791
.6103
.8530
.3805
.8244
.3840
.2514
.9871
.8853
.4914
are not
effect of
PF-LLR
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.8582
.2712
.5794
.1933
.6228
.6242
.8762
.8791
.1307
.3829
.4495
.9241
.8728
.3525
.5287
different
concentr
(no overall increase or decrease trend)
PF-LL = probability of obtaining a greater F if no quadratic effect of
concentration
PF-LR = probability of obtaining a greater F if no linear effect interaction with
replicate
PF-LLR = probability of obtaining a greater F if no quadratic effect interaction
with replicate
39
-------
111.6
T
I
I
i
Probit
MLH
Quadra
tip
1 ^ 5T8 52
26
24
j 22
O)
1 20
c
o 18
CO
£ 16
1 14
o
12
10
8
6
t
t
n
^
T"
T
I ' >
i
•
J.
-> ' T
' v*- '
1 \ v1
! ~n
, i
i i
1 T
1
|
1
l-r-
1
1
1
1
1
1
|
i
1
1
1
I
| —
1
T
1
1
I
iT
i*
i
1
T |
! T i
1 T
|_
J.
|
1
I
1
1
I
I-J"
I
- t
1 ~
j
1
r
u
i
\
~i
T
1
I
1 -
1
1
1
1
1
1
1
1
i
v
\
•
,
*
i j
. *
1~~
-
T
1
1
t?
1 -i-
±
_,
IT
i -
-*-
T
1
1
1 .
1 i
i""
T
i
ii
••
13 14 32 33 46 47 48 49 57 78 81 86
Test Number
Figure 7-1.
LC50's and confidence limits using in-
verse, probit maximum likelihood (MLH), and
quadratic regression with arc sin square root
transformation of dependent variables.
40
-------
duckweed data appear to suffer from the latter cause which
is not corrected by use of the h factor. However, the
computer program cannot distinguish the cause of the
large x2 values, so it automatically calls the subroutine
to apply the h factor. The h factor extends the confi-
dence limits and thus makes the limits closer to what
they should be. However, the h factor does not correct
for the lack of model fit and should not be used to do so.
The use of the h factor does not cause consistent over-
or underestimation of the confidence limits (Figure 7-1).
Little if any faith should be placed in confidence inter-
vals calculated in this manner. The heterogeneity factor
would have been eliminated in the analysis, but the SAS
PROC PROBIT procedure used does not allow for overriding
the heterogeneity subroutine (SAS 1979).
7.1. CuSOij
All CuSOif results are based on the mg/L of CuSCK
added as CuSCK-B^O. The approximate CuSCK toxicity can
be estimated by multiplying the LC50 by 0.64 which is the
percentage of CuSOi* in CuSOij-S^O. The approximate Cu++
toxicity can be estimated by multiplying by 0.25. Con-
centrations used in the paired tests were 1.68, 3.0, 5.4,
9.6, 16.8, and 30.0 mg/L CuS04-5H20.
Paired test results yielded mean LCSO's of 6.76 mg/L
and 11.71 mg/L for flow-through and static-replacement
tests, respectively (Table 7-3). An ANOVA indicated that
differences between LC50's for static-replacement and
flow-through tests were significant at the 0.03 proba-
bility level (Table 7-4). Flow-through test LCSO's
varied slightly more than static LCSO's. Coefficients
of variation were 0.10 and 0.06 for flow-through and
static-replacement tests, respectively. However, in both
types of tests, the within test and between tests varia-
tion was proportioned similarly (Table 7-5). Approxi-
mately 84% of the variation occurred between tests.
LCSO's for flow-through tests ranged from 3.67 to
10.46 mg/L (Table 7-3). The 95% confidence interval for
the mean of the flow-through test LCSO's ranged from 3.27
to 13.96 mg/L. These were within 38% of the mean LC50
(6.76 mg/L) in the log scale. The LCSO's in the absolute
scale are unsymmetrical because they were transformed
from the log scale. The confidence interval must be cal-
culated in the log scale, because it is incorrect to con-
vert variances calculated for transformed data back to the
absolute scale (Snedecor and Cochran 1967).
Static-replacement test LCSO's ranged from 7.03 to
15.0 mg/L (Table 7-3). The 95% confidence interval ranged
41
-------
Table 7-3. LCSO's from Weighted Quadratic Regression with
Arc Sin Square Root Transformation of Proportion
of Dead Fronds to Total Fronds, Day 7.
Toxicant
CuSCH
2,4,6-TCP
o-cresol
Test
32
33
46
47
48
49
85
86
18
19
35
36
54
55
64
65
74
75
88
89
94
95
42
43
51
52
No.
type1
F
S
F
S
F
S
F
S
F
S
F
S
F
S
F
S
F
S
F
S
F
S
F
S
F
S
LC50
8.84
12.49
6.25
15.00
10.46
14.28
3.67
7.03
0.87
0.93
1.85
2.01
1.68
1.45
2.75
2.92
2.02
2.04
2.18
2.85
3.12
3.11
495.44
498.05
577.65
562.00
LCL2
3.60
7.34
2.98
7.27
5.07
9.54
2.17
3.50
_
-
-
0.64
0.61
-
0.69
1.41
1.10
232.56
228.49
304.75
311.66
UCL3
23.08
19.14
13.91
26.48
17.99
20.11
7.34
13.31
3.62
3.18
3.83
4.02
3.27
5.31
5.80
4.45
4.58
4.43
5.21
5.25
5.62
851.83
865.05
-
42
-------
Table 7-3 (Continued).
Toxicant
,
LC50
LCL
UCL
60
61
70
71
ethylene glycol1* 37
39
45
50
66
F
S
F
S
S
S
S
S
S
440.33
691.74
382.60
436.87
17006
29794
28287
40139
21184
254.01
329.49
217.18
246.85
3099
16602
9935
26405
1948
653
563
48016
49353
-
-
—
.79
.64
1paired tests conducted during the same time period
F = flow-through test
S = static-replacement test
2LCL = lower confidence limit
3UCL = upper confidence limit
**no flow-through tests conducted
43
-------
Table 7-4.
ANOVA Table Comparing CuSCU Flow-through and
Static-Replacement Paired Tests Conducted
During the Same Time Period. Data entered
were logi0 LCSO's generated by quadratic
regression with arc sin square root
transformation.
Dependent Variable: logio LC50
Source
df
Sum of Squares
Mean Square
Model
Error
Corrected Total
7
24
31
1.22101261
0.13997257
1.36098518
0.17443037
0.00583219
Source
df
Type I SS
F Value
PR > F
Type
Date
Type-Date
0.45655418
0.67673112
0.08772731
78.28
38.68
5.01
Tests of Hypotheses Using the Type IV MS for Type-Date as an Error Term
Source df Type IV SS F Value
0.0001
0.0001
0.0077
PR
Type
Date
0.45655418
0.67673112
15.61
7.71
0.0289
0.0637
44
-------
Table 7-5. Nested ANOVA to Determine Between Tests
and Within Test Variation for CuSCK Tests.
Data entered were logic LCSO's generated by
quadratic regression with arc sin square root
transformation.
Paired Flow-through and Static-Replacement Tests
Variance
Source df
Sum of Mean
Squares Squares
Total 31 1.36099 0.04390
Test No. 7 1.22101 0.17443
Error 24 0.13997 0.00583
Mean 0.949164
Standard Deviation 0.076369
Coefficient of Variation 0.080459
Flow-through
Variance
Source df
Tests
Sum of Mean
Squares Squares
Total 15 0.55972 0.03731
Test No. 3 0.46940 0.15647
Error 12 0.09032 0.00753
Mean 0.829718
Standard Deviation 0.086756
Coefficient of Variation 0.104561
Variance
Component Percent
0.04798 100.00
0.04215 87.84
0.00583 12.16
Variance
Component Percent
0.04476 100.00
0.03723 83.19
0.00753 16.81
Static-Replacement Tests
Variance
Source df
Sum of Mean
Squares Squares
Total 15 0.34471 0.02298
Test No. 3 0.29506 0.09835
Error 12 0.04965 0.00414
Mean 1.068610
Standard Deviation 0.064326
Coefficient of Variation 0.060196
Variance
Component Percent
0.02769 100.00
0.02355 85.06
0.00414 14.94
45
-------
from 6.59 to 20.80 mg/L, which is within 23% of the mean
LC50. The static-replacement test LCSO's were generally
higher than the flow- through test LCSO's. The contract
did not allow research into the causes for this phenomenon.
A possible explanation is that the fronds in the static-
replacement tests were able to absorb and sequester a
portion of the Cu++. The reproductive rate of L. gibba
results in a doubling of the number of fronds every two
days. The new fronds may be able to absorb a portion of
the replaced Cu++ which was changed every three days.
Hutchinson (1975) stated that L. minor from natural
waters in New Jersey which contained 0.009 mg/L Cu++ con-
tained 32.5 ppm Cu++. Thus plants apparently are able
to take up and store some Cu++ without a detrimental
effect. If this sequestering of the Cu++ occurred in
the bioassays, the actual concentration of Cu++ in solu-
tion would decrease. This was not verified because no
chemical measurements of toxicant concentration were
allowed under this contract.
The data and Equations 6-11 and 6-12 were used to
calculate the number of tests and replicates necessary to
obtain confidence limits within a given percentage of the
mean in log units (Table 7-6) . Given the fact that the
ANOVA indicated a significant difference between types
(Table 7-4) , separate predictions for flow- through and
static- replacement tests probably should be used. How-
ever, given the closeness of the projections, the com-
bined projection may be adequate (Table 7-6) . The pro-
jection for obtaining confidence limits within 25% would
be most appropriate given the magnitude of the mean and
the costs of conducting the additional tests and repli-
cates required to narrow the limits to within 10% of the
mean LC50 (Table 7-6) .
was also used to assess the effect of fewer
fronds per test chamber in three sets of paired static-
replacement tests. One test in each pair contained 4
fronds per test chamber (Type I) and the other test con-
tained 12 fronds per chamber (Type II) . The Type I and
Type II LCSO's were 7.62 and 6.84 mg/L, respectively.
An ANOVA indicated a significant difference in these
LCSO's at the 0.55 probability level. Thus based on
CuSOi, tests, reducing the number of organisms from 12
fronds to 4 fronds would not cause a loss in information.
However, such a reduction would not necessarily result
in a significant cost savings as the number of test cham-
bers, amount of media, and labor needed to conduct the
tests would be the same, regardless of the number of
organisms used.
46
-------
Table 7-6.
Prediction of Minimum Number of Tests and Replicates Necessary to
Obtain Confidence Limits Within 10%, 25%, and 50% of Mean at a 95%
Probability Level (using logio LCSO's from quadratic regression with
arc sin square root transformation for paired flow-through and static-
replacement tests only).
Toxicant
CuSOi,
2,4, 6-TCP
o-cresol
ethylene glycol
Test Type
F & S
F
S
F & S
F
S
F & S
F
S
S
k
21
24
11
-120
-105
-150
3
3
3
3
10% r
5
6
3
5
11
5
1
1
1
2
k 25%
6
6
4
21
19
26
3
3
3
3
r
1
5
1
6
7
5
1
1
1
1
k
4
4
3
7
7
9
2
2
2
2
50% r
1
1
1
19
3
2
1
1
1
1
0
0
0
0
0
0
0
0
0
0
T2
.04215
.03723
.02355
.02741
.02484
.03472
.00666
.00610
.00665
.02369
0
0
0
0
0
0
0
0
0
0
s2
.00583
.00753
.00414
.00794
.00738
.00848
.00124
.00116
.00133
.01121
X
0.95
0.83
1.07
0.31
0.31
0.31
2.70
2.67
2.73
4.41
F = flow-through test
S = static-replacement test
k = number of tests
r = number of replicates
T2 = estimation of T2 which is between test
variation
s2 = estimation of a2 which is between replicates
within tests variation
x = mean logjo (LC50)
-------
7.2. 2,4,6-TCP
Concentrations for paired definitive flow-through and
static-replacement tests were 0, 0.34, 0.6, 1.10, 1.92,
3.36, and 6.0 mg/L. Additional concentrations were added
to some static tests, but were not used in comparing static-
replacement and flow-through tests. The paired static-
replacement and flow-through tests had mean LCSO's of 2.03
and 2.04 mg/L, respectively; an ANOVA indicated a statis-
tical difference between these numbers at the 0.64 proba-
bility level (Table 7-7). Given the high volatility of
TCP, a greater difference in static-replacement and flow-
through tests would be expected. The contract did not
allow for research into the reason for this lack of dif-
ference. Static-replacement test LCSO's ranged from a
low of 0.93 mg/L to a high of 3.11 mg/L TCP, while flow-
through test LCSO's ranged from 0.87 to 3.12 mg/L TCP
(Table 7-3). The 95% confidence intervals for static-
replacement and flow-through tests were 1.36 to 3.08 mg/L
and 1.45 to 2.88 mg/L, respectively. Actual LCSO's cal-
culated for both types of tests exceeded these confidence
limits (Table 7-3) .
Although nested ANOVA1s (Table 7-8) were calculated
separately for static-replacement and flow-through tests,
the combined nested ANOVA should be used to assess the
distribution of variance components because 1) an ANOVA
indicated no differences between LCSO's for flow-through
and static-replacement tests (Table 7-7) , and 2) the com-
bined data yield more degrees of freedom. The combined
data indicated 78% of the variation occurred among tests
as opposed to 22% within tests (Table 7-8). The actual
LCSO's support this breakout (Table 7-3). The LCSO's for
paired tests are much closer together than the LCSO's for
tests conducted at different times. The nested ANOVA
(Table 7-8) indicates that the between tests variation
is approximately 3X the within test variation.
Equations 6-11 and 6-12 and data from the nested
ANOVA indicated the need for 21 tests with 6 replicates
to obtain 95% confidence limits within 25% of the mean
LC50 (Table 7-6). The confidence limits calculated using
14 tests and 4 replicates were within 31% of the mean
LC50. Given the magnitude of the mean LC50 (2.0 mg/L),
confidence limits within 50% of the mean should be ac-
ceptable. To obtain these confidence limits, 7 tests
with 19 replicates would be needed (Table 7-6). It is
more cost-efficient in terms of time and labor to conduct
7 tests with 19 replicates than to conduct 21 tests with
6 replicates because the tests need to be conducted at
different points in time. Between tests variability is
a function of culture age, environmental conditions, and
experimental techniques and thus depends on the time
48
-------
Table 7-7. ANOVA Table Comparing 2,4,6-TCP Flow-through and
Static-Replacement Paired Tests Conducted During
the Same Time Period. Data entered were logi0
LCSO's generated by quadratic regression with
arc sin square root transformation.
Dependent Variable: logio LC50
Source df Sum of Squares Mean Square
Model 13 1.50256427 0.11558187
Error 41 0.32571568 0.00794428
Corrected Total 54 1.82827995
Source df Type I SS F Value PR > F
Type 1 0.00002411 0.00 0.9563
Date 6 1.45874967 30.60 0.0001
Type-Date 6 0.04379048 0.92 0.4915
Tests of Hypotheses Using the Type IV MS for Type-Date as an Error Term
Source df Type IV SS F Value PR > F
Type 1 0.00176607 0.24 0.6403
Date 6 1.44327740 32.96 0.0002
49
-------
Table 7-8. Nested ANOVA to Determine Between Tests
and Within Test Variation for 2,4,6-TCP
Tests. Data entered were logi0 LCSO's
generated by quadratic regression with
arc sin square root transformation.
Paired Flow-through and Static-Replacement Tests
Variance Sum of Mean
Source df Squares Squares
Total 54 1.82828 0.03386
Test No. 13 1.50256 0.11558
Error 41 0.32572 0.00794
Mean 0.308557
Standard Deviation 0.089131
Coefficient of Variation 0.288863
Variance
Component
0.03535
0.02741
0.00794
Percent
100.00
77.53
22.47
Flow-through Tests
Variance Sum of Mean
Source df Squares Squares
Total 26 0.76598 0.02946
Test No. 6 0.61833 0.10306
Error 20 0.14765 0.00738
Mean 0.309231
Standard Deviation 0.085920
Coefficient of Variation 0.277851
Variance
Component
0.03222
0.02484
0.00738
Percent
100.00
77.09
22.91
Static-Replacement Tests
Variance Sum of Mean
Source df Squares Squares
Total 27 1.06228 0.03934
Test No. 6 0.88421 0.14737
Error 21 0.17807 0.00848
Mean 0.307906
Standard Deviation 0.092084
Coefficient of Variation 0.299066
Variance
Component
0.04320
0.03472
0.00848
Percent
100.00
80.37
19.63
50
-------
period(s) in which tests are conducted. When tests are
conducted during a single time period, under the same ex-
perimental conditions and using organisms from the same
group of stock cultures, the results appear as a single
large test with massive replication. If tests must be
conducted during the same time period, then stock cultures
of various ages should be used and tests should be con-
ducted in separate growth chambers so that between tests
variability approximates that for tests conducted at dif-
ferent times.
7.3. 0-Cresol
Concentrations for paired flow-through and static-
replacement tests were adjusted during the study period
to obtain the best estimate of the LC50 and ECSO's.
Paired tests conducted during the same time period had
the same concentrations. The concentration ranges, ex-
cluding the control, were 56-1000 mg/L, 45-800 mg/L,
81-700 mg/L, and 84-625 mg/L for tests 42-43, 51-52,
60-61, and 70-71, respectively.
The static-replacement and flow-through test mean
LC50's were 540 and 464 mg/L, respectively; the ANOVA in-
dicated these were statistically different at the 0.27
probability level (Table 7-9). However, the high vola-
tility of o-cresol suggests that there should be a greater
difference between static-replacement and flow-through
tests. This lack of difference in LCSO's between test
types also occurred with 2,4,6-TCP (Section 7.2), but,
as noted previously, the contract did not allow research
to determine the reason. It is possible that the re-
placement schedule was sufficient to prevent a signifi-
cant decline in the concentrations in the static-replace-
ment test chambers.
The overall mean LC50 based on the combined static-
replacement and flow-through tests was 501 mg/L with a
95% confidence interval of 427 and 589 mg/L. In the log
scale these limits are within 3% of the mean. Given the
magnitude of the mean, confidence limits within 10% of
the mean would probably be satisfactory. Three tests
of one replicate each would be adequate to obtain this
level of precision (Table 7-6) . However, conducting
tests with only one replicate is not recommended. Rep-
licates should be used to allow calculation and comparison
of LCSO's and calculation of variances within tests.
Additionally, confidence intervals for each test cannot
be calculated without replication.
Because more tests were conducted than were actually
required, according to the predictions calculated
51
-------
Table 7-9.
ANOVA Table Comparing 0-Cresol Flow-through and
Static-Replacement Paired Tests Conducted During
the Same T^ime Period. Data entered were logic
LCSO's generated by quadratic regression with arc
sin square root transformation.
Dependent Variable: logic LC50
Source
Model
Error
Corrected Total
Source
Type
Date
Type -Date
df
7
23
30
df
1
3
3
Sum of Squares
0.19387378
0.02931974
0.22319352
Type I SS
0.03870274
0.10178716
0.05338389
Mean Square
0.02769625
0.00127477
F Value PR > F
30.36 0.0001
26.62 0.0001
13.96 0.0001
Tests of Hypotheses Using the Type IV MS for Type-Date as an Error Term
Source
df
Type IV SS
F Value
PR > F
Type
Date
0.03161614
0.10620705
1.78
1.99
0.2747
0.2932
52
-------
(Equations 6-11 and 6-12) and data in Table 7-10, the
actual test data were used to verify the prediction. A
random numbers table was used to construct ten sets of
three test groups with one replicate each from LCSO's cal-
culated for each replicate of each test. The three LCSO's
were averaged and a standard error was calculated. The
average confidence limits for the ten groups were within
9.5% of the mean, thus confirming the number of tests and
replicates predicted (Table 7-6).
7.4. Ethylene Glycol
Initial range-finding tests indicated definitive test
concentrations should range from 0 to 56,000 mg/L. The
concentrations used in all tests were 0, 1792, 3136, 5600,
10,080, 13,440, 17,920, 23,520, 31,360, and 56,000 mg/L.
No flow-through tests were conducted using ethylene
glycol because of the relatively high effect thresholds.
The amount of chemical used to produce the required con-
centration range in flow-through tests was cost-prohibi-
tive. However, static-replacement tests were conducted
to determine the LC50 and ECSO's. The mean LC50 based on
five 9-concentration, 4-replicate tests was 25,760 mg/L.
The minimum and maximum LCSO's generated were 17,006 and
40,139, respectively (Table 7-3). The 95% confidence
limits for the mean LC50 were 16,360 and 40,560, which
were within 4% of the mean; all of the LCSO's generated
with replicates pooled fell within these confidence
limits.
The nested ANOVA indicated that 68% of the variation
in ethylene glycol experiments occurred among tests as
opposed to 32% which occurred within tests (Table 7-11) .
Calculations using Equations 6-11 and 6-12 indicated that
to insure confidence limits within 25% of the mean, in
log units, three tests with one replicate would be suffi-
cient. However, the use of only one replicate per test
is not recommended. A minimum of two replicates should
be used to provide checks and to compare LCSO's. The
replicates can be pooled if no differences are noted in
the replicate LCSO's or raw data. Furthermore, with an
additional replicate, Equation 6-12 indicates that the
confidence limits should decrease to within 10% of the
mean LC50 (Table 7-6).
7.5. Di(2-ethylhexyl)phthalate
Initial range-finding tests indicated the upper con-
centration for definitive tests should be approximately
53
-------
Table 7-10. Nested ANOVA to Determine Between Tests
and Within Test Variation for O-Cresol
Tests. Data entered were Iog10 LC50's
generated by quadratic regression with
arc sin square root transformation.
Paired Flow-through and Static-Replacement Tests
Variance
Source df
Sum of Mean
Squares Squares
Total 31 0.22490 0.00725
Test No. 7 0.19507 0.02787
Error 24 0.02983 0.00124
Mean 2.699667
Standard Deviation 0.035257
Coefficient of Variation 0.013060
Flow-through
Variance
Source df
Tests
Sum of Mean
Squares Squares
Total 15 0.09056 0.00604
Test No. 3 0.07668 0.02556
Error 12 0.01388 0.00116
Mean 2.666768
Standard Deviation 0.034010
Coefficient of Variation 0.012753
Variance
Component
0.00790
0.00666
0.00124
Variance
Component
0.00726
0.00610
0.00116
Percent
100.00
84.26
15.74
Percent
100.00
84.06
15.94
Static-Replacement Tests
Variance
Source df
Sum of Mean
Squares Squares
Total 15 0.09971 0.00665
Test No. 3 0.08376 0.02792
Error 12 0.01595 0.00133
Mean 2.732566
Standard Deviation 0.036461
Coefficient of Variation 0.013343
Variance
Component
0.00798
0.00665
0.00133
Percent
100.00
83.33
16.67
54
-------
Table 7-11.
Nested ANOVA to Determine Between Tests
and Within Test Variation for Ethylene
Glycol Static-Replacement Tests. Data
entered were Iog10 LCSO's generated by
quadratic regression with arc sin square
root transformation.
Variance
Source
Total
Test No.
Error
df
18
4
14
Sum of
Squares
0.56084
0.40394
0.15690
Mean 4 .
Standard Deviation 0.
Coefficient of Variation 0.
Mean
Squares
0.03116
0.10099
0.01121
410965
105864
024000
Variance
Component
0.03490
0.02369
0.01121
Percent
100.00
67.89
32.11
55
-------
200,000 mg/L. Given this extremely high concentration
range, and consequently the high mortality threshold
concentration, mortality tests were not conducted. There-
fore, the concentration ranges were set to provide data
on reproduction and dry weight.
56
-------
8.0 REPRODUCTION DATA ANALYSIS
An EC50 based on reproduction could not always be
calculated using the concentration necessary for assess-
ment of the LCSO's. However, it was possible to calcu-
late ECSO's based on reproduction for some tests (Table
8-1). Two methods were used to calculate ECSO's (Sec-
tion 6.2). A comparison of the ECSO's generated by the
two methods indicated that the first method, based on
frond numbers, yielded a lower EC50 in 92% of the cases
than the second, based on the growth rate constant
(Table 8-1). However, comparison of the coefficients
of determination and the variance components of the two
test types indicated less variation and thus more preci-
sion when the growth rate constant was used. Therefore,
subsequent analyses were based on ECSO's calculated
using the growth rate constant.
8.1.
The concentrations used to assess the ECSO's based
on reproduction were the same as those used for calcula-
tion of the LCSO's (Section 7.1). As with the LC50, the
paired static-replacement tests yielded a higher EC50
(x = 3.51 mg/L) than the flow-through tests (x = 2.21
mg/L). Sequestering of a portion of the Cu++ by the
fronds was offered as a possible explanation (Section 7.1).
However, an ANOVA indicated that the static-replacement
and flow-through test ECSO's were not statistically dif-
ferent (Table 8-2), due to the relatively large variation
within tests.
Variation within tests is due largely to the repro-
ductive nature of the organisms. The new fronds form in-
side two small pockets on each side of the mother frond
(Hillman 1961a). It is impossible to determine the stage
of development of these fronds without harming the frond.
Therefore, fronds selected for the tests may have daughter
fronds ready to come forth or may have no daughter fronds.
If daughter fronds are present, the generation time is
shorter for the first generation, and more mother fronds
are available to produce additional fronds. This effect
is multiplied after several generations and results in
different numbers of fronds in each test at the end of
the test period, even if there is no effect of the
57
-------
Table 8-1.
Comparison of ECSO's from Quadratic Regression
for Paired Flow-through and Static-Replacement
Tests: Square Root Transformation versus Growth
Rate Constant (K) Transformation.
Test
Toxicant and
CuSCK 13
14
32
33
46
47
48
49
85
86
2,4,6-TCP 18
19
35
36
54
55
64
65
74
75
88
89
94
95
No.
Type1
F
S
F
S
F
S
F
S
F
S
F
S
F
S
F
S
F
S
F
S
F
S
F
S
EC50
/fronds
0.80
0.271
1.29
0.81
0.69
2.08
0.09
2.60
1.29
1.583
<0.013
0.02
<0.013
<0.013
<0.013
0.23"
<0.013
1.78
<0.01
0.043
0.02
<0.01
0.193
2
K
2.32
2.23
2.67
2.98
1.88
4.10
1.32
6.03
2.34
3.68
0.023
0.08
0.023
0.013
0.023
0.11
0.033
0.95
0.03
0.143
0.10
0.03
0.173
58
-------
Table 8-1 (Continued).
Test
Toxicant and
o-cresol 10
11
42
43
51
52
60
61
70
71
No.
Type1
F
S
F
S
F
S
F
S
F
S
/ fronds
82.5
230.5
90.5
135.5
101.6
168.2
104.3
286.2
118.5
137.4
EC502
K
143.7
260.4
158.4
193.1
155.6
237.8
158.5
370.1
139.1
230.9
ethylene glycol
37
39
45
50
66
S
S
S
S
S
13320
8950
10680
20000
35980
17140
15310
18970
27460
13160
di(2-ethylhexyl)
phthalate
40
41
44
53
63
90
93
S
S
S
S
S
S
S
171. 55
117.6
502.6
569.5
3071.2
136.3
365.1
735.1
663.5
7492.1
2495.6
7469.9
408.3
5489.5
1 Coupled entries indicate paired tests conducted during the
same time period.
F = flow-through test
S = static-replacement test
2 /fronds = square root of total fronds Day 7
K = Iog10(total fronds Day 7) - Iogi0(total fronds Day 0)
3 one replicate
4 average of two replicates
5 excluding cone = 6750 (otherwise EC50 = 0.84)
59
-------
Table 8-2, ANOVA Table Comparing CuSOi* Flow-through
and Static-Replacement Paired Tests Con-
ducted During the Same Time Period. Data
entered were logio ECSO's generated by
quadratic regression with growth rate con-
stant transformation .
Dependent Variable: logic EC50
Source
Model
Error
Corrected Total
Source
Type
Date
Type -Date
df
9
28
37
df
1
4
4
Sum of Squares
0. 94449280
0.27369970
1.21819250
Type I SS
0.37989297
0.18651996
0.37797986
Mean Square
0.10494364
0.00977499
0.02763584
F Value PR
38.86 0.
4.77 0.
9.67 0.
> F
0001
0046
0001
Tests of Hypotheses Using the Type IV MS for Type-Date as an Error Term
Source df Type IV SS F Value
PR > F
Type
Date
0.38353116
0.10944848
4. 06
0.29
0.1142
0.8714
60
-------
toxicant. For example, the static-replacement controls
over all paired tests varied from 36 to 159 fronds per
chamber. The mean was 100 with a standard error of 8.3.
The flow- through tests averaged 138 fronds with a stan-
dard error of 13.3 and ranged from 41 to 223 fronds per
chamber.
The number of tests and replicates needed to obtain
confidence limits within 10%, 25%, and 50% of the mean
were calculated using Equations 6-11 and 6-12 and the
data from the nested ANOVA's (Table 8-3). Although the
ANOVA testing for differences between types (Table 8-2)
indicated significant differences only above the 0.10
probability level, the predictions based on the separated
static-replacement and flow-through tests probably should
be used, as the predictions with the data pooled appear
biased by the static-replacement data. Additionally,
static-replacement and flow-through tests indicate oppo-
site distributions of the variation (Table 8-3) . In
flow- through tests the within test variation is approxi-
mately 3X the between tests variation while in static-
replacement tests the opposite appears true.
The predictions of the number of tests and replicates
needed to yield confidence limits within 25% of the mean
EC50 were the most reasonable (Table 8-4) . To decrease
the confidence limits to within 10% of the mean would not
be cost-effective.
was also used to determine if the number of
fronds per test chamber had a significant effect on growth
rates in static- replacement tests. Three sets of paired
tests with four replicates each were conducted, and an
ANOVA was calculated using ECSO's for each replicate. The
ANOVA indicated a significant difference between 4-frond
and 12- frond chambers at the 0.60 probability level. The
mean EC50 for both 4 and 12-frond tests was 2.75 mg/L.
The 12-frond tests had slightly higher between tests vari-
ation (78%) in the replicate ECSO's than the 4-frond tests
(61%), based on the nested ANOVA variance components.
8.2. 2,4,6-TCP
The same concentration ranges that were used for the
LC50 calculations were used to calculate the ECSO's for
2,4,6-TCP (Section 7.2). Although the quadratic regres-
sion models fit the data relatively well, the solutions
to the regression equations often yielded imaginary roots.
The imaginary roots occurred when the entire curve was
above or below the 50% effect level. The imaginary roots
are, of course, completely invalid for this type of analy-
sis and were treated as missing values in the data analysis,
61
-------
Table 8-3.
Nested ANOVA to Determine Between Tests and
Within Test Variation for CuSOi* Tests. Data
entered were logio ECSO's generated by quadratic
regression with growth rate constant transformation.
Paired Flow-through and Static-Replacement Tests
Variance Sum of Mean
Source df Squares Squares
Total 37 1.21819 0.03292
Test No. 9 0.94449 0.10494
Error 28 0.27370 0.00977
Mean 0.450277
Standard Deviation 0.098869
Coefficient of Variation 0.219573
Variance
Component
0. 03489
0.02511
0.00977
Percent
100.00
71.98
28.02
Flow-through Tests
Variance Sum of Mean
Source df Squares Squares
Total 17 0.24780 0.01458
Test No. 4 0.10490 0,02622
Error 13 0.14290 0.01099
Mean 0.344883
Standard Deviation 0.104844
Coefficient of Variation 0.304000
Variance
Component
0.01528
0.00428
0. 01099
Percent
100.00
28.04
71.96
Static-Replacement Tests
Variance Sum of Mean
Source df Squares Squares
Total 19 0.59050 0.03108
Test No. 4 0.45970 0.11493
Error 15 0.13080 0.00872
Mean 0.545132
Standard Deviation 0.093381
Coefficient of Variation 0.171299
Variance
Component
0.03527
0.02655
0.00872
Percent
100.00
75.28
24.72
62
-------
Table 8-4.
Prediction of Minimum Number of Tests and Replicates Necessary
to Obtain Confidence Limits Within 10%, 25%, and 50% of Mean at
a 95% Probability Level (using logi0 ECSO's from quadratic regression
with growth rate constant transformation for paired flow-through
and static-replacement tests only).
u>
Toxicant Test Type
CuSOn F
2,4,6-TCP F
o-cresol F
ethylene glycol
di(2-ethylhexyl)
phthalate
F = flow-through test
S = static- replacement
k = number of tests
& S
F
S
& S
F
S
& S
F
S
S
S
test
k
-50
17
-38
-55
2
>120
4
2
4
3
11
10%
r
1
40
10
102
3711
-
3
7
1
1
35
25%
k
10
5
9
11
2
-45
3
2
3
2
4
r
1
21
2
74
594
11
1
2
1
7
6
T2
S2
50%
k
5
3
4
5
2
13
3
2
2
2
3
r
2
22
4
16
149
13
1
1
2
1
1
= estimation
variation
= estimation
0
0
0
0
0
0
0
0
0
0
0
of T2
of a2
T2
.0251
.0043
.0266
.1837
.4966
.0185
.0144
.0124
.2385
which
which
s2
0.0098
0.0110
0.0087
0.6236
1.1784
0.3461
0.0043
0.0041
0.0045
0.0095
0.1652
is between
is between
X
0.4503
0.3449
0.5451
-1.1780
-1.6011
-0.8757
2.2849
2.1794
2.3905
4.2345
3.3138
test
r = number of replicates
replicates within test variation
x = mean login (EC50)
-------
Because of the number of replicates for which an EC50
could not be calculated, the data set for assessing means,
confidence limits, and differences between types is lim-
ited. Therefore, all conclusions regarding reproduction
in 2,4,6-TCP tests should be considered carefully. The
following analysis of the data is presented as a best es-
timate of the effect of 2,4,6-TCP on reproduction.
The mean ECSO's for static-replacement and flow-
through TCP tests were 0.13 and 0.03 mg/L, respectively-
The ANOVA testing for differences between static-replace-
ment and flow-through tests indicated no significant dif-
ference until the 0.10 probability level (Table 8-5).
Assuming acceptance of the 0.05 level as the breakpoint
for determining if the two types of tests were different,
the ECSO's for the two test types can be pooled. Pooling
is preferable because of the limited amount of valid rep-
licate data for TCP. The pooled data ranged from ECSO's
of 0.01 to 0.95 mg/L. The 95% confidence limits about
the pooled mean of 0.07 mg/L were 0.02 and 0.19 mg/L.
The predicted number of tests and replicates were high
even at the 25% and 50% levels (Table 8-4). The nested
ANOVA showed approximately 3X more variation between rep-
licates than between tests (.Table 8-6) , thus indicating
the need for high replication. However, this may be an
artifact of the data because of the few replicate ECSO's
within tests.
1.3. 0-Cresol
The concentration ranges used to calculate ECSO's
were the same as those used in the LC50 calculations (Sec-
tion 7.3). ECSO's for paired static-replacement and flow-
through tests were statistically different at the 0.05
probability level (Table 8-7). The mean ECSO's were 245
and 151 mg/L for static-replacement and flow-through
tests, respectively. In all paired tests the static-re-
placement EC50 was higher than the corresponding flow-
through test EC50 (Table 8-1), due possibly to the vola-
tility of o-cresol. However, the LCSO's for o-cresol
static-replacement and flow-through tests were statis-
tically different at or above the 0.27 probability level
(Table 7-9). The paired tests yielded LCSO's that were
almost identical in one case (tests 42-43), and in another
the flow-through test LC50 was higher than the static-
replacement test LC50 (tests 51-52, Table 7-3). Thus
volatility may not be the cause of the statistically
different ECSO's for o-cresol.
Static-replacement test ECSO's ranged from 193 to 370
mg/L o-cresol (Table 8-1). The 95% confidence limits were
64
-------
Table 8-5.
ANOVA Table Comparing 2,4,6-TCP Flow-through and
Static-Replacement Paired Tests Conducted During
the Same Time Period. Data entered were logic
ECSO's generated by quadratic regression with
growth rate constant transformation.
Dependent Variable; logio EC50
Source
df
Sum of Squares
Mean Square
Model
Error
Corrected Total
11
12
23
10.77867574
7.48257155
18.26124729
0.97987961
0.62354763
Source
df
Type I SS
F Value
PR > F
Type
Date
Type-Date
3.06962659
5.50123692
2.20781223
4.92
1.76
0.71
0.0465
0.1948
0.6286
Tests of Hypotheses Using the Type IV MS for Type-Date as an Error Term
Source
df
Type IV SS
F Value
PR > F
Type
Date
1.75579474
2.11242792
3.98
0.96
0.1027
0.5187
65
-------
Table 8-6.
Nested ANOVA to Determine Between Tests and
Within Test Variation for 2,4,6-TCP Tests.
Data entered were logic ECSO's generated by
quadratic regression with growth rate constant
transformation.
Paired Flow-through and Static-Replacement Tests
Variance
Source
Total
Test No.
Error
df
23
11
12
Sum of
Squares
18.
10.
7.
Mean
Standard Deviation
Coefficient of Variation
26125
77868
48257
-1.
0.
-0.
Mean Variance
Squares Component
0.79397 0.80728
0.97988 0.18373
0.62355 0.62355
177989
789650
670338
Percent
100.00
22. 76
77.24
Flow-through Tests
Variance
Source
Total
Test No.
Error
df
9
5
4
Sum of
Squares
5.22965
0.51602
4.71363
Mean -1.
Standard Deviation 1.
Coefficient of Variation -0.
Mean
Squares
0.58107
0.10320
1.17841
601146
085545
677980
Variance
Component
1.17841
-0.68923
1.17841
Percent
100.00
0.0
100.00
Static-Replacement Tests
Variance
Source
Total
Test No.
Error
Mean
Standard
df
13
5
8
Deviation
Sum of
Squares
9.96197
7.19303
2.76894
-0.
0.
Mean
Squares
0.76631
1.43861
0.34612
,875734
,588318
Variance
Component
0.84270
0.49659
0.34612
Percent
100.00
58.93
41.07
Coefficient of Variation -0.671799
66
-------
Table 8-7.
ANOVA Table Comparing 0-Cresol Flow-through and
Static-Replacement Paired Tests Conducted During
the Same Time Period. Data entered were logio
ECSO's generated by quadratic regression with
growth rate constant transformation.
Dependent Variable: login EC50
Source
df
Sum of Squares
Mean Square
Model
Error
Corrected Total
7
23
30
0.52908719
0.10087716
0.62996435
0.07558388
0.00438596
Source
df
Type I SS
F Value
PR > F
Type
Date
Type. Date
1
3
3
0.33733763
0.10033407
O.D9141549
76.91
7.63
6.95
0.0001
0.0010
0.0017
Tests of Hypotheses Using the Type IV MS for Type-Date as an Error Term
Source
df
Type I SS
F Value
PR > F
Type
Date
0.33094854
0.10001228
10,86
1.09
0.0459
0.4714
67
-------
156 to 387 mg/L. In the log scale, the confidence interval
obtained using 4 tests with 4 replicates each was within 8%
of the mean. The nested ANOVA (Table 8-8) and Equations
6-11 and 6-12 indicated 4 tests with 1 replicate would give
confidence limits within 10% of the mean (Table 8-4).
The ECSO's for flow-through tests ranged from 139 to
159 mg/L o-cresol (Table 8-1). The 95% confidence limits
were 140 and 162 mg/L which were within 2% of the mean in
the transformed units. These confidence intervals were
obtained using 4 tests with 4 replicates each. The predic-
tions indicated that 2 tests with 7 replicates each would
be adequate to obtain confidence limits within 10% of the
mean (Table 8-4).
8.4. Ethylene Glycol
Ethylene glycol was used only in static-replacement
tests; no flow tests were conducted (Section 7.4). Con-
centrations used for calculation of the EC50's were the
same as those used for the LC50's (Section 7-4). As with
mortality, ethylene glycol did not appear to affect growth
rates until it reached relatively high concentrations. In
fact, the lowest concentration used (1800 mg/L) actually
stimulated growth in most tests.
The ECSO's averaged 17,200 mg/L and ranged from 13,160
to 27,460 mg/L (Table 8-1). An ANOVA indicated significant
differences between test ECSO's at the 0.004 probability
level. The nested ANOVA indicated that the variance com-
ponents were relatively evenly distributed with 57% between
tests and 43% within tests (Table 8-9).
The 95% confidence limits about the mean were 12,000
and 24,000 mg/L. These limits are within 4% of the mean.
The predicted number of tests and replicates needed to obtain
confidence limits within 25% of the mean were 2 and 7, re-
spectively. For confidence limits within 10% of the mean,
3 tests with 1 replicate were indicated (Table 8-4). As
discussed previously, the use of only one replicate should
be discouraged because it does not allow calculation of
within test variation or a check on test results.
8.5. Pi(2-ethyIhexyl)phthalate
As with ethylene glycol, no flow-through tests were
conducted using di(2-ethyIhexyl)phthalate. The 7 static-
replacement test ECSO's ranged from 408 to 7492 mg/L with
a mean of 2060 mg/L. The ANOVA indicated significant
68
-------
Table 8-8.
Nested ANOVA to Determine Between Tests and
Within Test Variation for O-Cresol. Data
entered were logic ECSO's generated by quadratic
regression with growth rate constant transformation,
Paired Flow-through and Static-Replacement Tests
Variance Sum of Mean
Source df Squares Squares
Total 31 0.65059 0.02099
Test No. 7 0.54731 0.07819
Error 24 0.10328 0.00430
Mean 2.284915
Standard Deviation 0.065601
Coefficient of Variation 0.028711
Variance
Component
0.02277
0.01847
0.00430
Percent
100.00
81.10
18.90
Flow-through Tests
Variance Sum of Mean
Source df Squares Squares
Total 15 0.05318 0.00355
Test No. 3 0.00431 0.00144
Error 12 0.04887 0.00407
Mean 2.179352
Standard Deviation 0.063816
Coefficient of Variation 0.029282
Variance
Component
0.00407
0.00659
0.00407
Percent
100.00
0.0
100.00
Static-Replacement Tests
Variance Sum of Mean
Source df Squares Squares
Total 15 0.24082 0.01605
Test No. 3 0.18640 0.06213
Error 12 0.05442 0.00453
Mean 2.390479
Standard Deviation 0.067339
Coefficient of Variation 0.028170
Variance
Component
0.01893
0.01440
0.00453
Percent
100.00
76.05
23.95
69
-------
Table 8-9.
Nested ANOVA to Determine Between Tests and
Within Test Variation for Ethylene Glycol
Static-Replacement Tests. Data entered were
logio ECSO's generated by quadratic regression
with growth rate constant transformation.
Variance
Source
Total
Test No.
Error
df
19
4
15
Sum of
Squares
0.37785
0.23572
0.14213
Mean 4 .
Standard Deviation 0.
Coefficient of Variation 0.
Mean
Squares
0.01989
0.05893
0.00948
234461
097343
022988
Variance
Component
0.02184
0.01236
0.00948
Percent
100
56
43
.00
.61
.39
Table 8-10.
Nested ANOVA to Determine Between Tests and
Within Test Variation for Di(2-ethylhexyl)
phthalate Static-Replacement Tests. Data
entered were logio EC50's generated by qua-
dratic regression with growth rate constant
transformation.
Variance
Source
Total
Test No.
Error
Mean
Standard
df
24
6
18
Deviation
Sum of
Squares
9.03861
6.06532
2.97329
3.
0.
Mean
Squares
0.37661
1.01089
0.16518
313817
406427
Variance
Component
0.40363
0.23845
0.16518
Percent
100.00
59.08
40.92
Coefficient of Variation
0.122646
70
-------
differences between tests (P<0.002). The 95% confidence
limits around the mean EC50 were 700 and 6000 mg/L, which
are within 14% of the mean in log units. The variance com-
ponents indicated 59% of the variation occurred between
tests with 41% within tests (Table 8-10). The predictions,
based on this variance breakout and Equations 6-11 and
6-12, indicate 4 tests with 6 replicates would be needed
to keep confidence limits within 25% of the mean. To ob-
tain confidence limits within 10% of the mean would not
be cost-effective because 11 tests with 35 replicates each
would be needed (Table 8-4).
71
-------
72
-------
9.0 DRY WEIGHT DATA ANALYSIS
Decrease in total weight per test chamber (replicate)
was significant in almost all tests, even at the lowest
concentration. The difference between control weights and
the lowest concentration was so great that ECSO's could
not be calculated because determination of the LC50 was
given priority in establishing concentration ranges; lower
concentrations in flow-through tests would have made cal-
culation of the LCSO's impossible. If the concentration
range were adjusted for dry weight, the range would not
bracket the LC50. It was possible to extend the range of
concentrations in some of the static-replacement tests.
However, concentrations could be lowered only a limited
amount and still be in the log series necessary for cal-
culation of ECSO's. An additional complication encoun-
tered in data analysis was that o-cresol and ethylene
glycol were stimulatory at lower concentrations, causing
increases in dry weight.
9.1.
Dry weight ECSO's could not be calculated for CuSOi*
using either of the two methods discussed (Section 6.3).
CuSOit was relatively toxic and quickly suppressed growth.
The weight in the controls averaged 3X the weight in the
lowest concentration tested (0.39 mg/L). An EC10 based
on a 90% reduction in the control weight could be calcu-
lated for three of the static-replacement tests. The
EClO's were 3.75, 2.78, and 2.17 mg/L for tests 49, 57,
and 81, respectively, based on arc sin square root trans-
formation. The lack of data made statistical comparison
of static-replacement and flow-through tests impossible.
One set of static-replacement and flow-through tests
was conducted with priority given to the concentrations
for EC50 calculation based on dry weight. The concentra-
tions used were 0, 0.112, 0.20, 0.36, 0.64, 1.12, and 2.0
mg/L for flow-through tests. Static-replacement tests had
two additional concentrations (0.036 and 0.064 mg/L) to
extend the range and hopefully to better bracket the EC50.
The test results were poor. Data points for the
static-replacement test were scattered. The coefficient
of determination (R2) for the static-replacement test was
0.26 with a probability of a greater F at 0.013. The
73
-------
calculated EC50 of 0.74 mg/L CuSOif5H2O should be viewed
with care due to the low R2 value. The flow-through test
results were also poor, although the coefficient of deter-
mination was higher (0.61) with the probability of a
greater F at 0.0001. However, roots in the quadratic re-
gressions for both tests were imaginary.
9.2. 2,4,6-TCP
An EC50 could be calculated for only four of the
static-replacement tests using a logic transformation of
dry weight. The ECSO's were 0.01, 1.63, 0.13, and 0.39
mg/L for tests 55, 65, 76, and 92, respectively- No ECSO's
could be calculated for flow-through tests.
EC10's based on arc sin square root transformation of
dry weight in the test concentrations divided by the dry
weight of controls could be calculated for five static-
replacement tests and four flow-through tests. The static-
replacement tests averaged 2.37 mg/L and ranged from
0.24 to 4.43 mg/L TCP. The flow-through tests averaged
2.82 mg/L with a low of 1.62 and a high of 5.64 mg/L TCP-
Of all the above tests, two sets were paired; they yielded
EClO's of 3.24 and 1.62 mg/L (tests 74-75) and 1.88 and
2.32 mg/L TCP (tests 94-95) for static-replacement and
flow-through tests, respectively. Because of the limited
data, no valid conclusions could be made about flow-through
and static-replacement tests. In one case the static-re-
placement test EC50 was higher, while in the other the
flow-through test EC50 was higher. The between tests vari-
ability was so high that resources would probably best be
allocated by conducting several static-replacement tests
rather than a few flow-through tests.
9.3. O-Cresol
Dry weight data for o-cresol tests were better suited
for calculation of EC50's than data for the other toxi-
cants . The concentration ranges for calculation of the
ECSO's were the same as those used for calculation of
LCSO's (Section 7.3). Quadratic regression with arc sin
square root transformation gave a much higher R2 value
than did the log transformation (R2 =0.75 and 0.48, re-
spectively) , and therefore a better estimate of the ECSO's
(Table 9-1). Thus, as the ANOVA's based on this transfor-
mation were considered more appropriate, the following
discussion is restricted to the arc sin square root trans-
formed data.
74
-------
Table 9-1. ECSO's from Quadratic Regression Using logio Dry
Weight Transformation vs. Arc Sin Square Root of
the Ratio Dry Weight to Control Weight Trans-
formation for 0-Cresol Paired Tests.
Test No.
and Type
42
43
51
52
60
61
70
71
F
S
F
S
F
S
F
S
EC50
logio transformation
113.
97.
152.
121.
96.
134.
97.
174.
91*
7"
1"
7"
81*
4"
8"
6"
EC50
arc sin /~ transformation
127.
77.
25.
17.
41.
11.
6.
0.
6"
9"
32
O3
81
7"
93"
171
Jmean of 1 replicate
2mean of 2 replicates
3mean of 3 replicates
''mean of 4 replicates
75
-------
The ECSO's for static-replacement tests averaged 16.98
mg/L and ranged from 0.17 to 78 mg/L. This extremely wide
range caused a large variance and consequently produced
very wide confidence intervals. The 95% confidence inter-
vals were 0.88 and 330 mg/L which were within 105% of the
mean EC50.
Flow-through tests showed similar results. The mean
EC50 was 32.4 mg/L; test ECSO's ranged from 6.93 to 128
mg/L, with 95% confidence limits of 4.7 and 218 mg/L.
These were within 55% of the mean in log units. The ANOVA
indicated differences at the 0.11 probability level (Table
9-2). Thus, the pooled data (Table 9-3) should be used to
predict the number of tests and replicates needed to yield
confidence limits within 25% and 50% of the mean; 15 tests
with 14 replicates each would be required for confidence
limits within 25% of the mean, while 6 tests with 4 repli-
cates would yield confidence limits within 50% of the mean.
The pooled data based on 8 tests with an average of 3.25
replicates per test were within 40% of the mean EC50
(24.0 mg/L).
9.4. Ethylene Glycol
The same concentrations were used for calculations of
ECSO's and EC10's based on dry weight as were used for the
LC50 calculations (Section 7.4). ECSO's could be calcu-
lated for only two of the five static-replacement tests
conducted; the ECSO's for tests 39 and 45 were 9500 and
4400 mg/L, respectively. Ethylene glycol stimulated
growth at the lower concentrations, producing dry weights
above those in the control as well as scattering the data
points. Thus, the correlation between concentration and
dry weight was poor in most cases. Coefficients of deter-
mination ranged from 0.02 to 0.47 for those tests for
which the EC50 could not be calculated.
9.5. Pi(2-ethylhexyl)phthalate
ECSO's and EC10's could not be generated for di-
(2-ethylhexyl)phthalate. The toxicant was not soluble in
or miscible with water; therefore, it formed oil droplets
or globules, or completely covered the surface of the test
chambers, depending on the concentration. At higher con-
centrations di(2-ethylhexyl)phthalate formed a thin layer
completely across the beaker and coated the duckweed
fronds. The oily coating did not evaporate under the dry-
ing conditions used (Section 5.4.2) and therefore increased
dry weights. Consequently, it was not possible to make any
correlations between dry weight and the effect of di-
(2-ethylhexyl)phthalate on growth.
76
-------
Table 9-2.
ANOVA Table Comparing 0-Cresol Flow-through and
Static-Replacement Paired Tests Conducted During
the Same Time Period. Data entered were logic
ECSO's generated by quadratic regression with
arc sin square root of the ratio dry weight to
control weight transformation (control weight =
mean dry weight at concentration 0).
Dependent Variable: logic EC50
Source
df
Source
df
Sum of Squares
Mean Square
Model
Error
Corrected Total
7
18
25
9.62867483
3.22652895
12.85520378
1.37552498
0.17925161
Type I SS
F Value
PR > F
Type
Date
Type-Date
0.49060331
7.85501384
1.28305768
2.74
14.61
2.39
Tests of Hypotheses Using the Type IV MS for Type-Date as an Error Term
Source
df
Type IV SS
F Value
0.1154
0.0001
0.1029
PR > F
Type
Date
2.10953006
8.87448328
4.93
6.92
0.1129
0.0733
77
-------
Table 9-3. Nested ANOVA to Determine Between Tests and
Within Test Variation for O-Cresol Tests. Data
entered were logi0 ECSO's generated by quadratic
regression with arc sin square root of the ratio
dry weight to control weight transformation (control
weight = mean dry weight at concentration 0).
Paired Flow-through and Static-Replacement Tests
Variance Sum of Mean
Source df Squares Squares
Total 25 12.85520 0.51421
Test No. 7 9.37552 1.37552
Error 18 3.22653 0.17925
Mean 1.378473
Standard Deviation 0.423381
Coefficient of Variation 0.307138
Variance
Component
0.55334
0.37409
0.17925
Percent
100.00
67.61
32.39
Flow-through Tests
Variance Sum of Mean
Source df Squares Squares
Total 13 4.56571 0.35121
Test No. 3 3.28583 1.09528
Error 10 1.27988 0.12799
Mean 1.505649
Standard Deviation 0.357755
Coefficient of Variation 0.237608
Variance
Component
0.41011
0.28213
0.12799
Percent
100.00
68.79
31.21
Static-Replacement Tests
Variance Sum of Mean
Source df Squares Squares
Total 11 7.79889 0.70899
Test No. 3 5.85224 1.95075
Error 8 1.94665 0.24333
Mean 1.230101
Standard Deviation 0.493286
Coefficient of Variation 0.401012
Variance
Component
0.84595
0.60262
0.24333
Percent
100.00
71.24
28.76
78
-------
10.0 ROOT LENGTH DATA ANALYSIS
The root length data were not useful in calculating
ECSO's because of the variability within and between tests.
Normalization of the data was attempted using a logic
transformation of root length and concentration. All co-
efficient of determination (R2) values for linear regres-
sion were less than 0.5, with 14 of 20 less than 0.1. The
quadratic R2 values were higher, with a mean R2 of 0.77.
Only six were less than 0.7. However, in most cases
ECSO's were difficult to determine because of the shapes
of the curves.
One factor influencing the variability of the data
was that when the frond died, the roots fell off or were
broken off in handling. Detached roots were too fragile
to be measured. Thus the root length data set was not
complete. In addition, the number of roots measured in
each chamber varied widely. For example, an o-cresol
test (#43) had an average.of 150 roots/chamber with a
mean length of 21 mm at the lowest concentration, an aver-
age of 5.3 roots/chamber with a mean length of 6.7 mm at
the median concentration, and an average of 6 roots/cham-
ber with a mean length of 12 mm at the highest concentra-
tion. The control chambers averaged 113 roots/chamber
with a mean root length of 21 mm. The curve for this test
was a positive parabola. If taken literally the data
would indicate that the toxicant had no effect at the
lowest concentration, maximum effect near the median con-
centration, and a lesser effect as the concentration in-
creased above the median.
The other data yielded similar results: 2,4,6-TCP,
ethylene glycol, and di(2-ethylhexyl)phthalate tests pro-
duced both negative and positive parabolas and in one
case an almost horizontal line. CuSO^ was the only toxi-
cant to produce a reasonable curve, with decreasing root
length as concentration increased. The EC50 for these
CuSOij tests based on 50% reduction of the mean control
root length averaged 1.7 mg/L. This was similar to the
ECSO's for CuSOi, based on reproduction.
Given the problems with measuring root length and
analyzing the data, and the similarity in the ECSO's for
CuSOi* based on reproduction and root length, reproduction
rather than root length should be used as a means of
assessing toxicant effects.
79
-------
80
-------
11.0 COST ANALYSIS
The relative costs for conducting a normal series of
tests (7 concentrations x 4 replicates) were compared for
flow-through and static-replacement tests. Costs were
broken down into labor (estimated in man-hours) and non-
labor (estimated in 1980 dollars). The cost analysis
assumed that an experimenter conducting bioassays would
already have typical laboratory equipment such as analyt-
ical balances, miscellaneous volumetric and standard
glassware, magnetic stirrers, etc. The basic cost anal-
ysis was based on conducting a single test of 14-day
duration as outlined in Section 5.0.
11.1. Culture Costs
Costs of maintaining stock cultures are the same re-
gardless of the type of test conducted, because the same
number of organisms is required for both flow-through and
static-replacement tests. Initial cost of establishing
the cultures and maintaining suitable conditions for their
sustenance depends on the facilities available. Cultures
should be maintained under conditions identical to test
conditions. Therefore, it is assumed that space and tem-
perature requirements for cultures are covered under costs
for space and temperature control for the actual tests.
Normally three aquaria, approximately 16x25 cm (6 L),
are sufficient to provide enough organisms for conducting
a standard 7-concentration x 4-replicate test. Lighting
for the aquaria costs approximately $100. Organisms for
the tests can be obtained free. Cultures require approx-
imately 21 days growth to provide enough organisms for the
tests. Establishment and maintenance of the cultures,
from the time they arrive until the time they are used in
the tests, requires approximately 26 man-hours and $165 in
non-labor items such as lights, aquaria, and chemicals.
11.2. Static-Replacement Test Costs
The requirements for static-replacement tests are
nominal except for the costs of providing a temperature-
controlled growth chamber (25°+3°). The necessary items
81
-------
and approximate costs for conducting these tests are listed
in Table 11-1. The cost of space for the tests depends on
the facilities available. The actual space requirement is
rather small (20 sq.ft.), assuming additional space is
available for mixing chemicals, making assessments, etc.
Cost of the space required was not determined because of
the variables involved.
When the size of the static-replacement test is al-
tered, the only real change in costs is in labor. Addi-
tional replicates or concentrations do not require signi-
ficant increases in chemicals or other supplies. However,
the assessment time changes considerably. For example, to
conduct a 10-concentration x 10-replicate test, non-labor
increased from $523 to $634 (21%) and labor increased from
53 man-hours to 100 man-hours (89%).
11.3. Flow-through Test Costs
Costs of conducting flow-through tests are consider-
ably greater than for conducting static-replacement tests.
The major cost increase is due to the amount of chemicals
used in Hillman's M-medium, the labor necessary to mix
these chemicals, increased costs of disposable items used
with the dilutor panel, and the capital costs of equipment
(Table 11-2). Additionally, approximately 192 sq.ft. are
required. Part of this space could be used for static-re-
placement tests conducted concurrently with flow-through
tests. The major capital equipment expense is the dilutor
panel. Obviously this initial investment is lessened per
test as more tests are conducted, but for one or few tests
the cost may be prohibitive. The major increase in labor
for a flow-through test is due to the time involved in
mixing medium for test organisms and setup and breakdown
of the dilutor panel.
11.4. Cost Comparisons
The cost differential between static-replacement and
flow-through tests is due to additional equipment, nutrient
solution, space, and labor required to clean, set up, and
break down the flow-through tests. The costs of the test
organisms and assessment are the same for both types of
tests. Capital costs of flow-through tests are 11.5X the
static-replacement costs when the entire cost of the dilu-
tor system is considered. The cost differential decreases
to 3.6X if the panel is not considered. Flow-through tests
require approximately twice as much labor as static-re-
placement tests in order to set up, calibrate, and break
82
-------
Table 11-1. Approximate Costs for Conducting a Single
Static-Replacement Test.
7 x 41 10 x 102
Item Quantity Cost ($) Quantity Cost ($)
NON-LABOR
Beakers (250ml) 28 15 100 56
Chemicals3 - 243 - 243
Lights 2 50 4 100
Misc. - 50 70
Culture" - 165 - 165
Total 523 634
LABOR (Man-hours)
Test setup 4.5 8.5
Conduct/assess 9.5 28.25
Test breakdown 4.75 11.75
Measurements
Root lengths 5.6 20.0
Dry weight 2.5 6.0
Culture4 26.0 26.0
Total 52.85 100.5
1 7 concentrations with 4 replicates each.
2 10 concentrations with 10 replicates each.
3 Costs of minimum-size containers for reagents in
Hillman's M-medium. Actual pro-rated cost of
chemicals per test was $2.85.
** Establishment and maintenance.
83
-------
Table 11-2.
Approximate Costs for Conducting a Single
Flow-through Test, 7 Concentrations with
4 Replicates Each.
Item
Quantity
Cost ($)
NON-LABOR
Dilutor panel
Aquaria
Lights
Troughs
Toxicant and diluent
chambers
Chemicals1
Misc. glassware
Tygon tubing
Cleaning chemicals
Deionized water
Culture2
Total
1
30
8
2
3 (min.)
320 (ft.)
5000
150
200
300
600
379
250
145
50
60
165
7299
LABOR (Man-hours)
Test setup
Conduct/assess
Test breakdown
Measurements
Root lengths
Dry weight
Culture2
21.0
27.25
11.5
5.6
2.5
26.0
Total
93.85
1 Costs of minimum-size containers for reagents in Hillman's
M-medium. Actual pro-rated cost of chemicals per test
was $155.
2 Establishment and maintenance.
84
-------
down the dilutor panel. Additional time is needed to mix
the large volume of nutrient solution required by the flow-
through tests.
The capital costs of the static-replacement test can-
not easily be decreased; however, the labor costs could
easily be reduced 21% by eliminating root length and dry
weight measurements and by making assessments only on the
final day of the tests. Detailed assessment on Days 1,
5, and 6 are not necessary, and if labor is a limiting
factor, eliminating assessment on these days will save
3 man-hours per test.
Assuming a labor rate of $10/hr., the total cost of
conducting a static-replacement test without making assess-
ments on Days 1, 5, and 6, and not measuring dry weight or
root length, would be approximately $930. A comparable
flow-through test would cost approximately $8000. The
costs of space and a controlled environment were not con-
sidered in either estimate.
The results of tests conducted on 2,4,6-TCP and
o-cresol indicated no statistical difference between
static-replacement and flow-through mortality tests.
Although differences were detected for CuSOi* tests at
the 0.03 probability level, the additional costs of con-
ducting flow-through tests would not be warranted and
would be better allocated to conduct additional static-
replacement tests.
85
-------
86
-------
12.0 COMPARISON OF MORTALITY, REPRODUCTION,
DRY WEIGHT, AND ROOT LENGTH DATA
Mortality, reproduction, dry weight, and root length
all were evaluated as parameters for determining the 50%
effect concentrations for the toxicants used in this study
(Section 5.2). Unfortunately, tailored tests were not
conducted to evaluate the effect of the toxicants on each
parameter. Mortality was designated the most important
parameter; therefore, concentration ranges were estab-
lished to obtain the most accurate projections of the
LC50. The additional parameters were measured on all
test organisms, but in some cases the data could not be
used to calculate ECSO's. In most cases the lowest con-
centration used in the mortality testing was at or above
the upper range of concentrations needed to calculate
ECSO's based on dry weight, reproduction, and root length.
The root length data were the least useful. As pre-
viously discussed, the data could not be used to calculate
an EC50 (Section 10.0). Data sets for each test were in-
complete because duckweed roots fell off at the higher
toxicant concentrations.
To determine the relative value of the other three
parameters in calculating LCSO's and ECSO's, partial cor-
relation coefficients were calculated (Section 6.5).
Analyses conducted both by test and by toxicant yielded
similar results. The data for tests pooled by toxicant
are presented as an example of the results (Table 12-1).
The partial correlation coefficients indicate that dry
weight and reproduction (represented by the growth rate
constant K) contained similar information about the effects
of the toxicants on duckweed. The coefficients for
ryiyt-y3 wnich measure the relationship of the log concen-
tration and log dry weight without the effect of repro-
duction (K), indicate that log dry weight contains little
information beyond that contained in K (Table 12-1). Sim-
ilarly, the coefficients for ryay^.y2 indicated little
additional information in dry weight beyond that contained
in the mortality data. These two sets of coefficients
indicate that measuring dry weight in addition to mortal-
ity and reproduction (growth rates) does not contribute
significant information about the effects of the toxicants
on duckweed.
Considering the additional expense in labor and equip-
ment necessary to measure dry weight (Section 11.0), it
seems advisable to delete it from the data collection.
87
-------
00
CO
Table 12-1. Partial Correlation Coefficients to Determine the Correlation Between
Concentration, Dry Weight, Reproduction, and Mortality.
Toxicant
CuSOi*
2,4,6-TCP
o-cresol
ethylene glycol
di (2-ethylhexyl)
phthalate
Partial
r m r
0.
0.
0.
0.
0.
40
58
33
29
38
-0.
-0.
-0.
-0.
-0.
46
53
84
64
27
Correlation
0.
0.
0.
0.
0.
80
59
40
70
57
Coefficients1 ' 2
r r
y i y ** * y 2 y i y s * y 4
-0
-0
-0
-0
0
.26
.36
.72
.31
.066
-0.80
-0.63
-0.66
-0.80
-0.55
ryiy
-0.
-0.
-0.
0.
-0.
4- * V 3
043
53
241*
095
25
1yi = log i o (cone)
Y2 - arc sin / p
y3 = growth rate constant K
y^ = logio(dry weight)
2significance probability = 0.0001 except where indicated otherwise
3significance probability = 0.41
"*significance probability = 0.0011
Significance probability = 0.26
6
'significance probability = 0.29
-------
Except for data analysis, the collection of reproduction
data does not require any additional labor or non-labor
costs beyond that for mortality data. Partial correlation
coefficients indicate additional information beyond that
in mortality is contained in the reproduction data (K).
89
-------
90
-------
13.0 LITERATURE CITED
Berkson, J. 1944. Application of the logistic function
to bio-assay. J. Am. Statist. Assoc. 39:357-365.
Berkson, J. 1949. Minimum x2 and minimum likelihood
solution in terms of a linear transformation, with
particular reference to bio-assay. J. Am. Statist.
Assoc. 44:273-278.
Blackman, G.E. and R.C. Robertson-Cuningham. 1953.
Interactions in the physiological effects of growth
substances on plant development. J. Exp. Bot. 5:184-203.
Dixon, W.J. (ed.) 1974. BMD biomedical computer programs.
Univ. of Calif. Press, Berkeley. 773 pp.
Finney, D.J. 1964. Statistical methods in biological
assay. 2nd ed. Charles Griffin and Co. Ltd., London.
Finney, D.J. 1977- Probit analysis. 3rd ed. Cambridge
Univ. Press, Cambridge, MA. 333 pp.
Finney, D.J. 1978. Statistical method in biological
assay. 3rd ed. MacMillan Publ. Co., Inc., New York.
508 pp.
Hillman, W.S. 1961a. The Lemnaceae, or Duckweeds - a
review of the descriptive and experimental literature.
The Botanical Rev. 27(2) :221-287.
Hillman, W.S. 1961b. Experimental control of flowering
in Lemna. III. A relationship between medium composition
and the opposite photoperiodic responses of L. perpusilla
6746 and L. gibba G3. Am. J. Botany 48:413-419.
Hutchinson, G.E. 1975. A treatise on limnology. Volume III,
Limnological botany. John Wiley & Sons, New York. 660 pp.
Peltier, W. 1978. Methods for measuring the acute toxicity
of effluents to aquatic organisms. U.S. Environmental Pro-
tection Agency, Cincinnati, OH. EPA-600/4-78-012. 51 pp.
Shuster, J.J. and F.H. Dietrich. 1976. Quantal response
assays by inversion regression. Communications in Statis-
tics A. 5(4)293-305.
91
-------
Snedecor, G.W. and W.G. Cochran. 1967. Statistical methods,
6th ed. Iowa State Univ. Press, Ames, 10. 593 pp.
Statistical Analysis System (SAS). 1979. SAS user's
guide. SAS Institute Inc., Raleigh, NC. 494 pp.
Walbridge, C.T. 1977. A flow-through testing procedure
with duckweed (Lemna minor L.). U.S. Environmental Protec-
tion Agency, Environmental Research Laboratory, Duluth, MN.
EPA-600/3-77-108. 20 pp.
92
-------
TECHNICAL REPORT DATA
(Please read Instructions on she reverse before completing)
1. REPORT NO.
EPA 560/6-81-003
4. TITLE AND SUBTITLE
Comparison of Static-Replacement and
Flow-through Bioassays Using Duckweed,
Lemna gibba G-3
7. AUTHOR(S)
John A. Davis, Ph.D.
3. RECIPIENT'S ACCESSION NO.
5. REPORT DATE
January 1981
6. PERFORMING ORGANIZATION CODE
8. PERFORMING ORGANIZATION REPORT NO
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Breedlove Associates, Inc.
618 Northwest 13th Avenue
Gainesville, Florida 32601
10. PROGRAM ELEMENT NO.
B2BL2S
11. CONTRACT/GRANT NO.
68-01-5776
12. SPONSORING AGENCY NAME AND ADDRESS
U.S. Environmental Protection Agency
Office of Pesticides and Toxic Substances
Washington, D.C. 20460
13. TYPE OF REPORT AND PERIOD COVERED
Final
14. SPONSORING AGENCY CODE
15. SUPPLEMENTARY NOTES
16. ABSTRACT
Static-replacement and flow-through tests were conducted using CuSO • 5HaO
2,4,6-trichlorophenol, and o-cresol to determine if they gave similar
LC50's and ECSO's for duckweed, Lemna gibba G-3. Static-replacement
tests also were conducted using ethylene glycol and di(2-ethylhexyl)
phthalate. Mortality, reproduction, dry weight, and root length were
used to measure effect levels of the toxicants. LCSO's and ECSO's were
calculated using quadratic regression with log transformation of the
independent variable (concentration) and with several different trans-
formations for the dependent variables. ANOVA's were used to test for
differences between the two types of tests, tests within types, and
replicates within tests. A procedure also was provided for estimating
the number of tests and replicates necessary to obtain confidence limits
within a given percentage of the mean.
Mortality and reproduction produced the best results. The highest vari-
ation occurred among tests, regardless of type, and the smallest varia-
tion was generally within tests (i.e. among replicates). Therefore, the
best allocation of resources would be to replicate static-replacement
tests in time, using four replicates per test.
17.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b. IDENTIFIERS/OPEN ENDED TERMS
bioassay, duckweed,
Lemna gibba G-3,
copper sulfate, 2,4,6'
trichlorophenol,
o-cresol, ethylene
glycol, di (2-ethyl-
hexyl) phthalate
c. COSATI Held/Group
8. DISTRIBUTION STATEMENT
Release unlimited
cirr- 2220-i (Rev. -s-77
19 StC'j'ilTY CLASS tThi; Htipvrt'
\ unclassified
j20'TECUR! i'-f CLASs777,~r/5jI.-(\r
' unclassified
21 . NO. OF PAGES
105
22. PRICE
"1
our L :, i T i c »
' £ '- b SC -- E "
93
------- |