BEFORE THE ILLINOIS POLLUTION CONTROL BOARD

BEFORE THE ILLINOIS POLLUTION CONTROL BOARD
NOTICE OF FILING
BEFORE THE ILLINOIS POLLUTION CONTROL BOARD
POST HEARING COMMENTS OF PRAIRIE RIVERS NETWORK, SIERRA CLUB AND
THE ENVIRONMENTAL LAW AND POLICY CENTER
BEFORE THE ILLINOIS POLLUTION CONTROL BOARD
CERTIFICATE OF SERVICE
SERVICE LIST- R07-009

BEFORE THE ILLINOIS POLLUTION CONTROL BOARD

IN THE MATTER OF:

)

PROPOSED AMENDMENTS TO:

)

R07-009

35 Ill. Adm. Code 302.102(b)(6), 302.102(b)(8)

)

Rulemaking – Water

302.102(b)(10), 302.208(g), 309.103(c)(3),

)

405.109(b)(2)(A), 405.109(b)(2)(B), 406.100((d)

)

REPEALED 35 Ill. Adm. Code 406.203 Part 407, and

)

PROPOSED NEW 35 Ill. Adm. Code 302.208(h)

)

NOTICE OF FILING

TO: See Attached Service List

PLEASE TAKE NOTICE that the Environmental Law and Policy Center of the

Midwest (“ELPC”), Prairie Rivers Network and the Sierra Club today have electronically

filed POST HEARING COMMENTS OF PRAIRIE RIVERS NETWORK, SIERRA

CLUB AND THE ENVIRONMENTAL LAW AND POLICY CENTER.

Respectfully submitted,

____________________________

Albert F. Ettinger (Reg. No.

3125045)

Counsel for Environmental Law &

Policy Center, Prairie Rivers

Network and Sierra Club

DATED: June 7, 2007

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *

BEFORE THE ILLINOIS POLLUTION CONTROL BOARD

IN THE MATTER OF:

)

PROPOSED AMENDMENTS TO:

)

R07-009

35 Ill. Adm. Code 302.102(b)(6), 302.102(b)(8)

)

Rulemaking – Water

302.102(b)(10), 302.208(g), 309.103(c)(3),

)

405.109(b)(2)(A), 405.109(b)(2)(B), 406.100((d)

)

REPEALED 35 Ill. Adm. Code 406.203 Part 407, and

)

PROPOSED NEW 35 Ill. Adm. Code 302.208(h)

)

POST HEARING COMMENTS OF PRAIRIE RIVERS NETWORK, SIERRA CLUB AND

THE ENVIRONMENTAL LAW AND POLICY CENTER

Prairie Rivers Network (“PRN”), Sierra Club and the Environmental Law and Policy Center (“ELPC”)

continue to support the proposed changes to the Illinois water quality standards for sulfate and total

dissolved solids. As was discussed during the hearing, these standards were developed following extensive

new sulfate toxicity testing and numerous interest group discussions hosted by the United States

Environmental Protection Agency (USEPA) in which members of the Illinois Coal Association and other

organizations fully participated.

We remain concerned about a few elements of the proposal, particularly regarding the proposed changes to

the mixing rules that were not addressed during the USEPA-hosted discussions.

Our specific comments are as follows:

Mixing

PRN, Sierra Club and ELPC, through the pre-filed testimony of Glynnis Collins, proposed language for

302.102(b)(8) that would codify IEPA’s current practice with regards to allowed mixing in the situation in

which 3:1 dilution is not available but in which the receiving water has a higher flow than a zero 7Q1.1.

Somewhat to our surprise, IEPA has not accepted our proposal to write its current practice into the Board

rules suggesting that it might in the future want to deviate from its current practice of assuring a zone of

passage for aquatic life by insisting on at least 50% of the volume of flow.

IEPA also claims that the proposal made by PRN, Sierra Club and ELPC is arbitrary and has no scientific

basis. IEPA, however, admits that the current Board rule covering the situation where 3:1 dilution is

available was based on just the kind of consistent agency practice that is the basis for our proposed rule for

the situation where less than 3:1 dilution is available. (Transcript of Proceedings, April 23, 2007 p. 56-61)

Clearly, there is something wrong here. Either it was wrong for the Board to adopt its current rule regarding

the situations where more than 3:1 dilution is available or there is something wrong with IEPA’s thinking

regarding the PRN/Sierra Club/ELPC proposal as to how to treat the situation where less than 3:1 dilution

is available.

In fact, drawing reasonable lines based on past practice and experience is both proper and a very common

regulatory function. Administrative agency regulatory numerical standards are lawfully established if they

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *

are “within a zone of reasonableness.”

Hercules Inc. v. EPA

, 598 F.2d 91, 107-08 (D.C. Cir. 1978).

See

also

Reynolds Metal Co. v. United States EPA

, 760 F.2d 549, 558 (4th Cir. 1985) (upholding EPA

numerical standard). Indeed, almost all water quality standards are ultimately based on rules of thumb and

views of an appropriate safety factor. For example, there is no scientific proof that requires setting acute

toxicity standards using 50% of the LC 50 instead of 10% of the LC 10. The decision to do so was based on

drawing a reasonable line.

As to the problem now before the Board of setting limits on mixing necessary to provide the zone of

passage necessary to protect aquatic life, PRN/Sierra Club/ELPC, writing on a blank slate, might prefer

using the Board provision (25% of the volume of flow) for all cases including the many cases in which less

than 3:1 dilution is available. However, IEPA’s current use of 50% seems tolerable. While allowing a

substantial number of areas in many streams to fail to meet water quality standards, the current IEPA

practice reserves a substantial portion of the stream for passage of aquatic life.

However, it is not acceptable to allow IEPA to go on using an unpublished rule that is probably not

enforceable because it has not gone through the Board approval process or the procedures set forth by the

Administrative Procedure Act.

See Senn Park Nursing Center v. Miller

, 104 Ill.2d 169, 181,

470 N.E. 2d

1029

(1984). (agency could not use rule that had not been adopted pursuant to the Illinois Administrative

Procedure Act) Nor is it acceptable for IEPA to have rules for required mixing from which it is free to

deviate on an ad hoc basis because the discharger is providing a “vital function for society.” (Transcript

April 23 p. 56) As the Board is aware, there are other ways under the law, including site specific relief and

variances, to reconcile the vital functions of society with the need to protect water quality that do not give

IEPA permit writers unbridled discretion to compromise aquatic health based on their beliefs as to the vital

needs of society.

Eliminating the requirement for allowing a zone of passage where there is little dilution certainly is not

justifiable. The many small and medium sized streams that could be affected by loosening mixing

protections can be extremely important to the overall health of the environment.

See

“Hydrological

Connectivity of Headwaters Streams and Their Contributions to the Integrity of Downstream Waters,”

JAWRA, Vol. 43 pp1-280 (Feb. 2007),

available at

http://www.blackwell-synergy.com/toc/jawr/43/1

PRN/Sierra Club/ELPC continue to believe that the language of Section 302.102(8) should be changed to

state:

(8) The area and volume in which mixing occurs, alone or in combination

with other areas and volumes of mixing must not contain more than 25% of

the cross-sectional area or volume of flow of a stream except for those

streams where the dilution ratio is less than 3:1. In streams where the

dilution ratio is less than 3:1, other than streams that have a zero flow for at

least seven consecutive days recurring on average in nine years out of ten,

the volume in which mixing occurs, alone or in combination with other

volumes of mixing must not contain more that 50% of the volume of flow.

Interactions between sulfate toxicity and other dissolved solids

As explained in the pre-filed testimony of Ms. Collins, some

data suggest that when calcium is the primary

cation in a solution, it may serve to increase the toxicity of sulfate. Table 2 of the attached study, D.R.,

D.D. Gulley, J. R. Hockett, T. D. Garrison, and J.E.Evans. 1997. Statistical Models to Predict the Toxicity

of Major Ions to

Ceriodaphnia dubia

Daphnia magna

, and

Pimephales promelas

(fathead minnows).

Environmental Toxicology and Chemistry

16(10):2009-2019, shows that for the three species tested, mean

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *

LC50 values for sulfate when calcium was the predominant cation in solution was lower than mean LC50

values for sulfate when sodium was the predominant cation (1910 mg CaSO

/L of vs 3080 mg NaSO

/L).

When converted to the common term of toxicity per unit SO

, the relationship still holds true, with LC50

values of 1348 mg SO

/L vs. 2082 mg SO

/L for calcium and sodium, respectively. The lower LC50 value

for calcium indicates that the solution was more toxic to test organisms.

However, while noting this problem, we agree with the approach taken by IEPA at the April 23

hearing to

address these issues on a permit-by-permit basis by discouraging the use of Ca(OH)

in processing at mine

sites, and requiring specific monitoring of water chemistry and toxicity in situations where Ca(OH)

will be

used.

Chloride above 500 mg/L

The draft standard does not clearly address the situation in which chloride concentrations in the receiving

waters are greater than 500 mg/L and all of the participants at the hearing agreed that there may

occasionally be situations in which this will be the case. IEPA has stated that it will address the situation on

a case-by-case basis. PRN/Sierra Club and ELPC ask that the Board make clear in the rule that if chloride

exceeds 500 mg/L, IEPA shall develop sulfate limits to prevent any danger that the sulfate discharge will

increase the potential adverse effects on aquatic life caused by the violation of the chloride standard.

Respectfully submitted,

Albert Ettinger

Counsel for Prairie Rivers Network,

Sierra Club and the Environmental Law & Policy Center

DATED: June 7, 2007

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *

BEFORE THE ILLINOIS POLLUTION CONTROL BOARD

IN THE MATTER OF:

)

PROPOSED AMENDMENTS TO:

)

R07-009

35 Ill. Adm. Code 302.102(b)(6), 302.102(b)(8)

)

Rulemaking – Water

302.102(b)(10), 302.208(g), 309.103(c)(3),

)

405.109(b)(2)(A), 405.109(b)(2)(B), 406.100((d)

)

REPEALED 35 Ill. Adm. Code 406.203 Part 407, and

)

PROPOSED NEW 35 Ill. Adm. Code 302.208(h)

)

CERTIFICATE OF SERVICE

I, the undersigned, on oath state that I have served the attached POST HEARING

COMMENTS OF PRAIRIE RIVERS NETWORK, SIERRA CLUB AND THE

ENVIRONMENTAL LAW AND POLICY CENTER upon the persons listed in the

attached service list via U.S. Mail.

Respectfully submitted,

Albert F. Ettinger (Reg. No.

3125045)

Counsel for Environmental Law &

Policy Center, Prairie Rivers

Network and Sierra Club

DATED: June 7, 2007

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *

SERVICE LIST- R07-009

Dorothy Gunn, Clerk

Marie Tipsord, Hearing Officer

Illinois Pollution Control Board

100 W. Randolph St.

Suite 11-500

Mathew Dunn

Jonathan Furr

Illinois Attorney General’s Office

IDNR

Environmental Control Division

One Natural Resources Way

James R. Thompson Center

Springfield, IL 62701-1271

100 West Randolph Street

Chicago, IL 60601

Sanjay K. Sofat, Assistant Counsel

Illinois Environmental Protection Agency

1021 North Grand Avenue East

P.O. Box 19276

Springfield, IL 62794-9276

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *

Environmental Toxicology and Chemistry, Vol.

16,

No.

10,

pp.

2009-2019, 1997

Printed in the USA

0730-7268197 $6.00 + .OO

Produced Water Series

STATISTICAL MODELS TO PREDICT THE TOXICITY OF MAJOR IONS TO

CERIODAPHNIA D UBIA, DAPHNIA MA GNA

AND

PIMEPHALES PROMELAS

(FATHEAD MINNOWS)

DAVID R. MOUNT,*$ DAVID D. GULLEY,~~

J. RUSSELL HOCKETT,II TYLER D. GARRISONJI

and JAMES M. EVANS#

1U.S. Environmental Protection Agency, 6201 Congdon Boulevard, Duluth, Minnesota 55804

§University of Wyoming, Laramie, Wyoming 82070, USA

IIENSR Consulting and Engineering, Fort Collins, Colorado 80521, USA

#Gas Research Institute, Chicago, Illinois 6063 1, USA

(Received

April

1996;

Accepted

Februaiy

1997)

Abstract-Toxicity of fresh waters with high total dissolved solids has been shown to be dependent on tbe specific ionic composition

of the water. To provide

a predictive tool to assess toxicity attributable to major ions, we tested the toxicity of over 2,900 ion

solutions using the daphnids,

Ceriodaphnia dubia

and

Daphnia magna,

and fathead minnows

(Pimephales promelas).

Multiple

logistic regression was used to relate ion composition to survival for each of the three test species. In general, relative ion toxicity

was

K+

HCO;

Mg"

CI-

SO:-; Na+ and Ca2+ were not significant variables in the regressions, suggesting that the toxicity

Na+ and CaZ+ salts was primarily attributable to the corresponding anion. For

C. dubia

and

D. magna,

toxicity of C1-, SO:-,

and

K+

was reduced in solutions enriched with more than one cation. Final regression models showed a good quality of fit to the

data

(R2

0.767-0.861). Preliminary applications of these models to field-collected samples indicated a high degree of accuracy

for the

C. dubia

model, while the

D. magna

and fathead minnow models tended to overpredict ion toxicity.

Keywords-Ions

Total dissolved solids

Salinity

Toxicity

Ceriodaphnia dubia

INTRODUCTION

Natural fresh waters contain several ionic constituents at

greater than trace levels. Indeed, ions such as

Na+, CaZ+, C1-,

and others are required at a minimum level to support aquatic

life, and these major ions are components of most formulas

for "reconstituted" water used in aquatic toxicity testing

[1,2].

However, many natural and anthropogenic sources can increase

ion concentrations to levels toxic to aquatic life. Studies of

oil and gas produced waters

[3-51, irrigation drain waters [6,7],

shale oil leachates [8], sediment pore waters [9,10], and in-

dustrial process waters

[11,12] have shown toxicity caused by

elevated concentrations of common ions.

Typically, integrative parameters such as conductivity, total

dissolved solids (TDS), or salinity are used as a measure of

the concentrations of common ions in fresh waters. While for

a given ionic composition there is undoubtedly a correlation

between increasing conductivity or TDS and increasing tox-

icity, these parameters are not robust predictors of toxicity for

a range of water qualities. For example,

Burnham and Peterka

[13] noted that fathead minnows could tolerate TDS concen-

tratlons up to 15,000 mg/L in Saskatchewan lakes dominated

Na+ and SO:-, but populations did not persist above 2,000

mg/L in Na+/K+/HCOy-dominated lakes of Nebraska. In stud-

ies of irrigation drain waters, Dickerson et al. [7]

found

Cer-

iodaphnia dubia

50% lethal concentration (LC50) values cor-

responding to approximate conductivities of 3,500 to 4,000

p,S/cm (calculated), while Jop and Askew [I 11 showed major

ion toxicity to

C. dubia

in an industrial process water with a

To whom correspondeilce may be addressed.

t Deceased.

Presented in part at the 12th Annual Meeting, Society of Envi-

ronmental Toxicology and Chemistry, Seattle, WA, USA, November

3-7, 1991.

conductivity of only 1,800

p,S/cm (K.M. Jop, personal com-

munication). Studies by

Dwyer et al. [I41 demonstrated that

the toxicity of high TDS waters to

Daphnia magna

and striped

bass

Morone saxatilis

was

dependent on the specific ionic

composition of those waters.

Given the substantial differences in toxicity among major

ion salts

[I 51, these differing responses in waters with different

ionic compositions are to be expected. Still, they emphasize

the inadequacy of generic measures for assessing the potential

toxicity of major ions and the need for a broader understanding

of major ion toxicity. This paper presents research to develop

more comprehensive tools for assessing major ion toxicity.

Acute toxicity tests using three freshwater organisms were

conducted on solutions enriched with varying combinations of

major ions. Results of these tests were incorporated into mul-

tivariate logistic regression models that predict survival of the

three test species based on major ion concentrations.

MATERIALS AND METHODS

Test organisms

All organisms used in testing were obtained from in-house

cultures (ENSR, Fort Collins, CO, USA);

daphnids were less

than 24

h old at test initiation; while fathead minnows were

1 to 7 d old.

Ceriodaphnia dubia

were cultured in either mod-

erately hard reconstituted water (MHRW) or 20% mineral wa-

ter

[l] at 25'C, while

D. magna

were cultured in hard recon-

stituted water [I] at 20°C. Fathead

minnow brood stock were

cultured at 20 to

25'C in tap water that was pretreated with

activated carbon. Eggs and larva were held in MHRW, larva

were fed brine shrimp nauplii

(Artemia

sp.) twice daily until

they were used in testing.

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *

2010

Environ. Toxicol. Chem.

16, 1997

D.R. Mount et al.

Test procedures

Toxicity tests followed the general guidance of the U.S.

Environmental Protection Agency

(USEPA) [1,16] for con-

ducting acute whole effluent toxicity tests. All tests were con-

ducted in 30-ml plastic beakers containing 10 ml of test so-

lution and five organisms per chamber. Tests were conducted

under a

16-h:8-h light: dark photoperiod;

C. dubia

and fathead

minnows were tested at

25OC, while D.

magna

were tested at

20°C.

Dilution/control water for all tests was MHRW. Expo-

sure periods were 48 h for

C. dubia

and

D. magna

and 96 h

for fathead minnows, with daily observations of mortality. The

criteria for death were no visible movement and no response

to prodding.

Standard guidance for acute

efftuent toxicity testing [I] is

to withhold food during testing of daphnids, presumably be-

cause of concerns that the addition of food might alter the

toxicity of the sample. However, in water devoid of food

(e.g.,

reconstituted laboratory water), withholding food likely places

some stress on the test organisms. Moreover, effluents and

ambient waters, to which the results of these experiments ap-

ply, can be expected to contain bacteria, algae, and other

sources of food. Hence, addition of daphnid food

(yeastlcer-

ophylltrout chow [YCT] and algae [2]) to clean laboratory

water might better simulate the characteristics of

field-col-

lected samples. To assess the potential effect of feeding on

major ion toxicity, initial tests using

C. dubia

were conducted

both with and without feeding. Analysis of these initial ex-

periments (see Results) showed that the addition of food rep-

resented only a small influence on

dubia

survival. Because

the effect of feeding was small and its inclusion was believed

to provide a more representative test matrix, remaining

dubia

tests included feeding, as did all D.

magna

and fathead

minnow tests. For daphnid tests, 100

p1 of a 1:l mix of YCT

and algal suspension was added to each test chamber at test

initiation. For fathead minnow tests, 100

pl of concentrated

brine shrimp nauplii was added after 48 h of exposure, though

solutions were not subsequently renewed as recommended by

the

USEPA [I].

Because toxicity testing of salt solutions was to be com-

pleted over several months, we recognized the possibility that

systematic drift in test organism sensitivity could bias the re-

sults of toxicity tests conducted at different times. In anattempt

to account for this potential variability, each set of toxicity

tests included a reference toxicant test using

NaC1. LC50 val-

ues were computed for each of these tests and were included

in the statistical modeling as another independent variable.

Thus, if drifts in organism sensitivity did occur and were re-

flected in the response to

NaC1, they could be accounted for

in the regression modeling.

Chemical measurements

Concentrations of major ions were determined analytically

in all stock solutions used in testing. Ca2+,

Na+, Mg2+, and

K+ were determined using inductively coupled plasma emis-

sion spectroscopy (ICP) according to

USEPA method 200.7

[17]; C1- and SO:- concentrations were determined by anion

chromatography

[18]; and HCO; concentrations were deter-

mined indirectly by the measurement of phenolphthalein al-

kalinity

[19]. As HCOy is the predominate carbonate species

present in the pH range of interest (pH

6.5-9.0), alkalinity

equivalents were converted directly

ti HCO; concentration.

Dissolved oxygen (DO) and pH were measured in selected

test solutions during actual toxicity testing, primarily on so-

lutions near the threshold for acute toxicity. DO was measured

with a Yellow Springs Instrument model 54 DO meter (Yellow

Springs, OH, USA) while pH was measured with a Orion pH

meter model

SA250 (Boston, MA, USA). Measured DO con-

centrations were always within an acceptable range

(>40%

saturation) [I]. Measured pH varied according to the com-

ponents of the solution but was generally between pH 7.5 and

9.0.

Preparation of test solutions

Test solutions were prepared by dissolving individual ion

salts in MHRW. Salts used in testing were

NaCl, Na,SO,,

NaHCO,, KCl, K,SO,, KHCO,, CaCl,, CaSO,, MgCI,, MgSO,,

CaCO,, and MgCO,; all were of reagent grade or better (Sigma

Chemical Company, St. Louis, MO, USA). Stock solutions

were prepared from these salts by dissolving 10,000

mg/L of

a salt in

MHRW. CaSO, was not fully soluble at 10,000 mglL;

for this reason, CaSO, solutions were filtered through a 1-pm

glass fiber filter prior to testing and ion concentrations were

measured in filtered solutions. Test solutions using

CaCO, and

MgCO, had pH in excess of 10 and were acidified with HC1

or H2S04 until pH stabilized at approximately 8.5.

For tests evaluating only one salt (one cation and one an-

ion), test solutions were prepared by serially diluting the

10,000-mg/L stock solutions with MHRW to develop a series

of test concentrations spaced on a 0.5

dilution factor (i.e.,

10,000, 5,000, 2,500, 1,250 mg/L). For tests involving two

salts, solutions were prepared by combining equal volumes of

the two stock solutions, then diluting as necessary. As testing

proceeded and effect thresholds were determined, test con-

centrations were often spaced much more closely

(e.g., 2,500,

2,000, 1,500, 1,000, 500

mg/L) to better define responses near

the effect threshold.

All ion concentrations measured in the stock solutions were

compared to nominal values. If the measured concentrations

differed from the nominal value by more than

20%, the actual

measured concentrations were substituted for the nominal con-

centrations. Aside from CaSO,, which did not completely dis-

solve, substantial discrepancies between nominal and mea-

sured concentrations occurred in two instances, once for a

MgCI, stock solution and once for a CaCl, stock solution. In

some analyses, the measured concentrations of cations and

anions (expressed as milliequivalents or meq) in a salt solution

were not similar. Because charge balance is a

physicallchem-

ical requirement, such solutions were further evaluated to de-

termine which concentration (cation or anion) was closer to

the nominal value. In all cases, the cation concentration was

closer to the nominal value; based on this, the anion concen-

tration in the stock solution was changed to the concentration

(in

meq) of the corresponding cation.

To calculate ion concentrations in actual test solutions, the

concentrations in the applicable stock solutions were multi-

plied by the relative proportion of each solution in the test

solution. Because the dilution water (MHRW) also contained

small concentrations of each ion, these background concen-

trations were then added to the calculated contributions from

the stock solutions.

In cases where an

SO$- salt (e.g., Na2S04) was combined

with a Ca2+ salt

(e.g., CaCl,), the potential existed for super-

saturation of test solutions with respect to CaSO,. This po-

tential was confirmed by the appearance of white precipitates

in some test solutions. Because precipitation would affect the

dissolved ion concentrations in the test solutions, all ion

com-

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *

Statistical models to predict ion toxicity

Environ. Toxicol. Chem.

16, 1997 

201 1 

binations tested were checked for CaSO, supersaturation by

comparing the nominal test concentrations of Ca2+ and

SO:-

with the solubility product for CaSO, (226.5) calculated from

measured concentrations of CaZ+ and

SO:- in a saturated

CaSO, solution. If a particular solution was supersaturated with

respect to CaSO,, CaZ+ and

SO:- concentrations were reduced

on an equimolar basis until the concentrations reached the

calculated saturation point. These corrected concentrations

were then used for data analyses.

Replication

To incorporate intertest variability into the data set, em-

phasis was placed on replication between batches of tests con-

ducted through time rather than on having replicate chambers

tested simultaneously. Accordingly, most ion combinations

evaluated were tested on at least two and as many as five

different occasions (see results). The exception was for two

cationlone anion solutions tested with D. magna and fathead

minnows, and two

cationltwo anion solutions tested with C.

dubia; for these tests, duplicate chambers (10 animals total)

were tested simultaneously. When calculating LC50 values,

replicate tests conducted on different days were analyzed sep-

arately, but duplicate chambers tested simultaneously were

combined into one analysis.

Data collection, management, and analysis

Data generated by all toxicity tests were entered into a

database using

Paradox@ 3.1 software (Borland International,

Scotts Valley, CA, USA). Regression modeling was based on

individual ion concentrations rather than salt concentrations.

By converting salts to ion concentrations, we were able to

separate out the effects of individual cations and anions instead

of the effects of cation-anion pairs. Statistical modeling of the

toxicity data consisted of

stepwise logistic multiple regression

using the LR program within BMDP statistical software

[20].

Logistic regression relates binary observations (e.g., alive

or dead) to one or more independent variables (in this case,

ion concentrations). The completed regression predicts

aprob-

ability of survival based on concentrations of ions showing

relationships to survival. The linear logistic regression model

used is of the form

logit(P)

In[Pl(l

P)]

where

proportion surviving,

regression coefficient,

ion concentration, and

n

total number of significant

terms in the model.

During development of the final models, various data trans-

formations

(e.g., log) and independent variable interactions

(e.g., Cl

SO, interaction) were considered. Each potential

model was evaluated using the following criteria: (1) each

independent variable in the model must significantly improve

the fit of the model to the data

(a

0.05); (2) the model

should maximize Rz (maximize the amount of variance in the

data that is explained by the model) and minimize the number

of independent variables; and (3) the model should provide

reasonable predictions even when extrapolating outside the

limits of the data used to generate the model.

Data collection and model development were iterative pro-

cesses in which a series of statistical models (regressions) were

developed followed by supplemental data collection. To begin,

data were generated for single ion pairs or salts

(e.g., NqSO,,

CaCl,). Based on these data, an initial regression equation was

developed (F,). Next, additional toxicity data were generated

using combinations of two cations and one anion

(e.g., Na+,

Caz+, and SO:-) and one cation and two anions (e.g., Na',

C1-, and SO:-). The F, equation was then used to predict

survival for these additional data. In addition, a second re-

gression equation

(F,) was then developed using all data gen-

erated to date. The predictive abilities of both models were

then compared by examining the relationship between pre-

dicted and observed survival for all of the ion combinations

tested. If

F2 had notably better predictive ability than F,, we

concluded that important relationships in the data were not

accounted for in the

F, equation. The process was repeated by

testing more complex ion solutions and developing additional

regression equations, until the incorporation of additional data

did not substantially alter the basic equation. This iterative

process of data generation, model development, and additional

data generation continued throughout model development.

As part of this iterative process, characteristics of specific

points that had poor correlation between predicted and ob-

served survival were considered. In some cases, it was found

that such data points had poor agreement between replicate

tests of the same ion combination, hence it was impossible for

the regression equation to fit both responses. In these instances,

additional toxicity tests were conducted using that particular

combination of ions to better characterize the response. Of

2,904 total data points, 59 were discarded as spurious; of these,

46 were for C. dubia, 5 for D. magna, and

8 for fathead

minnows. Thirty-eight of the 59 discarded points were cases

where mortality (typically one or two dead out of five organ-

isms) was observed two or more concentrations below the

primary concentration response, suggesting that ion toxicity

may not have been the cause of mortality. Though these points

may represent innate variability in the survival of test organ-

isms, our intent was to represent mortality due to ion stress;

random mortalities at low ion concentrations tended to de-

crease the slope of the regression model and obscure the re-

sponse threshold. Of the remaining

discarded,points, 10 were

discarded because the CaSO, solution was not filtered prior to

testing (C. dubia);

I0 were from a K2S04 dilution series in

which there was erratic and substantial mortality without ev-

idence of a concentration response (C. dubia); and one was

from a test chamber that was spilled after the 24-h observation

(P. promelas).

In other cases, it was found that outlier points tended to

share certain characteristics. For example, it was noted that

for

C. dubia, early regressions showed poor predictive ability

for ion combinations containing

C1- opposed by two cations

(e.g., Na+ and Ca2+ with C1-); these solutions showed lower

toxicity than those with just one

C1- salt (e.g., NaCl). Further

testing with these ion combinations showed that this response

was reproducible. To account for this phenomenon, a new

variable called NumCat was created. The value of NumCat is

equal to the number of cations representing at least 10% of

the total molar concentration of cations and present at greater

than 100

mg/L. The development and implications of the

NumCat variable are discussed in detail in the Results.

In addition to the more rigorous statistical modeling de-

scribed above, LC50 concentrations were also calculated using

a computer program following the trimmed

Spearman-Karber

method 1211. Independent LC50 values were calculated for

each unique

(i.e., nonsimultaneous) test of ion toxicity. For

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *

2012

Environ. Toxicol. Chem.

16, 1997

D.R. Mount et al.

Table 1. Number of ion solutions tested for toxicity'

Reference

Number of cations/anionsb

toxicant

and

Species

111

112

211

212

311

411 Subtotal

controls

Total

Ceriodaphniadubia

464

449

438

401

108

1,887

232

2,119

Daphnia magna

354

147

65.

566

122

688

Fathead minnows

242

142

451

499

Replicate analyses counted separately.

Number of ions enriched above background concentrations.

ion combinations that were tested repeatedly, average LC5Os

were calculated as the arithmetic mean of the values. In some

cases, tests did not capture the effect threshold and an LC50

could only be expressed as a range

(e.g., LC50

625 mg/L).

Where this range did not conflict with the other calculated

values, the indefinite value was dropped and the mean was

calculated from the remaining values

(e.g., 500,700, and <625

would average to 600 with

2). If the indefinite value

represented an extreme value, the mean was calculated as an

inequality relative to the mean of the numerical values

(e.g.,

775, 700, and <625 would average to <700 with

3).

RESULTS AND DISCUSSION

In total, survival data were collected for 2,904 ion solutions,

excluding reference toxicant tests and controls (Table 1). Data

collection and modeling were conducted first for

dubia,

and

the resulting data set encompasses both greater replication and

a greater variety of ion combinations. The full data sets are

too extensive to provide here but are provided in print in Mount

and Gulley

[22].

To present the data in a more condensed form, LC50 values

were calculated for all ion solutions tested (Tables 2 and 3).

Coefficients of variation for LC50 values for individual ion

combinations were typical for acute toxicity tests

[I.], with

means of 17% for

dubia

(SD

14; range 0.0-61), 17%

for

magna

(SD

7.5; range 4;8-3 l), and 24% for fathead

minnows (SD

15; range 1.4-62).

The effect of feeding on the response of

dubia

was

assessed during the first three sets of tests conducted. In each

of these, toxicity .of each single salt solution was tested both

with and without the addition of food. Average LC50 values

for tests with and without feeding were similar (Fig.

l), al-

though there was a tendency for tests without feeding to have

slightly lower LC50 values. Logistic regression modeling of

these data confirmed this trend; feeding was judged a signif-

icant variable by the regression algorithm, with a positive co-

efficient indicating that feeding did increase overall survival.

However, the influence of feeding in the model was quite small,

explaining less than 1% of the overall variance. Because we

believed that the addition of food might provide a more natural

test matrix, all remaining tests were conducted with feeding.

To determine whether the results of reference toxicant tests

related to the responses observed in the concurrent exposures

to ion combinations, LC50 values were calculated for the ref-

erence toxicant tests from the first 11 test groups with

dubia

(total of 1,045 ion solutions tested). During this period, 48-h

LC50 values for

NaCl averaged 1,042 mglL as C1- with a

coefficient of variation equal to 24%. The LC50 value from

the concurrent reference toxicant test was included as an in-

dependent variable for each ion solution and thus considered

by the

stepwise logistic regression. In this analysis, the ref-

erence toxicant variable was not selected as being statistically

significant, explaining only 0.12% of the overall variance.

From this, we surmised that there was no consistent relation-

ship between the sensitivity of the test organisms (as

measwed

by the reference toxicant test) and the responses of organisms

in the concurrent ion exposures. For this reason, the reference

toxicant test results were not considered further in subsequent

regressions.

As described previously, the development of the final pre-

dictive models was an iterative process in which a series of

regression models was developed. Initial regressions were de-

veloped based on more limited data sets

(e.g., results from

toxicity tests using single salts only); as data collection pro-

ceeded to more complicated solutions (enrichment with three

and four ions), these equations were refined. Throughout the

project, 74 distinct models were developed and considered.

The majority of these models were discarded, either because

they were superseded by later models that incorporated larger

data sets, or were found to have undesirable characteristics

(e.g., poor predictive ability). Several of these analyses in-

volved experimentation with alternative variables or data trans-

formations. To illustrate the model development process, we

selected three intermediate models that demonstrate major ad-

vances in the model development, including the creation of a

new variable, referred to as NumCat. The three example mod-

els are referred to as the single salt, double salt, and double

salt with

NumCat models and are based on 48-h survival data

for

dubia.

The single salt model was developed relatively early in the

data collection process using 362 data points involving single

salt solutions only

(i.e., enriched with one cation and one

anion; Fig. 2). This regression equation fit the observed sur-

vival values very well, with an RZ value of 0.950. Significant

variables in this equation were the concentrations of K+, Mg2+,

HCO,, C1-, and SO:-; Na+ and Ca2+ were not significant vari-

ables indicating that the toxicity of

Na+

and CaZ+ salts could

be accounted for primarily by the toxicity of the

co-occurring

anion. No first-order interaction terms (e.g., K

CI) were se-

lected as significant.

Data collection was then expanded to include solutions with

one cation and two anions and two cations and one anion.

When the single salt model was used to predict survival for

this expanded data set (1,045 data points) it showed consid-

erably less predictive ability than it had for the smaller initial

data set. Accordingly, a new model was developed using data

from all test solutions. This double salt model had the same

significant variables as did the single salt model but did a

better job of predicting survival for the entire data set than

did the single salt model. Although it did have better predictive

ability for the combined data set, the

R2 value of 0.837 indi-

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *

Table 2. Mean 24-h (upper right) and 48-h (lower left) LC50 values for salt combinations tested with

Ceriodaphniaa 

NaCl

Na,S04

NaHCO,

KC1

K2SO4

KHCO,

CaCl,

CaS04

MgClz

MgSO4

24-h

NaCl

Na,SO,

NaHCO,

KC1

KlSO4

KHCO,

CaCl,

CaSO,

MgClz

MgSO4

NaCl

Na,SO,

NaHCO,

KC1

K,S04

KHCO,

CaCI,

CaSO,

MgClz

MgSO4

48-h

NaCl

NqSO4

NaHCO,

KC1

KzSO4

KHCO,

CaCl,

CaS04

MgClz

MgSO4

Values are arithmetic means

[n]

(range) expressed as total ion concentrations added in mgL. Tests with two salts involved 1: 1 combinations of stock solutions containing 10,000 mgL, except CaSO,

(1,970 mg/L).

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *

20 14

Environ. Toxicol. Chem.

16, 1997

D.R. Mount et al.

Table 3. Mean LC50 values for salt combinations tested with

Daphnia magna

and fathead minnows'

Daphnia magna

Fathead minnow

Salt

24-h

48-h

24-h

48-h

96-h

NaCl

6,380 [2]

4,770 [2]

8,280 [3]

6,510 [3]

6,390 [3]

(6,160-6,600)

(3,790-5,740)

(7,240-1 0,000)

(6,090-7,070)

(6,020-7,070)

Na,S04

6,290 [4]

4,580 [4]

>8,080 [3]

>7,960 [3]

7,960 [3]

(5,790-7,070)

(4,060-5.360)

(7,070-> 10,000)

(6,800-> 10,000)

(6,800-10,000)

NaHCO,

2,380 [4]

1,640 [4]

4,850 [2]

2,500 [2]

(850 [3]

(1,900-2,870)

(1,170-2,030)

(3,540-6,160)

(950-4,060)

(<3 10-1,220)

KC1

740 [5]

660 [5]

950 [3]

910 [3]

880 [3]

(580-880)

(440-880)

(750-1,090)

(750-1,020)

&SO4

850 [4]

720 [4]

990 [4]

860 [4]

680 [4]

(670-1,170)

(580-880)

(770-1,170)

(580-1,170)

(5 10-880)

KHCO,

670 [4]

650 [4]

940 [4]

820 [4]

<510 [4]

(440-880)

(380-820)

(750-1,340)

(750-880)

(<3 10-750)

CaC1,

3,250 [4]

2,770 [4]

>6,660 [3]

>6,560 [3]

4,630 [3]

(2,680-4,010)

(2,330-3,230)

(4,700-> 10,000)

(4,390-> 10,000)

(3,930-5,360)

CaSO,

1,970 [3]

>1,970 [3]

1,970 [2]

1,970-> 1,970)

1,970->1,970)

MgC12

1,560 [4]

1,330 [4]

3,520 [3]

2,840 [3]

2,120 [3]

(1,250-1,810)

(1,170-1,580)

(2,520-4,490)

(1,970-3,880)

(1,580-2,740)

%SO4

2,360 [4]

1,820 [4]

4,630 [3]

3,510 [3]

2,820 [3]

(2,180-2,500)

(1,540-2.330)

(3.1 80-7,070)

(3,000-4,350)

(2,610-3,080)

NaCI/Na,SO,

6,140 [2]

5,700 [2]

>9,040 [2]

>8,460 [2]

6,090 [2]

(5,360-6,930)

(5,360-6,030)

(8,080-> 10,000)

(6,930-> 10,000)

(6,030-6,160)

NaCI/NaHCO,

4,440 [2]

2,950 [2]

4,580 [2]

3,790 [2]

2,540 [2]

(3,520-5,360)

(2,830-3,080)

(3,540-5,630)

(2,330-5,250)

(2,330-2,750)

Na2S04/NaHC0,

4,480 [2]

3,180 [2]

5,350 [2]

5,050 [2]

4,060 [2]

(4,060-4,900)

(2,830-3,540)

(4,660-6,030)

(4,060-6,030)

(3,080-5,040)

KC1/K2S04

740 [2]

900 [2]

760 [2]

(600-880)

(790-1,020)

(630-880)

KCIIKHCO5,

740 [2]

800 [2]

770 [2]

(640-830)

(770-830)

(700-830)

K,SO,/KHCO,

630 [2]

1,060 [2]

720 [2]

(540-720)

(1,030-1,090)

(610-830)

CaCI,/CaSO,

3,250 [2]

2,950 [2]

>5,510 [I]

>5,510 [l]

>5,510 [I]

(3,140-3,360)

(2,760-3,150)

MgC12/MgS04

2,1 10 [2]

1,510 [2]

3,830 [2]

3,330 [2]

2,800 [2]

(1,940-2,280)

(1,340-1,680)

(3,790-3.870)

(3,300-3,370)

(2,240-3.370)

NaCVKCl

3,930 [I]

1,410 [l]

1,410 [I]

NaCIICaCI,

5,250 [I]

5,250 [l]

8,410 [I]

8,080 [I]

6,460 [I]

NaCI/MgCl,

3,820 [I]

3,070 [I]

5,250 [I]

3,520 [l]

3,160 [I]

KCIICaCl,

2,620 [I]

2,450 [I]

2,810 [l]

2,810 [I]

KCI/MgCI,

2,280 [I]

2,020 [I]

1,580 [l]

1,410 [I]

CaCI,/MgCl,

4,850 [l]

4,390 [I]

5,630 [l]

5,250 [I]

Na2S0,/K2S0,

4,800 [I]

4,610 [l]

1,580 [l]

1,580 [I]

Na,SO,/MgSO,

8,400 [I]

7,980 [I]

8,840 [I]

5,740 [l]

4,800 [I]

K,SO,ICaSO,

1,160 [I]

1,200 [I]

1,980 [l]

1,720 [l]

K,SO,/MgSO,

2,760 [I]

2,210 [l]

1,380 [l]

1,290 [l]

1,290 [I]

CaSO,/MgSO,

>6,470 [I]

NTb

NaHCO,KHCO,

1,220 [I]

1,040 [I]

1,140 [I]

820 [I]

740 [I]

"Values are ar~thrnetlc means [n] (range) expressed as total ion concentrations added In mgL. Tests with two salts involved 1:l combinations

of stock solutions containing 10,000 rngIL, except for CaSO, (1,970 mgL), MgCI, (5,480 mgL), and CaCl, (7,480 mg1L).

Not tested.

cated a lower quality of fit than was observed for the single

salt model fit to the initial, less complex data set.

There were two basic explanations for the decreased quality

of fit observed with the double salt model: (1) the larger data

set contained greater inherent variability (measurement error)

and hence it was not possible to achieve as high an

R2 value;

(2) there were important toxic interactions represented in

the three ion solutions that were not represented in the solu-

tions containing only a single salt (although the regression

algorithm had not selected

any

interaction terms as being sig-

nificant). When the ion combinations for which the model

made poor predictions were analyzed, some patterns were ap-

parent. In particular, it appeared that the model was

overpre-

dicting toxicity for solutions containing two

C1-

salts.

This phenomenon is perhaps best illustrated by data col-

lected for solutions of NaCl and

CaCl, tested both alone and

in combination. As explained above, the single salt model

indicated that the toxicity of

Na+ and Ca2+ salts could be

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *

Statistical models to predict ion toxicity

Environ. Toxicol. Chern.

16.

1997

2015

1,000

2,000

3,000

4,000

LC50

with

feeding (mglL)

Fig. 1. Average LC50 values for

Ceriodaphnia dubia

exposed to

single salts with and without feeding.

adequately explained on the basis of the anion concentration

alone; in other words, NaCl and

CaCI, had approximately the

same toxicity when expressed on the basis of

C1-. A plot of

these data (Fig. 2) supports this conclusion and also shows a

good fit of the single salt model to these data. However, when

NaCl and

CaC12 were tested in combination, the resulting so-

lution was less toxic (on the basis of

C1- concentration) than

either of the solutions tested singly. The single salt model was

unable to account for this decreased toxicity and, consequently,

made poor predictions for the combined

NaCl/CaC12 solutions

(Fig. 2). The same trend toward lower toxicity of

C1- in the

presence of two cations was also evident for solutions con-

taining

K+ or Mg2+.

The double salt model compensated for the lower toxicity

of two cation solutions but only partially. The double salt

model simply fit a shallow

response curve between the single

cation and two cation data, predicting a "mean" probability

of survival somewhere between the observed single salt and

two salt survival values. While this compromise provided a

better overall fit to the data than did the single salt model, it

was clearly not a good representation of the response. Given

that the regression algorithm did not find any interaction terms

to be significant, it appeared that a new variable was required

to provide a better fit to the data.

1,000

2,000

3,000

4,OOC

Chloride (rng/L)

Fig. 2. The 48-h survival of

Ceriodaphnia dubia

exposed to solutions

enriched with

NaCI, CaCl,, or a

1 combination of NaCl and CaCl,,

normalized to CI- concentration. Curves represent regression model

predictions for .the single salt, double salt, and double salt with

NumCat models. Values at 0% and 100% offset slightly for clarity.

We attempted without success to derive a continuous vari-

able that would respond appropriately to the relative concen-

tration of cations in solution and thus identify the two cation

solutions as different than

solut~ons with a single catlon. After

our lack of success with continuous variables, we created a

categorical variable called NumCat. The NumCat variable was

intended to simply represent the number of major cations in

the solution. For the initial

modellng trials, the NumCat vari-

able was arbitrarily defined as the number of cations in the

solution that represented at least 10% of the total molar cation

concentration and that were also present at a concentration

greater than 100

mg/L. Our expectation was that the NumCat

variable would show a significant interaction with

C1- and any

other ion whose toxicity was influenced by the number of

cations present. The resulting model, called the "double salt

with NumCat" model, showed a markedly improved fit (R2

0.899); significant terms were the original five ions in the

single and double salt models, plus NumCat and the

Num-

Cat

C1, NurnCat

SO,, and NumCat

K interaction terms.

The NumCat

C1 term allowed the model to better represent

the toxicity of

NaC1, CaCl,, and NaCl

CaCl, solutions shown

in Figure

2. NumCat also showed significant (positive) inter-

actions with

SO:- and K+, suggesting that the presence of two

cations (or one additional cation in the case of

K') ameliorated

the toxicity of these ions as well.

After subsequent data collection and analysis, two addi-

tional steps were taken to optimize the NumCat variable. First,

we conducted supplemental testing of

dubia

exposed to

mixtures of three and four

C1- salts (data not shown). Modellng

of these data (NumCat

3 or 4) yielded a substantial under-

prediction of toxicity. Direct inspection of these data confirmed

that the protective effect observed with two cations did not

seem to increase with the addition of three or four cations.

Accordingly, we chose to limit the NumCat variable to values

0, 1, or 2; for solutions where the

10% and

100-mg1L

criterion yielded values of 3 or 4, these values were reset to 2.

The second step involved rigorously evaluating the defi-

nition criteria for the NumCat variable. Although the NumCat

variable was clearly effective at increasing the predictive ca-

pability of the model, its original definition had been arbitrary.

To provide a stronger technical basis for defining NumCat, we

conducted a sensitivity analysis by varying the two

compo-

~ents of the NumCat definition, the relative molar concentra-

tion (originally

>lo%), and the absolute concentration (orig-

inally 100

mg1L). A complete matrix of relative concentration

(0, 5, 10,

15,20, and 25%) and absolute concentration (0, 100,

200, and 300

mg/L) was modeled using 48-h

dubia

data.

The resultant models were evaluated based on their R2 values

(Fig. 3). The NumCat criteria that produced the model with

the highest R2 (best fit of the model to the observed data) were

the 15% with

>lo0 mg/L (R2

0.8559) and the 10% with

>loo-mg/L (R2

0.8553) criteria. Given that the difference

in R2 was only 0.0006 (0.06% of the variance) and that we

had already worked extensively with the 10% and

100-mglL

criteria, we elected to continue using these criteria in finalizing

the model equations.

After completion of data collection, final regression equa-

tions were developed to predict

dubia

survival after 24 and

48 h of exposure. Through the course of these analyses, several

additional variables and data transformations were evaluated

and discarded. Aside from the feeding and reference toxicant

variables discussed previously, we evaluated the sum of all

ions, the sum of all cations, the sum of all anions, and

NurnAn

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *

2016

Environ. Toxicol. Chem.

16, 1997

D.R. Mount et al

Percent of total cations

ppvd~?

ern;?

p_prn&:?

Fig.

Effect of varying criteria for the definition of the NumCat

variable. Circled point represents

the criteria selected initially and

maintained for final derivation of the regression equations.

(the anion equivalent of NumCat). First-order interactions of

these variables and ion concentrations were also evaluated.

None of these variables was selected as significant by the

regression algorithm. Models based on log-transformed ion

concentrations consistently showed lower R2 values than those

based on untransformed data.

The final 24- and 48-h equations for

dubia

had K+,

HCO;,

Mg2+, C1-, and SO:- as significant variables (Table 4).

Additionally, NumCat and the interaction terms NumCat

C1,

NumCat

SO,, and NumCat

K were found to be significant.

As had been the case since early in the modeling process, Na+

and Ca2+ concentrations were not significant variables except

as they affected the calculation of NumCat.

R' for the final

regressions were 0.861 and 0.842 for the 24-h and 48-h equa-

tions.

Model development for

magna

proceeded along the

same lines as those described for

dubia.

The initial model

developed using only single salt data fit those data very well

(R'

0.97) but was not as good at predicting survival for

more complex ion mixtures. As was observed for

dubia,

solutions with multiple cations tended to be less toxic than

comparable solutions with only one cation. As a result, when

all

magna

data were analyzed, NumCat was again selected

as a significant variable, both by itself and through its inter-

actions with

C1-, SO:-, and K+ (Table 4). In fact, all significant

terms in the

dubia

double salt model with NumCat were

also significant for

magna.

Quality of fit for the D.

magna

models was slightly lower than for the C.

dubia

models, though

still quite good (0.812 and 0.799).

As for the daphnids, modeling of the fathead minnow data

indicated that toxicity was a function of

K+, MgZ+, HCO;, %I-,

and SO:- concentrations, as neither Na+ nor CaZ+ were selected

as significant variables (Table 4). The primary difference in

the fathead minnow equations was that NumCat was not a

significant variable either by itself or in interaction with other

terms. R2 values for the three regression equations were gen-

erally comparable to those for the other models, ranging from

0.767 to 0.832.

Because of the large number of independent variables, the

actual response surface of the regression models cannot be

easily visualized. Nonetheless, marginal plots of the regression

equations can be used to illustrate the relative sensitivity of

each species to the various ions (Fig. 4). These plots show

that

dubia

are, in general, the most sensitive of the three

species to major ion toxicity, while fathead minnows are the

least sensitive.

K+ was the most toxic ion to all species and

SO:- the least. The only. inconsistency between species was

that Mg2+ was more toxic than HCO; for

magna

and fathead

minnows, but the reverse was true for

dubia.

As a means to visually evaluate the fit of the data sets to

the regression equations, each regression equation was used

to predict the ion concentrations producing 50% survival for

each of the ion combinations tested during data collection.

These values were then plotted against the average observed

LC50 values from Tables 2 and 3 (Fig. 5). These plots indicate

good overall agreement between the calculated and predicted

LC50 values for all three species. Note, however, that this

analysis is not a direct evaluation of quality of fit for the models

because it actually compares a point estimate derived from

individual logistic regression equations with the arithmetic

mean of multiple point estimates for specific ion combinations

derived by a different method (trimmed Spearman-Karber

LC50 estimation

[2 I]); it is not a plot of raw data versus model

predictions. There are other biases in this comparison as well,

such

asdifferent weighting of observations. Nevertheless, the

concordance between the two methods does provide some as-

surance that the single multiple regression models provide a

reasonable representation of the responses to a broad range of

ion combinations.

The absence of interaction terms in the final regression

equations, aside from those involving NumCat, suggests that

Table 4. Regression coefficients for final regression equationsa

Ceriodaphnia dubia

Daphnia magna

Fathead miunow

24-h

48-h

24-h

48-h

24-h

48-h

96-h

Constant

K+

Mg2+

Cl-

so:-

HCO;

NumCat

NumCattK+

NumCat*CI-

NumCat*S0,2-

Model R2

"Units for ion variables are mg/L.

indicates that this particular variable was not significant and was excluded from the model.

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *

Statistical models to predict ion toxicity

Environ. Toxicol. Chem. 

16, 1997

2017

0

1000

2000

3000

4000

5001

Concentration

(mg/L)

Fig. 4. Marginal plots of regression equations for each of the ions

selected as significant. For

Ceriodaphnia.dubia 

and

Daphnia magna 

models, NumCat

assuming additivity among individual ion toxicities is suffi-

cient to describe the toxicities of the ion mixtures, at least

from an empirical standpoint. The apparent amelioration of

C1-, SO:-, and K+ toxicity by multiple cations could be con-

strued as less than additivity. Alternatively, given that

Na+ and

Ca2+ were not clearly identified as toxic by themselves, it might

be more appropriate to consider those ions as water quality

variables influencing toxicity, rather than as components of a

toxic mixture.

We had little precognition of the important role that the

NumCat variable-would play in

represdnting the combined

toxicity of major ions. In a study of high-TDS irrigation return

waters, Dwyer et al.

[14] demonstrated that increasing the

hardness

(Mg2+ and Ca2+) of an NaC1-dominated water de-

creased-toxicity to

magna

and striped bass. For D.

magna,

this decreased toxicity would be predicted based on the current

research, as the addition of hardness to these waters would

have increased the value of the NumCat variable, thereby in-

creasing predicted survival. However, results of our study also

show that the effect of multiple cations is not an effect of

hardness per se. For example, the

C. dubia

48-h LC50 values

for NaCl

aid CaC1, were almost identical when expressed on

the basis of

C1- concentration (1,187 and 1,172 mgL, re-

spectively; Table 2), even though the solutions had greatly

different hardness. Moreover, the addition of NaCl to KC1

increased the

K+ concentration at the

C. dubia

48-h LC50

12.m

Oaphnla magna

.....

._..'

10,000                       

._.-.

......

: 1,m

.......

'12,m

Fathead mlnnows

....

......

6.000 

5.m

Fig. 5. Relationship between the ion concentrations predicted to cause

50% mortality and the average of LC50 values for individual salts

and salt combinations (Tables

2 and 3). Line of unity (slope

added for reference.

Ceriodaphnia dubia

....

._..

....

.......

.....

from 329 mg KL for KC1 to 458 mg KL for an NaCl

KC1

mix (Table 2), even though hardness was the same in both

solutions.

Despite its importance in modeling the response of

C. dubia

and

D. magna,

the NumCat variable was not selected as sig-

nificant for fathead minnows. Given that the addition of MgZ+/

CaZ+ improved survival of striped bass

in high TDS solutions

tested by Dwyer et al.

[14], it seems that the protective effect

in multiple cations is not restricted to cladocerans. It is worth

noting that combinations of two cations and one anion were

only tested once (in duplicate) for fathead minnows. If by

chance those test results had a systematic bias, it might mask

the presence of a cation-related effect for fathead minnows;

coefficients of variation for fathead minnow

LC5Os were high-

er than for the other two species. More testing would be re-

quired to confirm or deny this possibility.

Though the effect of multiple cations is quite consistent

both within the

C. dubia

and

D. magna

data sets and with

other research

[14], it must be emphasized that its identification

and quantification through our modeling is empirical. Not-

withstanding the effectiveness of the categorical NumCat vari-

able in modeling our data set, it seems reasonable from a

physiological standpoint to assume that the effect is in reality

some type of continuous function, rather than the step function

represented by our

10% and

100-mgL criteria. In our mod-

eling efforts, we were unable to devise a continuous variable

that corresponded to the observed influence of multiple cat-

ions. Nonetheless, with continued research it seems likely that

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *

20 18

Environ. Toxicol. Chem. 16, 1997

D.R. Mount et al.

such a relationship could be uncovered and, if so, might pro-

vide a more rigorous representation of the actual relationship

than that provided by NumCat as currently defined. A better

understanding of the mechanisms of major

Ion toxicity would

likely enhance this effort.

As a related matter, even though we conducted a sensitivity

analysis to determine the

optlmum criteria for the NumCat

variable, this analysis was subject to bias from the structure

of our data set. Specifically, we tested binary combinations of

salt solutions in 1: 1 ratios only. As such, only certain areas

within the total sampling space (all possible ion combinations)

were represented in the data set. Thus, there is no assurance

that the ion combinations tested were near critical points in

the response surface that might alter the apparent thresholds

for response. While we believe the NumCat variable is a

slg-

nificant advance in understanding the response of cladocerans

to high TDS solutions, it is probably a somewhat crude rep-

resentation of the actual physiological response.

Because most chemical reactions are related to molar con-

centrations, an argument could be made for modeling survival

on the basis of molar concentrations rather than mass-based

concentration. In retrospect, it seems this would have made

little difference

in the outcome of the modeling. As the equa-

tions are based on first-order concentrations of single ions only,

transformation between mass-based and molar concentrations

is a simple algebraic manipulation and does not affect the

nature of the response surface. In fact, the equations in Table

4 can be converted to a molar basis by simply dividing each

coefficient by the molecular weight of each ion.

Conversion

to chemical activity, however, would be much more involved.

Ultimately, the test of the toxicity models we have gen-

erated lies in

their, ability to make accurate predictions for

samples outside those used to generate the original data set.

Thus far, the equations have performed well in predicting ma-

jor ion toxicity in field-collected samples, particularly so for

the

dubia

equations. For example, Mount et al. [4] showed

a strong correlation

(Rz

0.95) between predicted and ob-

served survival of

dubia

exposed to ambient swples from

a watershed receiving oil field-produced waters. The

dubia

regression model was a better predictor of survival than any

individual ion concentration,

illushating the ability of the mod-

el to predict the combined toxic effects of multiple ions. In a

separate analysis, Mount et al.

[15] showed a shong relation-

ship between predicted and observed survival of

dubia

exposed to six produced waters collected from coalbed meth-

ane operations in Alabama. Obviously, these comparisons as-

sume that major ions were the primary cause of toxicity in

the field-collected samples.

Another application of the ion toxicity models that may

prove equally or even more valuable lies in using model pre-

dictions to determine whether the presence of toxicants other

than major ions is indicated. Research by Tietge et al. [5] both

demonstrates this application and

prov~des a rigorous evalu-

ation of the predictive capability of the regression models. Six

produced waters from various fossil fuel production sites were

tested for toxicity and analyzed for major ion concenhations.

The ion toxicity models presented here were used to predict

survival of

C. dubia,

rnagna,

and fathead minnows based

on major ion concentrations. Differences between observed

and

predicted toxicity were used to make inferences as to

whether the observed toxicity could be wholly

explained by

the major ion concentrations alone, or if the presence of other

toxicants was indicated. The accuracy of these inferences was

then evaluated by conducting Phase

I TIE manipulations [16]

and by testing the toxicity of laboratory waters reconstituted

to the same major ion concentrations. This study indicated that

the

C. dubia

model provided highly accurate predictions, while

the fathead minnow and

rnagna

models tended to overpre-

dict ion toxicity. The tendency of the D.

rnagna

and fathead

minnow models to overpredict toxicity in field-collected sam-

ples was also noted in comparisons made by Mount et al.

[4].

Dickerson et al. [7] used the C.

dubia

and fathead minnow

models to evaluate toxicity in surface waters influenced by

irrigation drain water. Although independent tests were not

performed to confirm model predictions, it appeared that pre-

dictions by the

dubia

model correlated well with observed

toxicity. As in the study by Tietge et al.

[5], however, the

fathead minnow model seemed to overpredict toxicity; several

sites had higher observed survival than was predicted by the

fathead minnow model.

In summary, applications of the

C. dubia

models and, to

a lesser extent the

rnagna

and fathead minnow models,

have proven them to be highly effective and comprehensive

tools for evaluating major ion toxicity. To date, they have been

successfully applied to studies of ambient waters

[15], pro-

duced waters

[4,5], irrigation drain waters [7], water purifi-

cation byproducts

[23], municipal effluents, and effluents from

pulp and paper, refining, and manufacturing industries (J.R.

Hockett, unpublished data). Because the models represent the

combined toxicity of all seven ions, they have much broader

application than ion toxicity studies based on generic measures

like conductivity or TDS, or studies focusing on certain waters

or ion combinations. Application of these models can reduce'

the need for extensive characterization and fractionation ma-

nipulations during TIE studies of high TDS waters

[l ll. They

can also be used to project changes in toxicity resulting from

modifications in industrial processes, effluent treatment, or oth-

er remedial measures.

Acknowledgement-Primary support for this research was provided

by the Gas Research Institute through contract 5091-253-2160 with

ENSR Consulting and Engineering. Analytical Technologies, Inc.,

conducted major ion analyses. J.P. Fillo contributed to the conceptual

development of the research and K.A.

Barten assisted in toxicity data

collection. G.T.

Ankley, S.R. Broderius, and A.M. Farag provided

helpful comments on earlier versions of the manuscript. Helpful com-

ments by

hvo anonymous reviewers are greatly appreciated. This

manuscript was reviewed in accordance with U.S. EPA policy. Men-

tion of trade names does not indicate endorsement by the Gas Research

Institute, U.S. EPA, or the Federal Government.

REFERENCES

U.S. Environmental Protection Agency.

1991. Methods for

measuring the acute toxicity of

efAuents to freshwater and marine

organisms, 4th ed.

EPA/600/4-901027. Cincinnati, OH.

U.S. Environmental Protection Agency.

1994. Short-termmeth-

ods for estimating the chronic toxicity of effluents and receiving

waters to freshwater organisms.

3rd ed.

EPAl60014-911002. Wash-

ington, DC.

Boelter, A.M.,

F.N.

Lamming, A.M. Farag

and

H.L. Bergman.

1992. Environmental effects of saline oil-field discharges on sur-

face waters. Environ. Toxicol. Chem. 11:

1 187-1 195.

Mount, D.R., D.D. Gulley and J.M. Evans.

1993. Salinityltox-

icity relationships to predict the acute toxicity of produced waters

to freshwater organisms. Proceedings,

I st Society of Petroleum

EngineersRT.S. Environmental Protection Agency Environmental

Conference. San Antonio, TX, USA pp. 605-614

Tietge, J.E., J.R. Hockett

and

J.M. Evans.

1997. Major ion

toxicity of six produced waters to three freshwater species: Ap-

plication of ion toxicity models and TIE procedures. Environ.

Toxicol. Chem.

16:2002-2008.

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *

Statistical models to predict ion toxicity

Environ. Toxicol. Chem.

16, 1997

2019

6. Ingersoll, C.G., F.J. Dwyer, S.A. Burch, M.K. Nelson, D.R.

Buckler and J.B. Hunn. 1992. The use of freshwater and salt-

water animals to distinguish between the toxic effects of salinity

and contaminants in irrigation drain water.

Environ. Toxicol.

Chem.

11:503-511.

7. Dickerson, K.K., W.A. Hubert and H.L. ~er~man.

1996. Tox-

icity assessment of water from lakes and wetlands receiving ir-

rigation drainwater.

Environ. Toxicol. Chem. 15:

1097-1 101.

8. Meyer, J.S., D.A. Sanchez, J.A. Brookman, D.B.

McWhorter

and H.L. Bergman. 1985. The chemistry and aquatic toxicity of

raw oil shale leachates from Piceance Basin, Colorado.

Environ.

Toxicol. Chem.

4:559-572.

9. Hoke, R.A., J.P. Giesy, M. Zabik and M. Unger. 1993. Toxicity

of sediments and sediment pore waters from the Grand Calumet

River-Indiana

Harbor, Indiana area of concern.

Ecotoxicol. En-

viron. Saj:

26:8&112.

10. Hoke, R.A., W.R. Gala, J.B. Drake, J.P. Giesy and S. Flegler.

1992. Bicarbonate as a potential confounding factor in cladoceran

toxicity assessments of pore water from contaminated sediments.

Can.

Fish. Aquat. Sci.

49:1633-1640.

11. Jop, K.M. and A.M. Askew. 1994. Toxicity identification eval-

uation' using a short-term chronic test with

Ceriodaphnia dubia.

.Bull. Environ. Contam. Toxicol.

53:91-97.

12. van Compernolle, R., P.B. Dorn, S.G. Newman, J.A. King and

J. Gumulka. 1993. Demonstration of the absence of underlying

acute toxicity in the presence of acutely toxic levels of salts of

group

1A and IIA cations.

Abstract,

14th Amual Meeting, Society

of Environmental Toxicology and Chemistry, Houston, TX, USA,

November 14-18, p. 91.

13. Burnham, B.L. and J.J. Peterka. 1975. Effects of saline water

from North Dakota lakes on survival of fathead minnow

(Pi-

mephales promelas)

embryos and sac fry.

J.

Fish. Res. Board

Can.

32809-812.

14. Dwyer, F.J., S.A. Burch, C.G. Ingersoll and J.B. Hunn. 1992.

Toxicity of trace element and salinity mixtures to striped bass

(Morone saxatilis)

and

Daphnia magna. Environ. Toxicol. Chem.

11:513-520.

15. Mount, D.R., K.R. Drottar, D.D. Gulley,'J.P. Fillo and P.E.

O'Neil. 1992. Use of laboratory toxicity data for evaluating the

environmental acceptability of produced water discharge to sur-

face waters. In J.P. Ray

and ER. Engelhardt, eds.,

Produced Wa-

ter:

Technological/Environmental 

Issues and Solutions.

Plenum,

New York,

NY, USA, pp. 175-185.

16. U.S. Environmental Protection Agency. 1991. Methods for

aquatic toxicity identification evaluations. Phase I toxicity char-

acterization procedures, 2nd ed.

EPAl60016-911003. Washington,

DC.

17. U.S. Environmental Protection Agency. 1994. Methods for the

determination of metals in environmental samples. Supplement

1. EPA/600/R-941111. Washington, DC.

18. U.S. Environmental Protection Agency. 1983. Methods for

chemical analysis of water and wastes.

EPAl60014-791020. Cin-

cinnati, OH.

19. American Public Health Association. 1989.

Standard Methods

for the Examination of Water and Wastewater.

Washington, DC.

20. Dixon, W.J. 1985.

BMIP Statistical Software.

University of

California Press, Berkeley, CA, USA.

21.

HamiIton, M.A., R.C. Russo and R.V. Thurston. 1977.

Trimmed Spearman-Karber method for estimating median lethal

concentrations in toxicity bioassays.

Environ. Sci. Technol.

11:

714-719. Correction 12:417(1978).

22. Mount, D.R. and D.D. Gulley. 1992. Development of a salinity1

toxicity relationship to prediet acute toxicity of saline waters to

freshwater organisms. GRI-9210301. Gas Research Institute, Chi-

cago, IL, USA.

23. Florida Department of Environmental Protection. 1995. Pro-

tocols for determining major-seawater ion toxicity in membrane

technology water-treatment concentrate. Bureau of Laboratories,

Tallahassee, FL, USA.

Electronic Filing, Received, Clerk's Office, June 7, 2007

* * * * * * PC #8 * * * * * *