1                                              NCEA-C-1763
 2                                             November 2006
 3
   Toxicological Reviews of Cyanobacterial
           Toxins: Cylindrospermopsin
 9

10

11

12

13

14

15

16

17

18

19

20

21

22
23               National Center for Environmental Assessment
24                  Office of Research and Development
25                  U.S. Environmental Protection Agency
26                      Cincinnati, OH 45268

-------
1                                      DISCLAIMER
2
3
4          This report is an external draft for review purposes only and does not constitute Agency
5   policy. Mention of trade names or commercial products does not constitute endorsement or
6   recommendation for use.
                                             ii         DRAFT: DO NOT CITE OR QUOTE

-------
 1                               TABLE OF CONTENTS
 2
 3
 4                                                                           Page
 5
 6   LIST OF TABLES	v
 7   LIST OF FIGURES	vi
 8   LIST OF ACRONYMS	vii
 9   PREFACE	viii
10   AUTHORS, CONTRIBUTORS AND REVIEWERS	ix
11   ACKNOWLEDGMENTS	x
12
13   1.    INTRODUCTION	1
14
15   2.    CHEMICAL AND PHYSICAL INFORMATION	3
16
17   3.    TOXICOKINETICS	4
18         3.1.   ABSORPTION	4
19         3.2.   DISTRIBUTION	4
20         3.3.   METABOLISM	5
21         3.4.   ELIMINATION	5
22         3.5.   PHYSIOLOGICALLY BASED TOXICOKINETIC MODELS	6
23
24   4.    HAZARD IDENTIFICATION	7
25         4.1.   STUDIES IN HUMANS - EPIDEMIOLOGY, CASE REPORTS,
26               CLINICAL CONTROLS	7
27         4.2.   ACUTE, SHORT-TERM, SUBCHRONIC AND CHRONIC
28               STUDIES AND CANCER BIO AS SAYS IN ANIMALS - ORAL
29               AND INHALATION	8
30               4.2.1. Oral Exposure	9
31                     4.2.1.1. Acute Studies	9
32                     4.2.1.2. Short-Term Studies	10
33                     4.2.1.3. Subchronic Studies	11
34                     4.2.1.4. Chronic Studies	13
35               4.2.2. Inhalation Exposure	13
36         4.3.   REPRODUCTIVE/DEVELOPMENTAL STUDIES - ORAL AND
37               INHALATION	13
38         4.4.   OTHER STUDIES	14
39               4.4.1. Effects by Parenteral Exposure	14
40                     4.4.1.1. Studies of Purified Cylindrospermopsin	14
41                     4.4.1.2. Cell Extract Studies	14
42               4.4.2. Immunotoxicity	15
43               4.4.3. Tumor Initiation	15
44               4.4.4. Genotoxicity	16
                                         iii       DRAFT: DO NOT CITE OR QUOTE

-------
 1                              TABLE OF CONTENTS cont.
 2
 3
 4                                                                                 Page
 5
 6         4.5.   MECHANISTIC DATA AND OTHER STUDIES IN SUPPORT OF
 7               THE MODE OF ACTION	16
 8               4.5.1.  Liver Toxicity	16
 9               4.5.2.  Kidney Toxicity	18
10               4.5.3.  Interactions with DNA and RNA	19
11               4.5.4.  Structure-Activity Relationships	20
12         4.6.   SYNTHESIS AND EVALUATION OF MAJOR NONCANCER
13               EFFECTS	21
14               4.6.1.  Oral	21
15               4.6.2.  Inhalation	23
16               4.6.3.  Mode of Action Information	23
17         4.7.   WEIGHT-OF-EVIDENCE EVALUATION AND CANCER
18               CHARACTERIZATION	24
19               4.7.1.  Summary of Overall Weight-of-Evidence	24
20               4.7.2.  Synthesis of Human, Animal and Other Supporting Evidence	24
21         4.8.   SUSCEPTIBLE POPULATIONS AND LIFE STAGES	24
22               4.8.1.  Possible Childhood Susceptibility	24
23               4.8.2.  Possible Gender Differences	25
24               4.8.3.  Other Possible Susceptible Populations	25
25
26   5.     DOSE-RESPONSE ASSESSMENTS	26
27         5.1.   NARRATIVE DESCRIPTION OF THE EXTENT OF THE
28               DATABASE	26
29         5.2.   ORAL REFERENCE DOSE (RfD)	26
30               5.2.1.  Data Considered in Deriving Reference Values	26
31               5.2.2.  Acute Duration	26
32                      5.2.2.1. Choice of Principal Study and Critical Effect - with
33                              Rationale and Justification	27
34               5.2.3.  Short-Term Duration	27
35                      5.2.3.1. Choice of Principal Study and Critical Effect-with
36                              Rationale and Justification	27
37               5.2.4.  Subchronic Duration	27
38                      5.2.4.1. Choice of Principal Study and Critical Effect - with
39                              Rationale and Justification	27
40                      5.2.4.2. Methods of Analysis - Including Models (PBPK, MBD, etc.) ...28
41                      5.2.4.3. RfD Derivation - Including Application of UFs	28
42               5.2.5.  Chronic Duration	32
43                      5.2.5.1. Choice of Principal Study and Critical Effect - with
44                              Rationale and Justification	32
45               5.2.6.  Route-to-Route Extrapolation	33
46         5.3.   INHALATION REFERENCE CONCENTRATION (RfC) 	33
47         5.4.   CANCER ASSESSMENT	33
                                             iv        DRAFT: DO NOT CITE OR QUOTE

-------
 1                         TABLE OF CONTENTS cont.
 2
 3
 4                                                                  Page
 5
 6   6.    MAJOR CONCLUSIONS IN THE CHARACTERIZATION OF HAZARD
 7        AND DOSE RESPONSE	34
 8        6.1.   HUMAN HAZARD POTENTIAL	34
 9        6.2.   DOSE RESPONSE	34
10
11   7.    REFERENCES	35
                                            DRAFT: DO NOT CITE OR QUOTE

-------
 1                                     LIST OF TABLES
 2
 3
 4    No.                                     Title                                   Page
 5
 6    4-1   Tumor Initiating Activity of C. raciborskii Extracts	16
 7
 8    4-2   Summary Results of Major Oral Toxicity Studies of Pure
 9          Cylindrospermopsin in Experimental Animals	22
10
11    5-1   Relative Kidney Weights in Mice Exposed to Purified
12          Cylindrospermopsin for 11 Weeks	29
13
14    5-2   Summary of Benchmark Dose Modeling (Relative Kidney Weight)	30
                                              vi         DRAFT: DO NOT CITE OR QUOTE

-------
1                                  LIST OF FIGURES
2
3
4
5   No.                                   Title                                   Page
6
7   2-1   Chemical Structure of Cylindrospermopsin	3
8
9   5-1   Linear Model Fit to Relative Kidney Weight Data (High Dose Group Dropped)	31
                                           vii        DRAFT: DO NOT CITE OR QUOTE

-------
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
AIC
BMD
BMDL
BMDS
BMR
CYP450
DMSO
EPA
GFR
HPLC
i.p.
LD50
LOAEL
NOAEL
PBPK
POD
RfC
RfD
ROS
THP
TPA
UF
                LIST OF ABBREVIATIONS






Aikake's Information Criteria




Benchmark dose




Statistical lower confidence limit on the benchmark dose




Benchmark dose software




Benchmark response




Cytochrome P-450




Dimethyl sulfoxide




Environmental Protection Agency




Glomerular filtration rate




High performance liquid chromatography




Intraperitoneal




Dose lethal to 50% of the population




Lowest-observed-adverse-effect level




No-observed-adverse-effect level




Physiologically based pharmacokinetic




Point of departure




Reference concentration




Reference dose




Reactive oxygen species




Tamm-Harsfall protein




O-Tetradecanoylphorbol  13-acetate




Uncertainty factor
                            Vlll
DRAFT: DO NOT CITE OR QUOTE

-------
 1                                          PREFACE
 2
 3
 4          The Safe Drinking Water Act (SDWA), as amended in 1996, requires the Environmental
 5   Protection Agency (EPA) to publish a list of contaminants that, at the time of publication, are not
 6   subject to any proposed or promulgated national primary drinking water regulations, are known
 7   or anticipated to occur in public water systems, and may require regulations under SDWA. This
 8   list, known as the Contaminant Candidate List (CCL), was first published in 1998 and then again
 9   in 2005.  The 1998 and 2005 CCLs include "cyanobacteria (blue-green algae), other freshwater
10   algae, and their toxins" as microbial contaminants.
11
12          In 2001, a meeting was held among EPA, researchers from the drinking water industry,
13   academia and government agencies with expertise in the area of fresh water algae and their
14   toxins. The goal of this meeting was to convene a panel of scientists to assist in identifying a
15   target list of algal toxins that are likely to pose a health risk in source and finished waters of the
16   drinking water utilities  in the U.S. Toxin selection was based on four criteria: health effects,
17   occurrence in the United States, susceptibility to drinking water treatment and toxin stability.
18   Cylindrospermopsin was identified at this meeting as being a toxin of high priority  based on
19   those criteria.
20
21          The National Center for Environmental Assessment has prepared this Toxicological
22   Review of Cyanobacterial Toxins: Cylindrospermopsin as one in a series of dose-response
23   assessments to support  the health assessment of unregulated contaminants on the CCL.  The
24   purpose of this document is to compile and evaluate the available data regarding
25   Cylindrospermopsin toxicity to aid the Office of Water in regulatory decision making. It is not
26   intended to be a comprehensive treatise on the chemical or toxicological nature of
27   Cylindrospermopsin.
28
29          In Section 6, Major  Conclusions in the Characterization of Hazard and Dose Response,
30   EPA has characterized  its overall confidence in the quantitative and qualitative aspects of the
31   hazard and dose response by addressing knowledge gaps,  uncertainties, quality of data and
32   scientific controversies. The discussion is intended to convey the limitations of the assessment
33   and to  aid and guide the Office of Water in the ensuing steps of the human health risk assessment
34   of Cylindrospermopsin.
                                                ix         DRAFT: DO NOT CITE OR QUOTE

-------
 1                     AUTHORS, CONTRIBUTORS AND REVIEWERS
 2
 3
 4   AUTHORS
 5
 6   Belinda Hawkins, Ph.D. (Chemical Manager)
 7   National Center for Environmental Assessment
 8   Office of Research and Development
 9   U.S. Environmental Protection Agency
10   Cincinnati, OH
11
12   Stephen Bosch
13   Syracuse Research Corporation
14   Syracuse, NY
15
16   Marc Odin
17   Syracuse Research Corporation
18   Syracuse, NY
19
20   David Wohlers
21   Syracuse Research Corporation
22   Syracuse, NY
23
24
25   REVIEWERS
26
27   INTERNAL EPA REVIEWERS
28
29   Joyce Donohue, Ph.D.
30   Office of Water
31   Washington, DC
32
33   Elizabeth Hilborn, D.V.M., M.P.H.
34   National Health and Environmental Effects Research Laboratory
35   Office of Research and Development
36   Research Triangle Park, NC
37
38   James Sinclair, Ph.D.
39   Office of Water
40   Cincinnati, OH
41
42   Jeff Swartout
43   National Center for Environmental Assessment
44   Office of Research and Development
45   Cincinnati, OH
46
                                                        DRAFT: DO NOT CITE OR QUOTE

-------
1   EXTERNAL REVIEWERS
2
3
4   ACKNOWLEDGMENTS
5
                                   xi      DRAFT: DO NOT CITE OR QUOTE

-------
 1                                     1. INTRODUCTION
 2
 3
 4          This toxicological review presents background and justification for hazard and dose-
 5   response assessments of cylindrospermopsin. U.S. Environmental Protection Agency (EPA)
 6   toxicological reviews may include oral reference doses (RfD) and inhalation reference
 7   concentrations (RfC) for chronic and less-than-lifetime exposure durations and a carcinogenicity
 8   assessment.
 9
10          The RfD and RfC provide quantitative information for use in risk assessments for health
11   effects known or assumed to be produced through a nonlinear (possibly threshold) mode of
12   action. These reference values are defined as an estimate of an exposure, designated by duration
13   and route, to the human population (including susceptible subgroups), that is likely to be without
14   an appreciable risk of adverse effects. Reference values may be derived for acute (<24 hours),
15   short-term (up to 30 days), subchronic (up to 10% of average lifespan) and chronic (up to
16   lifetime) exposures, all considered to be continuous exposures throughout the duration  specified.
17   A reference value is derived from a BMDL (a statistical lower confidence limit on the
18   benchmark dose), a no-observed-adverse-effect level (NOAEL), lowest-observed-adverse-effect
19   level (LOAEL) or other suitable point of departure with uncertainty/variability factors applied to
20   reflect limitations of the data used. The RfD is expressed in units of mg/kg-day, and the RfC  in
21   units of mg/m3.
22
23          The carcinogenicity assessment provides information on the carcinogenic hazard
24   potential of the  substance in question and quantitative estimates of risk from oral exposure and
25   inhalation exposure. The information includes a weight-of-evidence judgment of the likelihood
26   that the agent is a human carcinogen and  the conditions under which the carcinogenic effects
27   may be expressed.  Quantitative risk estimates are presented in three ways.  The slope factor is
28   the result of application of a low-dose extrapolation procedure and is presented as the risk per
29   mg/kg-day.  The unit risk is the quantitative estimate in terms of either risk per |J,g/L drinking
30   water or risk per |J,g/m3 air breathed.  Another form in which risk is presented is a drinking water
31   or air concentration providing cancer risks of 1 in 10,000; 1 in 100,000; or 1 in 1,000,000.
32
33          Development of these hazard identification and dose-response assessments for
34   cylindrospermopsin has followed the general guidelines for risk assessment as set forth by the
35   National Research Council (NRC,  1983). EPA guidelines that were used in the development  of
36   this assessment include the following: Guidelines for the Health Risk Assessment of Chemical
37   Mixtures (U.S. EPA, 1986a), Guidelines for Mutagenicity Risk Assessment (U.S. EPA,  1986b),
3 8   Guidelines for Developmental Toxicity Risk Assessment (U.S. EPA, 1991), Guidelines for
39   Reproductive Toxicity Risk Assessment (U.S. EPA, 1996), Guidelines for Neurotoxicity Risk
40   Assessment (U.S. EPA, 1998a), Guidelines for Carcinogen Assessment (U.S. EPA, 2005a),
41   Supplemental Guidance for Assessing Susceptibility from Early-Life Exposure to Carcinogens
42   (U.S.  EPA, 2005b), Recommendations for and Documentation of Biological Values for Use in
43   Risk Assessment (U.S. EPA, 1988), (proposed) Interim Policy for Particle Size and Limit
44   Concentration Issues in Inhalation Toxicity (U.S. EPA, 1994a), Methods for Derivation of
45   Inhalation Reference Concentrations and Application of Inhalation Dosimetry (U.S. EPA,
46   1994b), Use of the Benchmark Dose Approach in Health Risk Assessment (U. S. EPA, 1995),
                                                          DRAFT: DO NOT CITE OR QUOTE

-------
 1    Science Policy Council Handbook: Peer Review (U.S. EPA, 1998b, 2000a), Science Policy
 1    Council Handbook. Risk Characterization (U.S. EPA, 2000b), Benchmark Dose Technical
 3    Guidance Document (U.S. EPA, 2000c) and Supplementary Guidance for Conducting Health
 4    Risk Assessment of Chemical Mixtures (U.S. EPA, 2000d) and^4 Review of the Reference Dose
 5    and Reference Concentration Processes (U.S. EPA, 2002).
 6
 7          Literature searches were conducted for studies relevant to the derivation of toxicity and
 8    carcinogenicity values for cylindrospermopsin. The following databases were searched:
 9    MEDLINE (PubMed), TOXLINE, BIOSIS, CANCERLIT, TSCATS, CCRIS, DART/ETIC,
10    EMIC, GENETOX, HSDB and RTECS. The relevant literature was reviewed through May
11    2006.
                                                      DRAFT: DO NOT CITE OR QUOTE

-------
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
                   2.  CHEMICAL AND PHYSICAL INFORMATION
       Cylindrospermopsin is a naturally occurring toxin produced by particular strains of
Cylindrospermopsis raciborskii and at least four other freshwater cyanobacterial species,
including Umezakia natans, Aphanizomenon ovalispomm, Anabaena bergii and Raphidiopsis
curvata (Fastner et al., 2003). The chemical structure of Cylindrospermopsin was not elucidated
until 1992.  It consists of a tricyclic guanidine moiety combined with hydroxymethyluracil
(Figure 2-1) (Humpage and Falconer, 2003; Ohtani et al., 1992), has a molecular formula of
CisH^iNsOyS and a molecular weight of 415.43 (Lewis, 2000). It is zwitterionic (i.e., a dipolar
ion with localized positive and negative charges) (Ohtani et al., 1992). Deoxycylindro-
spermopsin, an analog of Cylindrospermopsin in which the hydroxyl group on the uracil bridge
(C-7) has been removed, has been isolated from C. raciborskii and R. curvata (Li et al., 2001;
Norris et al., 1999). Another structural variant of Cylindrospermopsin, 7-epicylindrospermopsin,
was isolated from A. ovalisporum (Banker et al., 2000).
                   Figure 2-1. Chemical Structure of Cylindrospermopsin*
* Conformations of steriocenters within the structure are indicated as either R or S. The numbers 7 and 12 indicate
carbon positions for identification purposes.
       Cylindrospermopsin is a white powder that is highly soluble in water (Ohtani et al., 1992;
Sigma, 2006). It is also soluble in dimethylsulfoxide (DMSO) and methanol (Sigma, 2006).
Cylindrospermopsin is chemically stable in sunlight, at high temperatures and through a wide
range of pH values (Chiswell et al., 1999).  Additional chemical and physical property data are
not available in the open literature for Cylindrospermopsin (HSDB, 2006; Lewis, 2000; O'Neil,
2001). This substance is produced on a small scale for research purposes (Sigma, 2006).
                                                           DRAFT: DO NOT CITE OR QUOTE

-------
 1                                    3.  TOXICOKINETICS
 2
 3
 4   3.1.    ABSORPTION
 5
 6          No quantitative data were located regarding the rate or extent of absorption of
 7   cylindrospermopsin in humans or animals following oral, inhalation or dermal exposure.
 8   Absorption of cylindrospermopsin from the gastrointestinal tract of mice is demonstrated by the
 9   induction of hepatic and other systemic effects in 14-day and 11-week oral toxicity studies of
10   pure cylindrospermopsin (Humpage and Falconer, 2003; Shaw et al., 2000, 2001) (see Section
11   4.2.1).
12
13   3.2.    DISTRIBUTION
14
15          No information was located regarding the tissue distribution of cylindrospermopsin
16   following oral, inhalation or dermal exposure.  The distribution and elimination of
17   intraperitoneally (i.p.) administered 14C-cylindrospermopsin (>95% pure; extracted and purified
18   from lyophilized C. raciborskii cells) in normal saline was studied in male Quackenbush mice in
19   a series of experiments using sublethal and lethal dose levels of the chemical (Norris et al.,
20   2001). In one experiment, four mice were given a single sublethal dose of 0.1  mg/kg, and urine
21   and feces were collected for the following 48 hours. Most of the 14C was eliminated in the urine
22   and feces, as discussed in Section 3.4.  Analysis of liver, kidneys and spleen at 48 hours showed
23   mean 14C recovery of 13.1% of the dose in the liver and <1% in the other tissues.  Total recovery
24   of radiolabel was 85-90% of the administered dose in each of the four mice.
25
26          The second experiment reported by Norris et al. (2001) included 12 mice administered a
27   single 0.2 mg/kg dose of 14C-cylindrospermopsin, which is the approximate median lethal i.p.
28   dose (Norris et al., 2001).  14C content was determined in the urine and feces in all animals after
29   12 and 24 hours, and in the liver, kidneys and spleen in five mice that were euthanized after 5-6
30   days due to toxicity (effects not specified) and after 7 days in the surviving 7 mice that had no
31   signs of toxicity. Most of the 14C was eliminated in the urine and feces, as discussed in Section
32   3.4.  The overall mean recoveries of 14C in the liver, kidneys and spleen after 5-7 days were 2.1,
33   0.15 and <0.1% of the dose, respectively.  Comparison of data from four mice with signs of
34   toxicity and four mice without  signs of toxicity showed no clear relationship between toxicity
35   and patterns of tissue distribution,  although a trend toward decreased liver retention in the
36   surviving mice was suggested.
37
38          Norris et al. (2001) reported a third experiment, in which excretion and tissue distribution
39   were assessed in four mice that were given a 0.2 mg/kg i.p. dose of 14C-cylindrospermopsin and
40   evaluated after 6 hours (Norris et al., 2001). 14C was detected in all tissues that were examined
41   (liver, kidney, heart, lung, spleen, blood and bile), but occurred predominantly in the liver and
42   kidneys (20.6 and 4.3% of the dose, respectively). Approximately 60% of the administered dose
43   of 14C was eliminated in the urine and feces (see Section 3.4).
44
                                                           DRAFT: DO NOT CITE OR QUOTE

-------
 1   3.3.    METABOLISM
 2
 3          The distribution and elimination of i.p administered 14C-cylindrospermopsin (>95% pure;
 4   extracted and purified from lyophilized C. raciborskii cells) in saline was studied in a series of
 5   mouse experiments (Norris et al., 2001), as detailed in Sections 3.2 and 3.4. Urine, fecal, liver
 6   and kidney samples from these studies were extracted with methanol to precipitate proteins, and
 7   the 14C in the supernatant was fractionated using high performance liquid chromatography
 8   (HPLC) for the detection of metabolites.  No attempt was made to fractionate or otherwise
 9   identify the 14C in the protein precipitate.  Analysis of methanol extracts of urine samples
10   collected for 12 hours following a single dose of 0.1 mg/kg (4 mice) or 0.2 mg/kg (12 mice)
11   suggested that a large part  (72%) of the excreted 14C was present as cylindrospermopsin (as
12   determined by retention times).  Some (-23.5%) of the urinary 14C was detected in protein
13   precipitated by the methanol, suggesting the presence of a protein-bound metabolite. The
14   authors did not indicate whether the level of protein in the urine was normal or abnormal. Most
15   (94.3%) of the 14C in an aqueous extract of the feces had the same retention time as
16   cylindrospermopsin, but only one mouse dosed with 0.2 mg/kg was tested. Analysis of liver
17   tissue showed the presence of 14C in both methanol extract and protein precipitate.  When
18   fractionated by HLPC, the extracted 14C had the same elution characteristics seen in some of the
19   urine methanol extracts, suggesting the presence of the same metabolite. The authors could not
20   rule out the possibility that the non-extractable 14C in the liver was protein-bound
21   cylindrospermopsin itself,  although the evidence for metabolic activation of cylindrospermopsin
22   in other studies (Runnegar et al.,  1995; Shaw et al., 2000) suggested that it might also be a
23   metabolite. The methanol-extractable metabolite was not found in kidney tissue. No
24   identification of metabolites was performed.
25
26          There is evidence indicating that the hepatic cytochrome P-450 (CYP450) enzyme
27   system is involved in the metabolism  and toxicity  of cylindrospermopsin. As discussed in
28   Section 4.5.1, pretreatment of hepatocytes with known inhibitors of CYP450 diminished the in
29   vitro cytotoxicity of cylindrospermopsin (Froscio  et al., 2003; Runnegar et al.,  1995). Similarly,
30   pretreatment of mice with a CYP450 inhibitor protected against the acute lethality of
31   cylindrospermopsin (Norris et al., 2002). Additionally, a main target of cylindrospermopsin
32   toxicity is the periacinar region of the liver, which is where CYP450-mediated  xenobiotic
33   metabolism occurs (Shaw et al., 2000, 2001).
34
35   3.4.    ELIMINATION
36
37          No information was located regarding the elimination of cylindrospermopsin following
38   oral, inhalation or dermal exposure. The elimination of i.p administered 14C-cylindrospermopsin
39   (>95% pure; extracted and purified from lyophilized C. raciborskii cells) in saline was studied in
40   male Quackenbush mice in a series of experiments using sublethal and lethal dose levels of the
41   chemical (Norris et al., 2001). In one experiment, four mice were given a single sublethal dose
42   of 0.1 mg/kg, and urine and feces were collected for the following 48 hours.  The mean
43   cumulative excretion of 14C in the first 12  hours after dosing was 62.8% of the  administered dose
44   in the urine and 15.5% in the feces. There was little additional excretion of 14C in either the
45   urine or feces following 12 additional hours. The  15.5% mean fecal excretion value reflects a
46   very high fecal excretion in one of the four animals (nearly 60% of the dose compared to <5% in
                                                          DRAFT: DO NOT CITE OR QUOTE

-------
 1   the other mice); the authors considered the possibility that the high value in the one animal
 2   resulted from the injection entering the upper gastrointestinal tract, but concluded that this
 3   possibility was unlikely given the injection technique used, the recovery of 4.7% of the injected
 4   dose in the liver after 48 hours and a similarly high fecal excretion of 14C in another animal in
 5   the third experiment in this study (discussed below). Total mean recovery in the urine, feces,
 6   liver, kidneys and spleen was 85-90% of the 14C dose in each of the four mice.
 7
 8          The second experiment reported by Norris et al. (2001) included 12 mice administered a
 9   single 0.2 mg/kg dose of 14C-cylindrospermopsin, which is the approximate median lethal i.p.
10   dose (Norris et al., 2001).  Five of the 12 dosed animals died within 5-6 days (signs of toxicity
11   not reported).  14C content was determined in the urine and feces in all animals after 12 and 24
12   hours. Results were similar to those obtained with a sublethal dose (reported above),  except that
13   there was some continued urinary and fecal excretion over the second 12 hours of the monitoring
14   period.  The mean cumulative urinary and fecal excretion of 14C was 66.0 and 5.7% of the dose
15   within 12 hours, and 68.4 and 8.5% of the dose within 24 hours, respectively. The mean total
16   recovery in the urine and feces after 24 hours was 76.9% of the administered dose.  The overall
17   mean recoveries of 14C in the liver,  kidneys and spleen after 5-7 days were 2.1, 0.15 and <0.1%
18   of the administered dose, respectively. Comparison of data from four mice with signs of toxicity
19   and four mice without signs of toxicity showed no clear relationship  between toxicity and
20   patterns of excretion, although trends toward increased urinary excretion and decreased fecal
21   excretion in surviving mice were suggested.
22
23          Norris et al. (2001) reported a third experiment in which four mice were given a 0.2
24   mg/kg i.p. dose of 14C-cylindrospermopsin and evaluated for 6 hours (Norris et al., 2001). The
25   mean cumulative urinary and fecal excretion of 14C after 6 hours was 48.2 and 11.9% of the
26   administered dose, respectively. One of the  four mice showed more  than 40% of the dose in the
27   feces (additional data not reported).
28
29   3.5.   PHYSIOLOGICALLY BASED TOXICOKINETIC MODELS
30
31          No physiologically based toxicokinetic models have been developed for
32   cylindrospermopsin.
                                                           DRAFT: DO NOT CITE OR QUOTE

-------
 1                              4. HAZARD IDENTIFICATION
 2
 3
 4   4.1.    STUDIES IN HUMANS - EPIDEMIOLOGY, CASE REPORTS, CLINICAL
 5          CONTROLS
 6
 7          An outbreak of a hepatoenteritis-like illness occurred in 148 residents of aboriginal
 8   descent in the Palm Island community in Queensland, Australia, in 1979 (Blyth, 1980; Griffiths
 9   and Saker, 2003). The total number of people exposed was not reported.  Of the 148 cases, 138
10   were children (mean age 8.4 years, range 2-16 years, 41% male and 59% female) and 10 were
11   adults (sex and age not reported).  The majority of the cases in the outbreak, called the "Palm
12   Island mystery disease," required hospitalization.  The clinical symptoms included fever,
13   headache, vomiting, profuse bloody diarrhea, hepatomegaly and renal damage as indicated by
14   loss of water, electrolytes, proteins, ketones and carbohydrates. Many of the individuals required
15   intravenous therapy for electrolyte imbalance and, in some cases, for hypovolemic and acidotic
16   shock. The findings may indicate increased susceptibility of children unless the 138 children
17   were from the households of the 10 adults (not indicated) or if there was differential exposure
18   between the children and the adults (not indicated); the child:adult ratio is approximately 14:1.
19   A few days prior to the outbreak, the major drinking water supply for the island, Solomon Dam
20   reservoir,  had been treated with unreported levels of copper sulfate to control a dense algal
21   bloom; only households connected to the reservoir were  affected by the illness. Retrospective
22   analyses, including epidemiological and ecological assessments, implicated the predominant
23   cyanobacterial species in the reservoir, C. raciborskii, as the likely source of the illness (Griffiths
24   and Saker, 2003; Hawkins et al., 1985). Intraperitoneal injection of cell extracts of C. raciborskii
25   from the reservoir caused damage to the liver, kidneys and other organs in mice (Hawkins et al.,
26   1985), and the toxin was later identified as cylindrospermopsin (Ohtani et al., 1992).  Some
27   symptoms of acute oral exposure to high concentrations  of copper sulfate, including headache,
28   nausea, vomiting  and diarrhea (HSDB, 2006), are similar to those observed during the outbreak.
29   The only information that was located regarding a potential role of the copper sulfate treatment
30   in the outbreak is  an indication that its algalcidal mode of action,  cell lysis, could have
31   contributed to the release of cylindrospermopsin and other cellular toxins into the water
32   (Griffiths  and Saker, 2003).
33
34          Hayman (1992) investigated reports of disease (sometimes called "Barcoo fever") in the
35   Australian outback dating back as far as 1887. He concluded that the reported symptoms were
36   similar to  those of the Palm Island mystery disease and that they might have been  caused by
37   exposure to C. raciborskii.  No additional information was located regarding effects in humans
38   known or  suspected to be associated with exposure to cylindrospermopsin.
39
40          An outbreak of acute liver failure occurred in patients at a renal dialysis clinic in Caruaru,
41   Brazil (Carmichael et al., 2001). Following routine hemodialysis treatment during a week in
42   February 1996, 116 of 131 patients experienced headache,  eye pain, blurred vision, nausea and
43   vomiting.  Subsequently,  100 of the affected patients developed acute liver failure and, of these,
44   76 died. Analysis of the clinic's water treatment system (samples of carbon, sand and
45   cation/anion exchange resin from in-house filters) for microcystins and cylindrospermopsin
46   showed the presence of both cyanotoxins. Analyses of blood sera and liver samples revealed
                                                           DRAFT: DO NOT CITE OR QUOTE

-------
 1   microcystins, but not cylindrospermopsin, although the method used to extract
 2   cylindrospermopsin from these samples may have been inadequate. Based on a comparison of
 3   victims' symptoms and liver pathology using animal studies of microcystins and
 4   cylindrospermopsin, it was concluded that the major contributing factor to death of the dialysis
 5   patients was intravenous exposure to microcystins.
 6
 7          The skin irritant potential of cylindrospermosin was evaluated using skin-patch testing in
 8   humans (Pilotto et al., 2004).  Both whole and lysed preparations of laboratory-grown C.
 9   raciborskii cells were applied to the skin of 50 adult volunteers using adhesive patches divided
10   into 10 individual filter pad-containing chambers. Each volunteer was exposed to one patch for
11   whole cells and one patch for lysed cells with each patch containing six cell concentrations, two
12   positive controls (1 and 5% solutions of sodium lauryl sulfate), and two negative controls
13   (culture media and an empty patch). The concentrations (densities) of cells were consistent with
14   those found in C. raciborskii-contaming water bodies used for recreational water activities.
15   Patches were removed after 24 hours and erythematous reactions were graded as 0 (no reaction
16   or erythema), 1 (minimal or very weak spotty erythema), 2 (mild diffuse erythema), 3 (moderate
17   diffuse erythema) or 4 (severe diffuse erythema with edema) by a dermatologist blinded to the
18   cell type and concentration. The distribution of clinical gradings by patch type (control/active),
19   cell type and cell concentration was assessed using logistic regression modeling. Due to a
20   relatively  small number of high-level gradings, each  observation was dichotomized into no
21   reaction (grade 0) and a positive reaction (1, 2, 3 or 4) prior to modeling.  The subjects were
22   more likely to have skin reactions to the active patches than to the negative control patches for
23   both whole cells (odds ratio (OR) = 2.13, 95% confidence interval (CI) 1.79-4.21, p<0.001) and
24   lysed cells (OR = 3.41, 95% CI 2.00-5.84, p<0.001).  The mean percentages of subjects having a
25   reaction were 20% (95% CI 15-31%) for all subjects (n=50) and 11% (95% CI 6-18%) for
26   subjects not reacting to negative controls (n=39).  The irritation was mild and resolved within 24
27   to 72 hours. There was no evidence of a statistically significant increasing dose-response
28   relationship between skin reactions and increasing cell concentrations for either whole or lysed
29   cells, although there was a slight reduction in response with increasing cell concentration for the
30   whole cells (OR = 0.966, 95% CI 0.936-0.997, p = 0.03). Additionally, there was no evidence
31   for a threshold effect (i.e., a particular concentration  above which there were frequent or strong
32   reactions).
33
34   4.2.    ACUTE, SHORT-TERM, SUBCHRONIC AND CHRONIC STUDIES AND
35          CANCER BIOASSAYS IN ANIMALS - ORAL AND INHALATION
36
37          Toxicity studies in animals have been performed using pure cylindrospermopsin isolated
38   and purified from cell extracts of C. raciborskii or other cylindrospermopsin-producing
39   cyanobacteria.  Studies have also been performed in which the administered material consisted of
40   whole cell extracts, lyophilized (freeze-dried) cells in suspension and cell-free extracts of
41   sonicated  freeze-dried cells.  These studies are included in this report because they contribute
42   salient information to the overall toxicological database for cylindrospermopsin. However, due
43   to confounding factors discussed below, the studies of cell extracts are not useful for dose-
44   response assessment of cylindrospermopsin and are considered supplemental information for
45   hazard identification.
46
                                                           DRAFT: DO NOT CITE OR QUOTE

-------
 1          Most of the cell extract studies were performed using laboratory cultures of
 2   cyanobacterial cells, but there is no clear means of predicting the cylindrospermopsin content in
 3   a particular extract.  Studies with cylindrospermopsin and other cyanobacterial toxins indicate
 4   that growth conditions can significantly contribute to the level of toxin produced by a given
 5   species and strain, and that toxin concentration can also vary depending on the method used to
 6   produce a material for toxicological testing (Chiswell et al., 1999; WHO, 1999). The
 7   extracellular fraction of cylindrospermopsin can sometimes exceed the intracellular fraction
 8   (Griffiths and Saker, 2003). For example, at different stages of a C. raciborskii bloom,
 9   extracellular cylindrospermopsin ranged from 19 to 98% of the  total amount in water (Chiswell
10   et al., 1999).  Similarly, during a bloom of A. ovalisporum, >85% of the cylindrospermopsin was
11   extracellular (Shaw et al., 1999). In these studies, intracellular concentration of
12   cylindrospermopsin was determined by taking the difference between the concentration in a
13   sample of filtered water and the concentration in a sample of water that was frozen to release the
14   toxin contained in the cells. Extracts obtained by removing intact cells may or may not contain
15   toxin or may have variable amounts of toxin. For example, Falconer et al. (1999) found that the
16   cylindrospermopsin content in four different batches of cell-free extracts of C. raciborskii varied
17   from  1.3 to 5.4 mg/g extract. Additionally, cell extracts containing cylindrospermopsin can also
18   contain other potentially toxic substances. The 24-hour i.p. LD50 (dose lethal to 50% of the
19   population) of purified cylindrospermopsin in male CH3 mice was  2.1 mg/kg (Ohtani et al.,
20   1992), whereas the value for a cell extract in male Swiss mice was 0.29 mg/kg (Hawkins et al.,
21   1997), nearly an order of magnitude lower.  Hawkins et al. (1997) proposed that the difference in
22   potency could reflect the presence of other toxins in the cell extract that were not present in the
23   purified cylindrospermopsin (see Section 4.4.1).
24
25   4.2.1.  Oral Exposure
26
27          4.2.1.1.  Acute Studies
28
29          4.2.1.1.1.  Studies of Purified Cylindrospermopsin
30
31          No information regarding the acute oral toxicity of purified cylindrospermopsin was
32   identified in the materials reviewed for this document.
33
34          4.2.1.1.2.  Cell Extract Studies
35
36          Twelve male MF1 mice were administered a saline suspension of freeze-dried C.
37   raciborskii cells (strains PHAWT/M or PHAWT/1) by gavage in single reported doses ranging
38   from 4.4 to 8.3 mg/kg (cylindrospermopsin-equivalent), and observed for the following 8 days
39   (Seawright et al., 1999). The following dose levels were tested  (one mouse per level except as
40   noted): 4.4, 5.3, 5.7 (two mice), 5.8, 6.2, 6.5, 6.7, 6.8, 6.9, 8.0 and 8.3 mg/kg; there was no
41   control group. Eight of the 12 mice died.  The lowest lethal dose was 4.4 mg/kg, the highest
42   nonlethal dose was 6.9 mg/kg and the average lethal dose was approximately 6 mg/kg. Deaths
43   occurred 2-6 days after treatment, and histological examinations showed effects that included
44   fatty liver with periacinar coagulative necrosis, acute renal tubular necrosis, atrophy of the
45   lymphoid tissue of the spleen and thymus, subepicardial and myocardial hemorrhages in the
                                                           DRAFT: DO NOT CITE OR QUOTE

-------
 1   heart and ulceration of the esophageal section of the gastric mucosa. Some of the animals also
 2   developed thrombohemorrhagic lesions in one or both eye orbits.
 3
 4          An aqueous suspension of a cell-free extract of freeze-dried and sonicated C. raciborskii
 5   cells (strain AWT 205) was administered to an unspecified number of male Swiss mice in a
 6   single gavage dose of 1400 mg extract/kg (Falconer et al., 1999). The cylindrospermopsin
 7   content of the extract was not specified, but ranged from 1.3 to 5.4 mg/g extract in concurrent i.p.
 8   experiments,  indicating that the cylindrospermopsin-equivalent gavage dose was likely in the
 9   range of 1.8-7.6 mg/kg.  This dose level was not fatal, but caused severe liver and kidney
10   pathology.  Histological changes were not detailed, but patterns of damage were reported to be
11   similar to those observed following i.p. administration (see Section 4.4.1). Additional
12   information on the design and results of the oral study were not provided.
13
14          Another gavage study reported that the minimum lethal dose of a saline extract of freeze-
15   dried C. raciborskii cells (strain AWT 205) in Swiss mice was 2500 mg extract/kg (Falconer and
16   Humpage, 2001). Based on a reported cylindrospermopsin content of 5.5 mg/g extract, the
17   equivalent dose of cylindrospermopsin was 13.8 mg/kg.
18
19          Groups of four Quackenbush mice were administered a cell-free extract of freeze-dried
20   and sonicated C. raciborskii cells (strain AWT 205) in water in a single gavage dose of 0, 1, 2, 4,
21   6 or 8 mg cylindrospermopsin/kg and observed for the following 7 days (Shaw et al., 2000,
22   2001).  All animals were evaluated for gross pathological and histological (liver, kidney, spleen,
23   heart, lungs and thymus) changes. Hepatic effects were observed at all dose levels, as shown by
24   foamy hepatocellular cytoplasmic changes  at 1 and 2 mg/kg, lipid infiltration with some
25   hepatocyte necrosis in the periacinar region at 4 mg/kg, and uniformly pale and mottled livers
26   with lipid infiltration throughout and cell necrosis mainly in the periacinar region at 6 mg/kg.
27   Mortality occurred in 2/4 mice at 6 mg/kg (in 5 days) and 4/4 mice  at 8 mg/kg (in 24-48 hours).
28   Additional information on the experimental design and results was not reported.
29
30          4.2.1.2.  Short-Term  Studies
31
32          4.2.1.2.1.  Studies of Purified Cylindrospermopsin
33
34          Groups of four Quackenbush mice were administered purified cylindrospermopsin by
35   daily gavage  for 14 days (Shaw et al., 2000, 2001).  The cylindrospermopsin was purified (purity
36   not reported) from an extract of freeze-dried C.  raciborskii cells (strain AWT 205). All animals
37   were evaluated for gross pathological and histological (liver, kidney, spleen, heart, lungs and
38   thymus) changes. The authors identified the following effect levels: a NOAEL of 0.05 mg
39   cylindrospermopsin/kg-day and a LOAEL of 0.15 mg cylindrospermopsin/kg-day for lipid
40   infiltration in the liver, and a NOAEL of 0.3 mg cylindrospermopsin/kg-day (highest tested dose)
41   for lymphophagocytosis in the spleen.  Additional information on the experimental design and
42   results was not reported.
43
                                                10         DRAFT: DO NOT CITE OR QUOTE

-------
 1          4.2.1.2.2.  Cell Extract Studies
 2
 3          Groups of four Quackenbush mice were administered an aqueous cell-free extract of
 4   freeze-dried and sonicated C. raciborskii cells (strain AWT 205) by daily gavage for 14 days
 5   (Shaw et al., 2000, 2001).  All animals were evaluated for gross pathological and histological
 6   (liver, kidney, spleen, heart, lungs and thymus) changes. The authors identified the following
 7   effect levels: a NOAEL of 0.05 mg cylindrospermopsin/kg-day and a LOAEL of 0.15 mg
 8   cylindrospermopsin/kg-day for lipid infiltration in the liver, and a LOAEL of 0.05 mg
 9   cylindrospermopsin/kg-day for lymphophagocytosis in the spleen. Additional information on the
10   experimental design and results was not reported.
11
12          4.2.1.2.3.  Other Studies
13
14          Six Quackenbush mice and two Wistar rats were exposed for 21 days to drinking water
15   containing 800 |J,g/L cylindrospermopsin (Shaw et al., 2000, 2001). The water was "sourced"
16   from a dammed impoundment. The reported approximate daily dose based on water
17   consumption was 0.2 mg cylindrospermopsin/kg-day in both species.  Gross pathological and
18   histological (liver, kidney, spleen, heart, lungs and thymus) examinations showed no effects,
19   indicating that 0.2 mg/kg-day was a NO AEL in the rats and mice.  Additional information on the
20   experimental design and results was not reported.
21
22          4.2.1.3. Subchronic Studies
23
24          4.2.1.3.1.  Studies of Purified Cylindrospermopsin
25
26          Groups of male Swiss albino mice (10 per dose, 6 in the highest dose group) were
27   administered purified cylindrospermopsin in water by gavage in doses of 0, 30, 60, 120 or 240
28   |j,g/kg-day for 11 weeks (Humpage and Falconer, 2003). The cylindrospermopsin was purified
29   (purity not reported) from an extract of freeze-dried C. raciborskii cells (strain AWT 205).
30   Endpoints monitored throughout the study included food and water consumption and body
31   weight. A clinical examination that focused on physiological and behavioral signs of toxicity
32   was conducted after 9 weeks of exposure. Hematology (all animals; red cell counts, hemoglobin,
33   packed cell volume, and white cell total and differential counts), serum chemistry (five
34   mice/group except all six mice at the high dose; total  protein, albumin, globulin, glucose,
35   creatinine, urea, total bilirubin, total bile acids, cholesterol, triglycerides, sodium, potassium,
36   calcium, bicarbonate, creatinine kinase, alanine and aspartate aminotransferases [ALT and AST,
37   respectively], and alkaline phosphatase) and urine (five mice/group excluding high dose; specific
38   gravity, protein, glucose, ketones, creatinine, sodium, potassium, chloride, calcium, bicarbonate,
39   phosphate, pH, volume and presence of blood) evaluations were performed near or at the end of
40   the treatment period.  Postmortem examinations included organ weights (liver, spleen, kidneys,
41   adrenal glands, heart, testis, epididymis and brain) and comprehensive histological evaluations.
42   The histological examinations were conducted in accordance with Organization for Economic
43   Cooperation and Development recommendations and performed on the following tissues: liver,
44   kidney, heart, lungs, thymus, thyroid, trachea, salivary glands, adrenal glands, epididymis, testis,
45   prostate, gall bladder, esophagus, stomach, duodenum/small intestine, large intestine, pancreas,
                                                11         DRAFT: DO NOT CITE OR QUOTE

-------
 1   spleen, urinary bladder, eyes, lymph nodes, aorta, cerebrum, cerebellum, spinal cord (cervical,
 2   thoracic and lumbar) and peripheral nerve.
 3
 4          No deaths were reported. The mean final body weight was 7-15% higher than controls in
 5   all dose groups, but the increases were not dose-related and were statistically significant only at
 6   30 and 60 |j,g/kg-day (Humpage and Falconer, 2003). There were no significant changes in food
 7   consumption; however, water intake was significantly reduced in all dose groups (data not
 8   reported).  Relative kidney weight was increased in a significant, dose-related manner beginning
 9   at 60 |j,g/kg-day (12-23% greater than controls), while relative liver weight was significantly
10   increased only at the high dose of 240 |j,g/kg-day (13% greater than controls). Information on
11   absolute kidney and liver weights was not reported.  Absolute testis weights were significantly
12   increased at >60 |j,g/kg-day (data not reported), but these changes were not significant when
13   normalized to body weight.  The hematology, serum chemistry and urine evaluations showed no
14   clear exposure-related changes in any endpoint (including serum indicators of liver injury),
15   except for significant decreases in urine protein concentrations (g/mmol creatinine) at >120
16   |j,g/kg-day and urine specific gravity at 240 |j,g/kg-day (data presented graphically).  The
17   postmortem examinations showed "minor increases in histopathological damage to the liver" at
18   >120 |j,g/kg-day and proximal renal tubular damage at 240 |j,g/kg-day, but additional information
19   regarding the type, severity and incidences of the liver and kidney lesions was not reported.
20
21          Cylindrospermopsin is known to inhibit protein synthesis in the liver (see Section 4.5.1).
22   Serum albumin, a major product of liver protein synthesis, was not decreased in this study
23   (Humpage and Falconer, 2003), but the most sensitive effects, decreased urinary protein at >120
24   |j,g/kg-day and increased relative kidney weight at >60 |j,g/kg-day, are both potential indicators of
25   suppressed protein synthesis. As hypothesized by the authors, the decrease in urinary protein is
26   consistent with decreased availability of protein and the increase in kidney weight may reflect a
27   compensatory hyperplasia, such that the kidney, as a protein-synthesizing organ, is stimulated to
28   grow in an attempt to maintain homeostasis in response to a chemically-induced decrease in
29   protein synthesis. Information supporting the hypothesis that the decrease in urinary protein
30   excretion reflects a  specific effect of Cylindrospermopsin on protein synthesis, as well as the
31   possibility that it reflects a functional change in the nephron, is discussed in Section  4.5.2.
32   Because the renal effects observed  by Humpage and Falconer (2003) are consistent with a known
33   mode of action of Cylindrospermopsin, and plausibly represent part of the progression of effects
34   leading to toxicity (Section 4.5.2), they are considered to be adverse. This study, therefore,
35   identifies aNOAEL and LOAEL of 30 and 60 |j,g/kg-day, respectively.
36
37          4.2.1.3.2.  Cell Extract Studies
38
39          Groups of male Swiss albino mice (10 per dose except 12 controls and 5 at high-dose)
40   were exposed to a cell-free extract  of sonicated and frozen C. raciborskii cells (strain AWT 205)
41   in the drinking water at reported Cylindrospermopsin doses of 0, 216, 432 or 657 |j,g/kg-day for
42   10 weeks (doses based on actual water consumption) (Humpage and Falconer, 2003). Food and
43   water consumption  and body weight were measured throughout the study. Urinalyses (12
44   unspecified parameters) were performed after 5 and  10 weeks. Serum chemistry (15 unspecified
45   parameters) evaluations and examinations of unspecified major organs (organ weight, gross
                                                 12        DRAFT: DO NOT CITE OR QUOTE

-------
 1   pathology and histopathology) were performed at the end of the exposure period.  Hematology
 2   was not evaluated.
 3
 4          Final body weights were significantly reduced at 432 and 657 |j,g/kg-day (9 and 7% less
 5   than controls, respectively), and relative liver and kidney weights were significantly increased in
 6   a dose-related manner at 216-657 |j,g/kg-day (27-47 and 30-43% greater than controls,
 7   respectively).  Other statistically significant effects included increased serum total bilirubin at
 8   >216 |j,g/kg-day, decreased serum total bile acids at >216 |j,g/kg-day and decreased urine protein
 9   concentration (g/mmol creatinine) at >432 |j,g/kg-day. There were no clear exposure-related
10   changes in any other serum or urine endpoints and no  additional indicators of liver or kidney
11   injury. Results of the postmortem pathology examinations were not reported. The low dose of
12   216 |j,g/kg-day is a LOAEL for this study, based on increased relative liver and kidney weights,
13   increased serum bilirubin and decreased serum bile  acids. An increase in serum bilirubin is
14   indicative of liver dysfunction or bile duct blockage as it reflects the ability of the liver to take
15   up, process and secrete bilirubin into the bile.  Serum bile acids can be decreased due to an
16   inhibition of bile acid synthesis or an interference with bile  acid resorption in the gastrointestinal
17   tract; bile acids are synthesized from cholesterol in the liver, conjugated, excreted in the bile and
18   resorbed in the ileum.
19
20          Quackenbush mice were administered drinking water containing a cell-free extract of
21   freeze-dried and sonicated C. raciborskii cells (strain AWT 205) for 90 days (Shaw et al., 2000,
22   2001). Gross pathological and histological (liver, kidney, spleen, heart, lungs and thymus)
23   examinations showed no effects at dose levels as high as 0.15 mg cylindrospermopsin/kg-day
24   (the highest tested dose), indicating that a NOAEL of 0.15 mg/kg-day was identified. Additional
25   information on the experimental design and results was not  reported. The 0.15 mg/kg-day
26   NOAEL in Quackenbush mice is only slightly below the 216 |j,g/kg-day (0.22 mg/kg-day)
27   LOAEL for liver and kidney effects in the 10-week study with  Swiss mice summarized above
28   (Humpage and Falconer, 2003); however, the LOAEL is based on different measured endpoints
29   (liver and kidney weights, serum bilirubin and serum bile acids) than the NOAEL
30   (histopathology).
31
32          4.2.1.4. Chronic Studies
33
34          No information regarding the chronic oral toxicity of cylindrospermopsin was identified
35   in the materials reviewed for this document.
36
37   4.2.2.  Inhalation Exposure
38
39          No information regarding the inhalation toxicity of cylindrospermopsin was identified in
40   the materials reviewed for this document.
41
42   4.3.   REPRODUCTIVE/DEVELOPMENTAL STUDIES -  ORAL AND INHALATION
43
44          No information regarding the reproductive or developmental toxicity of
45   cylindrospermopsin was identified in the materials reviewed for this document.
46
                                                 13         DRAFT: DO NOT CITE OR QUOTE

-------
 1   4.4.   OTHER STUDIES
 2
 3   4.4.1.  Effects By Parenteral Exposure.
 4
 5          4.4.1.1. Studies of Purified Cylindrospermopsin
 6
 7          Acute lethality values have been determined for Cylindrospermopsin purified from
 8   extracts of cultured C. raciborskii or U. natans cells (Ohtani et al., 1992; Shaw et al., 2000,
 9   2001; Terao et al., 1994).  In male CHS mice, 24-hour and 5- to 6-day LD50 values of 2.1 and 0.2
10   mg/kg, respectively, have been reported for a single i.p. dose of purified Cylindrospermopsin
11   (purity not reported) (Ohtani et al., 1992).  Another study found that a single 0.2 mg/kg i.p. dose
12   of purified Cylindrospermopsin (purity not reported) caused 50% moribundity after 31 hours in
13   Quackenbush mice (Shaw et al., 2000, 2001). The main pathological findings in the moribund
14   animals were lipid infiltration and cell necrosis in the liver. Terao et al. (1994) also found that
15   the liver was the main target of toxicity in male ICR mice  administered a single 0.2 mg/kg i.p.
16   dose of purified Cylindrospermopsin (purity not reported),  although treatment-related lesions
17   were additionally noted in the thymus, kidney and heart. A time series of ultrastructural tissue
18   examinations indicated four sequential phases of liver changes: inhibition of protein  synthesis,
19   membrane proliferation, fat droplet accumulation and cell  death.
20
21          4.4.1.2. Cell Extract Studies
22
23          The results of acute i.p. studies of extracts of freeze-dried and sonicated C. raciborskii
24   cells are generally similar to those of the i.p.  studies of purified Cylindrospermopsin.  A single
25   0.2 mg/kg cylindrospermopsin-equivalent dose caused  50% moribundity in Quackenbush mice
26   after 98 hours (Shaw et al., 2000, 2001). Other single-dose LD50 values, expressed as
27   cylindrospermopsin-equivalent doses, included 24-hour and 7-day values of 0.29 and 0.18
28   mg/kg, respectively, in male Swiss mice (Hawkins et al., 1997).  This 24-hour LDso was lower
29   than the 24-hour i.p. LD50 of 2.1 mg/kg for purified Cylindrospermopsin in mice (Ohtani et al.,
30   1992), leading the authors to suggest that the extract contained more than one toxin.  The liver
31   was the main target organ in the extract studies, although lesions also occurred in other tissues,
32   including kidney, adrenal gland, lung and intestine (Hawkins et al., 1985, 1997; Shaw et al.,
33   2000,2001).
34
35          A single dose i.p. LDso value of 64 mg freeze-dried culture/kg was determined in mice
36   observed for 24 hours (Hawkins et al., 1985). Falconer et  al. (1999) assessed the acute lethality
37   and liver and kidney effects of four different  batches of cell-free extracts of sonicated freeze-
38   dried C. raciborskii cells in male Swiss albino mice treated by single i.p. injection. Reported
39   24-hour and 7-day LDso values for the four batches were 50-110 and 20-65 mg extract/kg,
40   respectively. The Cylindrospermopsin content in the four batches varied from 1.3 to  5.4 mg/g
41   extract, indicating that the cylindrospermopsin-equivalent  LD50 values were 0.07-0.6 mg/kg
42   (24-hour) and 0.03-0.4 mg/kg (7-day). Liver damage was  characterized by cellular vacuolation,
43   intercellular spaces and darker nuclear and cytoplasmic staining. Kidney damage included
44   proximal tubule epithelial necrosis and presence of proteinaceous material  in the distal tubules.
45   There was no clear correlation between Cylindrospermopsin batch concentration and the LD50
                                                 14        DRAFT: DO NOT CITE OR QUOTE

-------
 1   values or severity of liver or kidney lesions, leading the study authors to suggest that more than
 2   one toxin was present in the extract.
 3
 4   4.4.2.  Immunotoxicity
 5
 6          No information was located regarding effects of cylindrospermopsin on immune
 7   function, although immune system tissues appear to be a target of short-term, high-level
 8   exposures. Massive necrosis of lymphocytes occurred in the cortical layer of the thymus of male
 9   ICR mice given a single 0.2 mg/kg i.p. dose of cylindrospermopsin purified (purity not reported)
10   from cultured U. natans cells (Terao et al., 1994). Effects observed in MF1 mice administered a
11   single gavage dose of a suspension of freeze-dried C. raciborskii cells, in the lethal dose range of
12   4.4-8.3 mg cylindrospermopsin/kg, included atrophy in lymphoid tissue of the spleen (follicular
13   lymphocyte loss due to lymphophagocytosis) and thymus (degeneration and necrosis of cortical
14   lymphocytes) (Seawright et al., 1999). These effects were considered by the study authors to be
15   normal responses of the immune system to the stress of severe intoxication. Lympho-
16   phagocytosis was observed in the spleen of Quackenbush mice exposed to a cell-free extract of
17   freeze-dried and sonicated C. raciborskii cells by gavage at a nonlethal dose level of 0.05 mg
18   cylindrospermopsin/kg-day for 14 days (Shaw et al., 2000, 2001).
19
20   4.4.3.  Tumor Initiation
21
22          The tumor initiating activity of cylindrospermopsin was tested in male Swiss mice using
23   O-tetradecanoylphorbol 13-acetate (TPA) as the promoter (Falconer and Humpage, 2001).  Mice
24   were administered a gavage dose of saline (27 mice) or 500 mg/kg of a saline extract of fireeze-
25   dried C. raciborskii cells (strain AWT 205) (34 mice) every other week for three doses.  Other
26   groups received a single dose of 1500 mg extract/kg (14  mice) or two doses of 1500 mg
27   extract/kg separated by 2 weeks (17 mice). Most (70%)  of the 2 x 1500 mg extract/kg group
28   died within 1 week of the second dose, leaving five survivors for use in the rest of the study.
29   Based  on a reported cylindrospermopsin content  of 5.5 mg/g extract, the cylindrospermopsin-
30   equivalent doses in the 500 and 1500 mg extract/kg groups were 2.75 and 8.25 mg/kg,
31   respectively.  Two weeks after the final dose, the saline and  500 mg extract/kg groups were
32   divided into subgroups of 13-18 mice that were fed liquid food containing TPA dissolved in
33   DMSO, or food containing DMSO alone, for 24 hours twice weekly for 30 weeks.  All of the
34   mice in both  1500 mg extract/kg groups were similarly exposed to TPA-containing liquid food
35   (no 1500 mg/kg mice were exposed to food containing DMSO alone). Histological examinations
36   of the liver, kidneys, spleen and grossly abnormal organs were performed on all groups at the
37   end of the 30-week promotion period.  Neoplastic changes were found in none of the 27 control
38   mice and in a total of 5 cylindrospermopsin-treated mice, a difference that was not statistically
39   significant. There was no pattern to the neoplasic changes, as they occurred in different animals,
40   target organs and treatment groups, as detailed in Table 4-1.
                                                15        DRAFT: DO NOT CITE OR QUOTE

-------
Table 4-1. Tumor Initiating Activity of C. raciborskii Extracts
Oral Treatment (mg extract/kg)
Saline/DMSO
Saline/TPA
3 x 500/DMSO
3 x 500/TPA
1 x 1 500/TPA
2xl500/TPA
Number of Mice
14
13
18
16
14
5
Histological Finding*
No neoplasia observed
No neoplasia observed
1 hepatocellular carcinoma
1 lymphoma
No neoplasia observed
2 hepatocellular dysplastic foci
1 fibroblastic osteosarcoma
No neoplasia observed
 2
 o
 J
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
* All findings were in different animals
Source: Falconer and Humpage (2001)
4.4.4. Genotoxicity

       Purified cylindrospermopsin caused an increase in the frequency of micronuclei in the
human lymphoblastoid cell line WIL2-NS (Humpage et al., 2000).  Both centromere-positive
and centromere-negative micronuclei were induced, suggesting that whole chromosome loss, as
well as DNA strand breaks, contributed to the in vitro cytogenetic damage. DNA strand
breakage was also observed in the liver of Balb/c mice following a  single 0.2 mg/kg i.p. dose of
purified cylindrospermopsin (Shen et al., 2002). Covalent binding  of cylindrospermopsin or a
metabolite to DNA (adduct not identified) was detected in the liver of Quackenbush mice given a
single i.p. injection of a cell-free extract of C. raciborskii (dose levels not reported) (Shaw et al.,
2000).  Purified cylindrospermopsin caused cell growth inhibition and altered cell morphology,
but no apoptosis or DNA strand breaks, in Chinese hamster ovary Kl cells in vitro (Fessard and
Bernard, 2003).

4.5.    MECHANISTIC DATA AND OTHER STUDIES IN SUPPORT OF THE MODE
       OF ACTION

4.5.1. Liver Toxicity

       The liver is widely regarded as the main target of cylindrospermopsin toxicity, and
consequently, most mechanistic studies have assessed hepatic endpoints. The specific
mechanism for the liver toxicity is not clearly understood, although it has generally been
considered to involve cylindrospermopsin-induced inhibition of protein synthesis.
                                                16
                                                     DRAFT: DO NOT CITE OR QUOTE

-------
 1   Cylindrospermopsin was shown to be a potent inhibitor of protein synthesis in an in vitro rabbit
 2   reticulocyte globin synthesis assay (Terao et al., 1994). Ultrastructural liver changes in mice
 3   treated with a single 0.2 mg/kg i.p. dose of purified Cylindrospermopsin had features in common
 4   with those dosed with the protein synthesis inhibitor cycloheximide, particularly detachment of
 5   ribosomes from the rough endoplasmic reticulum, suggesting that protein synthesis inhibition
 6   plays a role in Cylindrospermopsin hepatotoxicity in vivo (Terao et al., 1994). However, unlike
 7   the liver in the cycloheximide-dosed mice, the liver of those treated with Cylindrospermopsin
 8   showed membrane proliferation, fat droplet accumulation and reduced amount of total P450 in
 9   microsomes, indicating that mechanisms other than protein synthesis inhibition must also
10   contribute to Cylindrospermopsin toxicity.
11
12          Cylindrospermopsin-induced depletion of mouse hepatic glutathione was demonstrated in
13   vivo (Norris et al., 2002), although the study authors did not consider the effect to be of sufficient
14   magnitude to represent the primary mechanism of Cylindrospermopsin toxicity.
15   Cylindrospermopsin also caused decreased glutathione levels, as well as decreased synthesis of
16   glutathione and protein,  in cultured rat hepatocytes (Runnegar et al., 1994, 1995, 2002).
17   Inhibition of glutathione synthesis was the predominant mechanism for the reduction in
18   glutathione; other mechanisms, including increased consumption of glutathione, increased
19   formation of oxidized glutathione, increased glutathione efflux, hidden forms of glutathione,
20   decreased glutathione precursor availability and decreased cellular ATP were effectively ruled
21   out (Runnegar et al., 1995). Glutathione depletion occurred at non-toxic Cylindrospermopsin
22   concentrations and preceded the onset of observable toxicity at higher concentrations (Runnegar
23   et al., 1994). Pretreatment with the CYP450 inhibitor, a-naphthoflavone, partially protected
24   against cytotoxicity and cellular glutathione depletion, indicating involvement of the CYP450
25   enzyme system in Cylindrospermopsin metabolism and that one or more metabolites might be
26   more active than the parent compound in inhibiting glutathione synthesis (Runnegar et al.,  1995).
27   In vitro studies in mouse hepatocytes provided no indication that reductions in glutathione  levels
28   by Cylindrospermopsin led to increased levels of reactive oxygen species (ROS) (Humpage et  al.,
29   2005).
30
31          Cylindrospermopsin induced time- and concentration-dependent toxicity and inhibition of
32   protein synthesis in hepatocytes isolated from male Swiss mice (Froscio et al., 2003).  The
33   broad-spectrum CYP450 inhibitors proadifen (SKF525A) and ketoconazole diminished the
34   induction of cytotoxicity by Cylindrospermopsin, but did not diminish the inhibition of protein
35   synthesis. These findings suggest that the cytotoxic effects of Cylindrospermopsin might be
36   linked more to CYP450-mediated bioactivation than to inhibition of protein synthesis by the
37   parent compound.  Similarly, pretreatment of male Quackenbush mice with the broad-spectrum
38   CYP450 inhibitor piperonyl butoxide protected against the acute lethality of Cylindrospermopsin
39   (Norris et al., 2002). In  a study using inhibitors of specific CYP450 isoforms, furafylline
40   (CYP1A2) and omeprazole (CYP3A4 and CYP2C19) protected against Cylindrospermopsin
41   cytotoxicity in an in vitro mouse hepatocyte system; unspecified inhibitors of CYPs 2A6, 2D6
42   and 2E1 were not found to be cytoprotective (Humpage et al., 2005). Additional support for the
43   involvement of CYP450 in the hepatotoxicity of Cylindrospermopsin is the finding that liver
44   histopathology is mainly induced in the region (periacinar) where CYP450-catalyzed xenobiotic
45   metabolism occurs (Shaw et al., 2000, 2001).
46
                                                17         DRAFT: DO NOT CITE OR QUOTE

-------
 1   4.5.2.  Kidney Toxicity
 2
 3          No studies were located that specifically investigated the involvement of protein
 4   synthesis inhibition or other modes of action in cylindrospermopsin-induced toxicity in the
 5   kidney or other non-hepatic target tissues (e.g., spleen and thymus). As detailed in Section
 6   4.2.1.3.1, the kidney was the most sensitive target in mice that were exposed to
 7   cylindrospermopsin by daily gavage for 11 weeks (Humpage and Falconer, 2003). Renal effects
 8   in the mice included increased relative kidney weight at >60 |j,g/kg-day, decreased urinary
 9   protein at >120 |j,g/kg-day and decreased urine specific gravity and proximal renal tubular
10   lesions at 240 |j,g/kg-day.  The authors hypothesized that the decrease in urinary protein is
11   consistent with decreased availability of protein and that the increase in kidney weight may
12   reflect a  compensatory hyperplasia, such that the kidney, as a protein-synthesizing organ, is
13   stimulated to grow in  an attempt to maintain homeostasis in response to a cylindrospermopsin-
14   related decrease in protein synthesis. Information supporting the hypothesis that the decrease in
15   urinary protein excretion reflects a specific effect of cylindrospermopsin on protein synthesis, as
16   well as the possibility that it reflects a functional change in the nephron, is discussed below.
17   Also discussed is evidence suggesting a dose-severity progression of kidney effects.
18
19          Potential mechanisms for a decrease in urinary protein  include a decrease in glomerular
20   filtration (i.e., filtered load) of protein, an increase in resorption of filtered protein  and a decrease
21   in secretion of nephrogenic protein.  A decrease in glomerular  filtration of protein  (e.g., |j,g
22   protein/day) could result from a decrease in serum protein concentration or a decrease in
23   glomerular filtration rate (mL/day, GFR). The predominant serum protein in urine of healthy
24   animals (e.g., mice, rats and humans) is albumin (-50% of serum proteins in urine).  In the
25   Humpage and Falconer (2003) study, serum albumin concentration increased in mice exposed to
26   cylindrospermopsin, and serum creatinine (a marker of GFR) apparently was unchanged; it was
27   measured but not discussed in the results. Therefore, it is unlikely that glomerular filtration of
28   serum proteins decreased in response to cylindrospermopsin (if a change occurred, it is likely to
29   have been an increase in the rate of filtration of albumin).  Furthermore, serum proteins normally
30   account for approximately  15% of total urinary protein (Pesce  and First, 1979).  The decrease in
31   urinary excretion of protein observed in Humpage and Falconer (2003) was substantially larger
32   than this (-50%), indicating that the decrease in urinary protein cannot derive solely from a
33   decrease in excretion (i.e., glomerular filtration) of serum proteins.
34
35          No information is presented in Humage and Falconer (2003) that would allow an
36   assessment of tubular resorption of filtered protein (e.g., plasma-to-urine clearance of protein,
37   excretion of low-molecular weight proteins such as  p2|J,globulin or retinal binding protein).
38
39          In healthy mammals, the dominant protein in urine (-50%) is the nephrogenic Tamm-
40   Horsfall  protein (THP, uromucoid) (Bachmann et al., 1991, 2005).  In the absence of a decrease
41   in filtration or increased resorption of filtered serum protein, the substantial decrease in urinary
42   protein (i.e., -50%) observed by Humpage and Falconer (2003) would almost certainly have to
43   involve decreased excretion of THP, since it is the predominant protein in urine. Although there
44   are numerous possible mechanisms for an acute change  in THP excretion (Bachman et al., 1991),
45   long-term maintenance of lower (i.e., steady-state) rate of urinary excretion of THP requires a
46   decreased rate of synthesis of THP (Bachman et al.,  1991, 2005; Schoel and Pfleiderer, 1987).
                                                 18        DRAFT: DO NOT CITE OR QUOTE

-------
 1   THP is synthesized exclusively in the thick ascending limb of the loop of Henle (TAL);
 2   therefore, a sustained change in THP excretion is likely to reflect a functional change in this
 3   region of the nephron. Increases  and decreases in THP have been observed in various kidney
 4   diseases, and in association with experimental treatments that induce hypertrophy of the TAL,
 5   including increased dietary protein (Bachmann et al., 1991). Depletion of THP from the kidney
 6   may, in itself, be adverse.  Mice deficient in THP (i.e., THP knockout mice) display impaired
 7   urine concentrating ability, up-regulation of distal nephron electrolyte transport proteins and
 8   increased susceptibility to urinary tract infections (Bachmann et al., 2005; Bates et al., 2004).
 9   The decrease in urine specific gravity in animals exposed to cylindrospermopsin in the Humpage
10   and Falconer (2003) study may be indicative of impaired urine concentrating ability and,
11   possibly, related to impaired function of the TAL (i.e., impairment of transport activity in this
12   region of the nephron impairs urine concentrating ability) and/or decreased  synthesis of THP.
13
14          Additional kidney effects  in the Humpage and Falconer (2003) mouse study included
15   proximal renal tubular damage (type and severity of lesions not reported) at the high dose.
16   Clinical effects in the Palm Island outbreak in which humans apparently ingested drinking water
17   containing elevated levels of cylindrospermopsin included renal damage, as indicated by loss of
18   water, electrolytes, proteins, ketones and carbohydrates (Blyth,  1980; Griffiths and Saker, 2003)
19   (Section 4.1). Proteinuria would  be expected with proximal tubular damage, as this is the site of
20   resorption of filtered protein.  Proteinuria was not observed by Humpage and Falconer (2003),
21   but information on the type and severity of the tubular damage was not reported. Proteinuria did
22   occur in the humans, although other mechanisms could have caused it (e.g., glomerular injury
23   will produce high molecular weight proteinuria). The evidence for proximal tubular damage and
24   functional impairment (e.g., proteinuria, glucosuria) together strengthen the argument that the
25   kidney is a target of cylindrospermopsin and, when considered with decreased protein excretion
26   at lower doses, suggests a dose-severity progression.
27
28   4.5.3. Interactions with DNA and RNA
29
30          Based on structural characteristics of cylindrospermopsin (its nucleoside structure and
31   potentially reactive guanidine and sulfate groups), it has been speculated that cylindrospermopsin
32   may exert its toxic effects via pathways that include reactions with DNA and/or RNA (see
33   Humpage et al., 2000; Shen et al., 2002). Covalent binding between DNA and
34   cylindrospermopsin, or a metabolite, occurred in mouse liver in vivo (Shaw et al., 2000). DNA
35   adducts were detected, but not identified, using the 32P-postlabeling assay; this involved
36   extraction of the DNA, hydrolysis into individual nucleotides, labeling of the nucleotides using
37   32P-ATP,  separation of adducted nucleotides using two-dimensional thin layer chromatography
38   and visualization of adduct spots  by autoradiography.  Cylindrospermopsin also induced DNA
39   strand breakage in mouse liver in vivo (Shen et al., 2002) and increases in micronuclei occurred
40   in treated binucleated cells of the WIL2-NS lymphoblastoid cell-line (Humpage et al., 2000).
41   Two mechanisms were suggested for causing the cytogenetic damage: one at the level of DNA to
42   induce strand breaks and the other at the level of kinetochore/spindle function to induce loss of
43   whole chromosomes (Humpage et al., 2000; Shen et al., 2002).  The broad-spectrum CYP450
44   inhibitors omeprazole and SKF525A inhibited cylindrospermopsin-induced DNA damage in
45   primary cultured mouse hepatocytes at subcytotoxic concentrations, suggesting that CYP-derived
                                                19         DRAFT: DO NOT CITE OR QUOTE

-------
 1   metabolites are responsible for cylindrospermopsin genotoxicity and that genotoxicity is a
 2   primary effect of the chemical (Humpage et al., 2005).
 3
 4          Cylindrospermopsin-induced up-regulation of the tissue transglutaminase (tTGase) gene
 5   was demonstrated in liver RNA of Balb/c mice following i.p. injection of a single 100 ng/kg
 6   dose of cylindrospermopsin (Shen et al., 2003). tTGase is a unique member of the TGase (EC
 7   2.3.2.13) family that catalyzes the post-translational modification of proteins via Ca2+-dependent
 8   cross-linking reactions (Shen et al., 2003). The up-regulation of tTGase can lead to liver injury
 9   (Grenard et al., 2001; Mirza et al., 1997), and has been implicated in diverse biological
10   processes, such as induction of apoptosis (Piacentini et al., 2002; Zhang et al., 1995), cell death
11   and differentiation (Shen et al., 2003; Fesus et al., 1987) and adhesion and morphological
12   changes of cells (Shen et al., 2003; Akimov and Belkin, 2001).
13
14   4.5.4.  Structure-Activity Relationships
15
16          Natural cylindrospermopsin, synthetic (racemic) cylindrospermopsin and  selected
17   synthetically-produced cylindrospermopsin structural analogues were assessed for effects on
18   protein synthesis in both the rabbit reticulocyte lysate system and cultured  rat hepatocytes
19   (Runnegar et al., 2002). No significant differences were observed in levels of protein synthesis
20   inhibition elicited by natural cylindrospermopsin and its diol analogue, indicating that the sulfate
21   group might not be a necessary component of cylindrospermopsin-induced protein synthesis
22   inhibition.  Additionally, the orientation of the hydroxyl group at C7 in the carbon bridge does
23   not appear to be important, since the C7 epimer of cylindrospermopsin and its corresponding diol
24   exhibited protein synthesis inhibition similar to that elicited by synthetic (racemic)
25   cylindrospermopsin. The cyclopentyl ring and the methyl and hydroxyl groups on the adjacent
26   hexyl ring may be important structural features, because the analogue lacking these features was
27   500-1000-fold less effective in the inhibition of protein synthesis.
28
29          The uracil portion of cylindrospermopsin appears to play an important role in
30   cylindrospermopsin toxicity.  Banker et al. (2001) found that the acute lethality of
31   cylindrospermopsin to mice was eliminated by chlorination or partial cleavage of the uracil
32   moiety (resulting in 5-chloro-cylindrospermopsin and cylindrospermic acid, respectively), as
33   shown by a 5-day i.p. LD50 value of 0.2 mg/kg for cylindrospermopsin and 10-day i.p. LD50
34   values of >10 mg/kg for 5-chloro-cylindrospermopsin and >10 mg/kg for cylindrospermic acid.
35
36          Deoxycylindrospermopsin, an analogue of cylindrospermopsin isolated and purified from
37   C. raciborskii, was tested for  toxicity in male white Quackenbush mice treated by i.p. injection
38   (Norris et al., 1999). Deoxycylindrospermopsin did not appear to be toxic during 5 days
39   following administration of a 0.8 mg/kg dose, whereas Ohtani et al. (1992) reported a 5- to 6-day
40   i.p. LDso value of 0.2 mg/kg for cylindrospermopsin in male CD3 mice. Although this
41   comparison suggests that deoxycylindrospermopsin is significantly less toxic than
42   cylindrospermopsin, differences in study designs (e.g., the use of different strains of mice) could
43   have contributed to the difference in toxicity.
44
                                                 20        DRAFT: DO NOT CITE OR QUOTE

-------
 1   4.6.    SYNTHESIS AND EVALUATION OF MAJOR NONCANCER EFFECTS
 2
 3   4.6.1.  Oral
 4
 5          Information on the health effects of cylindrospermopsin in humans is limited to
 6   observations on the Australian Palm Island poisoning incident that involved acute and/or short-
 7   term drinking water exposure to C. raciborskii, a non-infectious cyanobacterium (Blyth, 1980;
 8   Griffiths and Saker, 2003).  The clinical picture of the illness is well-defined and includes fever,
 9   headache, vomiting, bloody diarrhea, hepatomegaly and kidney damage with loss of water,
10   electrolytes and protein, but no data are available on exposure levels of cylindrospermopsin that
11   induced these effects.
12
13          The preponderance of information on noncancer effects of cylindrospermopsin in animals
14   is available from oral and i.p. administration studies in mice that were exposed to purified
15   compound or extracts of C. raciborskii cells. These studies indicate that the liver and kidneys
16   are main targets of toxicity and that cylindrospermopsin also causes significant lesions in other
17   organs, particularly the spleen and thymus. Considering both animal and human kidney data, the
18   evidence suggests a dose-severity progression of renal effects ranging from decreased protein
19   synthesis at low doses to functional impairment at high doses. The cell extract studies provide
20   limited dose-response information for cylindrospermopsin because concentrations vary between
21   cultures and strains and, in some cases, may contain other toxins, as discussed in the introduction
22   to Section 4.2. The available oral toxicity studies  of purified cylindrospermopsin are
23   summarized in Table 4-2.
24
25          No studies have been performed assessing the acute oral toxicity of purified
26   cylindrospermopsin.  Studies in which mice were administered single gavage doses of
27   suspensions or cell-free extracts of C. raciborskii cells at near-lethal to lethal levels found severe
28   damage to the liver (fatty and necrotic changes), kidneys (acute tubular necrosis), spleen and
29   thymus (atrophy of lymphoid tissue), heart (hemorrhages) and gastric mucosa (ulceration of the
30   esophageal section) (Falconer et al., 1999; Seawright et al., 1999; Shaw et al., 2000, 2001).
31
32          A limited amount of information on the short-term oral toxicity of cylindrospermopsin is
33   available from inadequately reported 14- and 21-day studies.
34
35          Histological examinations of small numbers of mice that were administered daily gavage
36   doses of purified cylindrospermopsin for 14  days showed effects in the liver (fatty infiltration)
37   and spleen (lymphophagocytosis) (Shaw et al., 2000, 2001).  Fatty infiltration in the liver was the
38   more sensitive effect based on a reported NOAEL of 0.05 mg/kg-day and LOAEL of 0.15
39   mg/kg-day. Small numbers of mice and rats were exposed to cylindrospermopsin for 21 days in
40   drinking water from a dammed impoundment at a reported approximate dose of 0.2 mg/kg-day
41   (Shaw et al., 2000, 2001). No histopathological changes were noted, indicating a NOAEL of 0.2
42   mg/kg-day in drinking water. The adequacy of the 14- and 21-day effect levels cannot be
43   assessed due to a lack of any additional reported information on the design and results of these
44   studies.
45
                                                21         DRAFT: DO NOT CITE OR QUOTE

-------
Table 4-2. Summary Results of Oral Toxicity Studies of Pure Cylindrospermopsin in Experimental Animals*
Species
Sex
Average
Daily Dose
(mg/kg-day)
Exposure
Duration
NOAEL
(mg/kg-day)
LOAEL
(mg/kg-day)
Responses
Comments
Reference
Acute Exposure
No suitable acute studies are available
Short-Term Exposure
Mouse
NR
0.05,0.15,
0.3
(gavage)
14 days
0.05
0.15
Lipid infiltration in liver.
Low confidence in NOAEL and
LOAEL. A full report of this study
has not been published; this table
provides essentially all available
information on experimental design
and results.
Shaw et al.,
2000, 2001
Subchronic Exposure
Mouse
M
0, 0.03,
0.06,0.12,
0.24
(drinking
water)
1 1 weeks
0.03
0.06
Increased relative kidney
weight with decreased
urinary protein at >0. 12
mg/kg-day.
Well-designed study with endpoints
that included food and water
consumption, body weight, clinical
signs, hematology, serum chemistry,
urinalysis, organ weights (eight
organs) and histology
(comprehensive). Ten mice/level
(six in high dose group).
Humpage and
Falconer, 2003
Chronic Exposure
No suitable chronic studies are available.
* Oral studies using suspensions or cell-free extracts of C. raciborskii cells are discussed in Section 4.2.1.
NR = Not reported
                                                                   22
DRAFT: DO NOT CITE OR QUOTE

-------
 1          A comprehensive subchronic toxicity study was conducted in which mice were exposed
 2   to five dose levels of purified cylindrospermopsin (0, 30, 60, 120 or 240 ng/kg) by daily gavage
 3   for 11 weeks (Humpage and Falconer, 2003). Histopathological effects were observed in the
 4   liver at >120 |j,g/kg-day ("minor increases in histopathological damage") and kidneys at 240
 5   |j,g/kg-day (proximal tubular damage), but no other information on the lesions, including
 6   incidence data, was reported. There were no changes in liver weight at doses below 240
 7   |j,g/kg-day or serum indices of liver damage (e.g., serum ALT, AST and alkaline phosphatase) in
 8   any of the dose groups.  Relative kidney weight was increased at >60  |j,g/kg-day and urine
 9   protein was decreased at >120 |j,g/kg-day. These effects are considered to be adverse because
10   they are consistent with a known mode of action of cylindrospermopsin (inhibition of protein
11   synthesis) and represent part of the spectrum of effects leading to toxicity, as discussed  in
12   Section 4.5.2. Based on the increase in kidney weight, the subchronic NOAEL and LOAEL
13   values are 30 and 60 |j,g/kg-day, respectively.
14
15          No information was located regarding the chronic toxicity, neurotoxicity or
16   developmental/reproductive toxicity of cylindrospermopsin.
17
18   4.6.2.  Inhalation
19
20          No information was located regarding the inhalation toxicity of cylindrospermopsin.
21
22   4.6.3.  Mode of Action Information
23
24          The liver and kidneys appear to be the main targets of cylindrospermopsin toxicity. The
25   mechanism for liver toxicity is incompletely characterized, but involves inhibition of protein
26   synthesis (Froscio et al., 2003; Terao et al., 1994).  Available evidence indicates that the protein
27   synthesis inhibition is not decreased by broad-spectrum CYP450 inhibitors, suggesting that it is
28   mediated by the parent compound (Froscio et al., 2003). Hepatocytotoxicity occurs  at higher
29   levels of cylindrospermopsin and appears to be CYP450-dependent, indicating the involvement
30   of metabolites and other mechanisms  (Froscio et al., 2003; Humpage et al., 2005; Norris et al.,
31   2002). Studies specifically investigating the inhibition of protein synthesis in the kidneys are not
32   available, although the results of the 11-week oral toxicity  study in mice (Humpage  and
33   Falconer, 2003) are consistent with an inhibition of protein synthesis.  Effects in this study
34   included decreased urinary protein and, at a higher dose, proximal renal tubular lesions. As
35   discussed in Section 4.5.2, the decrease in urinary protein excretion at low doses could reflect a
36   specific effect of cylindrospermopsin  on protein synthesis or, possibly, a functional change in the
37   nephron. The proximal renal tubular damage in mice (Humpage and Falconer, 2003), as well as
38   the clinical findings of renal insufficiency in the Palm Island human poisoning incident  (Blyth,
39   1980; Griffiths and  Saker, 2003), suggest that cytotoxic mechanisms may predominate in the
40   kidney at higher doses.
41
42          Genotoxic effects of cylindrospermopsin include DNA adduction and strand  breakage in
43   mouse liver (Shaw et al., 2000; Shen et al., 2002) and micronuclei formation in a lymphoblastoid
44   cell line (Humpage et  al., 2000). Broad spectrum CYP450 inhibitors inhibited
45   cylindrospermopsin-induced DNA damage in mouse hepatocytes at sub-cytotoxic concentrations
                                                23         DRAFT: DO NOT CITE OR QUOTE

-------
 1   (Humpage et al., 2005), suggesting that metabolites are responsible for cylindrospermopsin
 2   genotoxicity and that genotoxicity is a primary effect of the chemical.
 3
 4   4.7.    WEIGHT-OF-EVIDENCE EVALUATION AND CANCER
 5          CHARACTERIZATION
 6
 7   4.7.1. Summary of Overall Weight-of-Evidence
 8
 9          No information is available on the carcinogenicity of cylindrospermopsin in humans, and
10   no cancer studies of purified cylindrospermopsin have been conducted in animals.  A test of an
11   extract of C. raciborskii cells suggests that cylindrospermopsin has no tumor initiating activity in
12   mice (Falconer and Humpage, 2001).  A limited amount of data indicate that cylindrospermopsin
13   or a metabolite can covalently bind to DNA (Shaw et al., 2000) and cause cytogenetic damage,
14   as shown by induction of micronuclei (Humpage et al., 2000) and DNA strand breakage (Shen et
15   al., 2002). In accordance with the Guidelines for Carcinogen Risk Assessment (U.S. EPA,
16   2005a), the weight of evidence descriptor for the carcinogenic hazard potential of
17   cylindrospermopsin is "Inadequate Information to Assess Carcinogenic Potential."
18
19   4.7.2. Synthesis of Human, Animal and Other Supporting Evidence
20
21          No information was located regarding the carcinogenicity of purified cylindrospermopsin
22   in humans or animals.  There was no  indication that cylindrospermopsin had tumor initiating
23   activity in a test in which mice were administered a cell-free extract of freeze-dried C.
24   raciborskii cells by gavage followed by oral exposure to the tumor promoter TPA (Falconer and
25   Humpage, 2001).
26
27          The nucleotide structure of cylindrospermopsin, as well as the presence of potentially
28   reactive guanidine and sulfate groups, suggests the possibility of interference with DNA and/or
29   RNA synthesis and induction of mutations.  Covalent binding between DNA and
30   cylindrospermopsin (or a metabolite) (Shaw et al., 2000, 2001) and DNA strand breakage (Shen
31   et al., 2002) have been demonstrated in mouse liver, and micronuclei were induced in human
32   WIL2-NS lymphoblasts (Humpage et al., 2000). Although the available data indicate that DNA
33   strand breakage could be a key mechanism for cylindrospermopsin-induced cytogenetic damage
34   (Humpage et al., 2000; Shen et al., 2002), insufficient data are available to speculate on the
35   carcinogenic potential of cylindrospermopsin.
36
37   4.8.    SUSCEPTIBLE POPULATIONS AND LIFE STAGES
38
39   4.8.1. Possible Childhood Susceptibility
40
41          As discussed in Section 4.1, cylindrospermopsin has been implicated in the Palm Island
42   outbreak of a hepatoenteritis-like illness in 148 Australians (Blyth, 1980; Griffiths and Saker,
43   2003).  Of the  148 cases, 138 were children (mean age 8.4 years, range 2-16 years, 41% male
44   and 59% female) and 10 were adults (sex and age not reported). There are no reported
45   indications that the 138 children were from the households of the 10 adults or that the children
                                               24         DRAFT: DO NOT CITE OR QUOTE

-------
 1    and adults received different exposures, suggesting a possible increased sensitivity of children
 2    (the child:adult ratio is approximately 14:1).
 3
 4    4.8.2. Possible Gender Differences
 5
 6           There is no information on possible gender differences in the disposition of, or response
 7    to, cylindrospermopsin.
 8
 9    4.8.3. Other Possible Susceptible Populations
10
11           No data were located regarding populations that might be unusually susceptible to
12    cylindrospermopsin.  It is conceivable that individuals with liver and/or kidney disease might be
13    more susceptible than the general population because of compromised detoxification
14    mechanisms in the liver and impaired excretory mechanisms in the kidney.
                                                25         DRAFT: DO NOT CITE OR QUOTE

-------
 1                            5. DOSE-RESPONSE ASSESSMENTS
 2
 3
 4   5.1.    NARRATIVE DESCRIPTION OF THE EXTENT OF THE DATABASE
 5
 6          Studies on the absorption, tissue distribution, metabolism and elimination of
 7   cylindrospermopsin following oral, inhalation or dermal exposure have not been performed.
 8   Gastrointestinal absorption of cylindrospermopsin is indicated by the induction of systemic
 9   effects in oral toxicity studies.  Studies in which cylindrospermopsin was administered to mice
10   by acute i.p. injection indicate that it is largely distributed to the liver and rapidly eliminated in
11   the urine as unmetabolized compound. The liver and kidneys are the main targets of
12   cylindrospermopsin toxicity. Possible modes of action include inhibition of protein synthesis,
13   CYP450-mediated bioactivation to a reactive intermediate and covalent binding between parent
14   compound or a metabolite and DNA and/or RNA.
15
16          The only information on the toxicity of cylindrospermopsin in humans is from reports of
17   a poisoning outbreak that is attributed to the consumption of drinking water containing toxin-
18   producing C. raciborskii. Although the clinical picture of this hepatoenteritis-like illness is well
19   defined and includes bloody diarrhea, swollen liver and impaired kidney function, there are no
20   data on exposure levels that induced these effects.
21
22          Most of the available data on the toxicity of cylindrospermopsin in animals are available
23   from oral and i.p. studies that tested purified compound or extracts of C. raciborskii cells. These
24   studies are generally consistent in indicating that cylindrospermopsin causes lesions in the liver
25   and other organs, particularly the kidneys, spleen and thymus. The cell extract studies are not
26   useful for dose-response assessment of cylindrospermopsin due to the confounding factors
27   discussed in the introduction to Section 4.2. The database on oral toxicity of pure
28   cylindrospermopsin is limited by a small number of studies and insufficient reporting.  No
29   studies have been performed assessing the acute oral toxicity of pure cylindrospermopsin. Data
30   on the short-term oral toxicity of pure compound are available from inadequately reported
31   14-day gavage and 21-day drinking water studies in mice and rats. The reports of these studies
32   identify NOAELs and LOAELs for histopathology, but the adequacy of these effect levels
33   cannot be verified  due to a virtual lack of any additional information on the experimental designs
34   and results. Data on the subchronic oral toxicity of pure cylindrospermopsin are available from a
35   well-designed and  reported 11-week study in mice that provides a suitable basis for derivation of
36   a subchronic oral RfD value. No chronic toxicity, reproductive toxicity, developmental toxicity
37   or carcinogenicity  studies of pure cylindrospermopsin have been performed.
38
39          No information is available on the inhalation toxicity of cylindrospermopsin.
40
41   5.2.    ORAL REFERENCE DOSE (RfD)
42
43   5.2.1. Data Considered in Deriving Reference Values
44
45          Data considered in deriving oral RfDs for each duration of exposure are summarized in
46   Table 4-1 (Section 4.6.1).
                                                26         DRAFT: DO NOT CITE OR QUOTE

-------
 1   5.2.2.  Acute Duration
 2
 3          5.2.2.1. Choice of Principal Study and Critical Effect - with Rationale and
 4                   Justification
 5
 6          Derivation of an acute oral RfD for cylindrospermopsin is precluded by insufficient data.
 7   The only information on the toxicity of cylindrospermopsin in humans is the outbreak of a
 8   hepatoenteritis-like illness that is attributed to the consumption of drinking water containing C.
 9   raciborskii (Blyth, 1980;  Griffiths and Saker, 2003; Hawkins et al., 1985; Ohtani et al., 1992).
10   Although the clinical picture of the illness is well defined, measured or estimated exposure levels
11   have not been reported. No acute oral toxicity studies of purified cylindrospermopsin have been
12   performed in animals.  Single-dose studies of suspensions or cell-free extracts of C. raciborskii
13   cells were conducted in mice, but only near-lethal to lethal dose levels were tested (Falconer et
14   al., 1999; Seawright et al., 1999; Shaw et al., 2000, 2001).
15
16   5.2.3.  Short-Term Duration
17
18          5.2.3.1. Choice of Principal Study and Critical Effect - with Rationale and
19                   Justification
20
21          Derivation of a short-term oral RfD for cylindrospermopsin is precluded by insufficient
22   data.  The only information relevant to the short-term toxicity of cylindrospermopsin in humans
23   is qualitative data on the outbreak of the hepatoenteritis-like illness that is attributed  to the
24   consumption of drinking water containing C. raciborskii (Blyth, 1980; Griffiths and  Saker, 2003;
25   Hawkins et al., 1985; Ohtani et al., 1992). As discussed in Sections 4.2.1.2 and 4.5.1, a limited
26   amount of information is  available on the short-term oral toxicity of cylindrospermopsin from
27   poorly reported 14-day gavage and 21-day drinking water studies (Shaw et al., 2000, 2001).  The
28   14-day study reported a NOAEL of 0.05 mg/kg-day and LOAEL of 0.15 mg/kg-day  for liver
29   fatty infiltration in mice, and the 21-day study reported a free-standing NOAEL of 0.2
30   mg/kg-day for histopathology in mice and rats. The appropriateness  of these effect levels cannot
31   be assessed due to inadequate information on the design and results of the studies.
32
33   5.2.4.  Subchronic Duration
34
35          5.2.4.1. Choice of Principal Study and Critical Effect - with Rationale and
36                   Justification
37
38          The comprehensive 11-week subchronic study in mice (Humpage and Falconer, 2003),
39   detailed in Section 4.2.1.3, is the only subchronic study of purified cylindrospermopsin and
40   provides a suitable basis for RfD derivation.  The LOAEL was 60 |j,g/kg-day for increased
41   relative kidney weight. At  120 |j,g/kg-day, there was a significant decrease in urinary protein
42   concentration and minor histopathological changes in the liver. Decreased urinary protein and
43   increased relative kidney  weight are both potential indicators of suppressed protein synthesis, a
44   known mode of action of cylindrospermopsin. The decrease in urinary protein is consistent with
45   decreased availability of protein and the increase in kidney weight may reflect a compensatory
46   hyperplasia, such that the kidney, as a protein-synthesizing organ, is stimulated to grow in an
                                                 27        DRAFT: DO NOT CITE OR QUOTE

-------
 1   attempt to maintain homeostasis in face of a chemical-related decrease in protein synthesis
 2   (Humpage and Falconer, 2003).  Information supporting this hypothesis, as well as the
 3   possibility that the decrease in urinary protein excretion reflects a functional change in the
 4   nephron, is discussed in Section  4.5.2. Because the changes are consistent with a known mode
 5   of action and represent part of the progression of effects leading to toxicity, they  are considered
 6   to be adverse and indicate that the LOAEL and NOAEL are 60 and 30 |j,g/kg-day, respectively.
 7
 8          5.2.4.2.  Methods of Analysis - Including Models (PBPK, BMD, etc.)
 9
10          A point of departure (POD) can be determined using the kidney weight data and BMD
11   modeling, but BMD analysis of the urinary protein and histopathology data is precluded by
12   insufficient data.  In particular, the urinary protein data are limited by inadequate reporting
13   (mean concentrations and errors are conveyed in a bar graph with no numerical values
14   specifically reported, no indication if the error bars represent standard deviation or standard
15   error, and no indication of numbers of animals) and the pathology findings are limited by  a lack
16   of incidence data.
17
18         In accordance with current BMD technical guidance (U.S. EPA, 2000c), available
19   continuous-variable models in the EPA Benchmark Dose Software (BMDS version 1.3.2; linear,
20   polynomial, power and Hill models) were fit to the data for changes in mean relative kidney
21   weight shown in Table 5-1. Statistical tests in the BMDS showed that variance was
22   homogeneous across dose groups. BMDs and BMDLs were calculated using 1 standard
23   deviation above the control mean as the benchmark response level (BMR), while assuming
24   homogenous variance across  groups. Using data from all dose groups, an adequate fit to the data
25   was obtained with the Hill model (Table 5-2), but the BMDS was not able to compute a BMDL.
26   After dropping the high dose group, there were insufficient degrees of freedom remaining to fit
27   the Hill model, but the linear model adequately fit the data and produced an estimated BMD of
28   43.1 |j,g/kg-day and BMDL of 33.1 |j,g/kg-day. The two-degree polynomial and power models
29   defaulted to the same linear model, albeit with lower p-value and/or higher Aikake's Information
30   Criteria (AIC) due to the greater number of parameters in these models. The BMD modeling
31   results are summarized in Table  5-2 and detailed in Appendix A, and  the fit of the linear model
32   to the data is shown in Figure 5-1.  The BMDL of 33.1  |j,g/kg-day is similar to the 30 |j,g/kg-day
33   NOAEL for increased kidney weight and is used as the POD for the RfD.
34
35          5.2.4.3.  RfD Derivation - Including Application of UFs
36
37          The BMDL of 33.1 |j,g/kg-day for increased relative kidney weight was used as the point
38   of departure (POD) for the subchronic RfD. Dividing the BMDL of 33.1 |j,g/kg-day by a
39   composite uncertainty factor  (UF) of 1000 results in a subchronic RfD for cylindrospermopsin of
40   3xlO"5 mg/kg-day.
41
                                                28        DRAFT: DO NOT CITE OR QUOTE

-------
Table 5-1. Relative Kidney Weights in Mice Exposed to Purified Cylindrospermopsin for 1 1 Weeks

Relative kidney weight
(mean + standard deviation)
Oral Dose (|j,g/kg-day)
0
1.48 + 0.10
(10)a
30
1.57 + 0.14
(10)
60
1.66 + 0.16b
(9)
120
1.82 + 0.12C
(9)
240
1.78 + 0.17C
(6)
2   a Values in parentheses are the number of animals evaluated in each group
3   b Statistically significant difference from controls (p<0.05)
4   c Statistically significant difference from controls (p<0.001)
5   Source: Humpage and Falconer (2003)
                                                               29
DRAFT: DO NOT CITE OR QUOTE

-------
Table 5-2. Summary of Benchmark Dose Modeling (Relative Kidney Weight/
Model Fit to Means
df
p-Value for
Model Fit
AIC for
Fitted
Model
BMD
(|ig/kg-day)
BMDL
(Hg/kg-day)
Relative kidney weight, all dose groups
(p=0.59 for test of homogenous variance, indicating assumption of homogenous variance is
appropriate)
Linear
2-Degree polynomial (pos
betas)
Power (power >=1)
Hill (power >=1)
3
2
2
1
0.01
0.003
0.003
0.21
-120.85
-120.85
-116.85
-124.74
106.07
106.07
106.07
43.19b
76.56
76.56
76.56
NAC
Relative kidney weight, high dose group dropped
(p=0.52 for test of homogenous variance, indicating assumption of homogenous variance is
appropriate)
Linear
2-Degree polynomial (pos
betas)
Power (power >=1)
Hill (power >=1)
2
1
1
0
0.98
0.84
0.84
NAd
-116.47
-116.47
-112.47
-110.51
43.90
43.90
43.90
41.20
33.07
33.07
33.07
21.72
2   a Modeling conducted assuming homogenous variance and using BMR of 1 standard deviation
3   b Optimum BMD may not have been found (i.e. bad completion code in the BMDS optimization
4   routine)
5   c BMDL computation failed
6   d The Chi-Square test for fit is not valid due to insufficient degrees of freedom (df)
7   NA = not available
                                             30
DRAFT: DO NOT CITE OR QUOTE

-------
                              Linear Model with 0.95 Confidence Level
1
2
o
J
4
5
       cu
       tn
       c
       WJ
       cu
       CC
       (U
          1.9
          1.8
          1.7
          1.5
          1.4
               Linear
        22:31 05/122005
                             BMDL
                                BMD
                           20
                             40
 60
dose
80
100
120
Figure 5-1. Linear Model Fit to Relative Kidney Weight Data (High Dose Group Dropped)
                       Source: Humpage and Falconer (2003)
                                                        DRAFT: DO NOT CITE OR QUOTE

-------
 1          Subchronic RfD           =      BMDL-UF
 2                                     =      33.1 ng/kg-day-1000
 3                                     =      0.00003 mg/kg-day or 3x105 mg/kg-day
 4
 5          The composite UF of 1000 includes a factor of 10 for interspecies extrapolation, a factor
 6   of 10 to account for interindividual variability in the human population and a factor of 10 for
 7   database limitations, as follows.
 8
 9       •  A default 10-fold UF is used to account for the interspecies variability in extrapolating
10          from laboratory animals to humans. No information is available on the toxicity of
11          purified cylindrospermopsin in humans, and no data on toxicokinetic differences between
12          animals and humans in the disposition of ingested cylindrospermopsin are available.

13       •  A 10-fold UF is used to account for variation in sensitivity within human populations
14          because there is insufficient information on the degree to which humans of varying
15          gender, age, health status or genetic makeup might vary in the disposition of, or response
16          to, ingested cylindrospermopsin.  As discussed in Section 4.1, data from the Palm Island
17          outbreak of a hepatoenteritis-like illness (Blyth, 1980; Griffiths and Saker, 2003) suggest
18          a possible increased sensitivity of children to cylindrospermopsin.

19       •  A 10-fold UF is used to account for deficiencies in the database.  There is no information
20          on the longer-term toxicity of cylindrospermopsin in humans.  Other database
21          deficiencies include a lack of particular kinds of animal studies on purified
22          cylindrospermopsin, including a chronic study, subchronic or chronic studies in a second
23          species and reproductive and developmental toxicity studies.

24          The NOAEL/LOAEL approach and an UF of 1000 would also yield an RfD of 0.00003
25   mg/kg-day due to the similarity of the NOAEL and BMDL for increased kidney weight (30 and
26   33.1 |j,g/kg-day, respectively).
27
28   5.2.5.  Chronic Duration
29
30          5.2.5.1. Choice of Principal Study and Critical Effect - with Rationale and
31                  Justification
32
33          Derivation of a chronic oral RfD for cylindrospermopsin is precluded by insufficient
34   data. No information is available on the chronic toxicity of cylindrospermopsin by any route of
35   exposure. The 11-week study (Humpage and Falconer, 2003) used to derive the subchronic RfD
36   was considered for use in the derivation of a chronic RfD; however, this approach was rejected
37   due to the lack of information on the potential progression of cylindrospermopsin-induced
38   adverse effects with increased exposure duration. The use of the POD from the 11-week
39   subchronic study for the derivation of a chronic RfD would require  the application of a
40   subchronic-to-chronic UF to account for the uncertainties involved in extrapolating across
41   exposure durations. The application of a full subchronic-to-chronic UF of 10, along with UFs of
42   10 in three other areas of uncertainty (interspecies UF, intraspecies UF, database UF), would
43   result in a total composite uncertainty factor of 10,000. A composite uncertainty factor of this
                                                32         DRAFT: DO NOT CITE OR QUOTE

-------
 1    magnitude suggests that the database is insufficient to support the derivation of an RfD for
 2    chronic exposure; therefore, no chronic oral RfD is derived.
 3
 4    5.2.6. Route-to-Route Extrapolation
 5
 6           Derivation of acute, short-term, and chronic RfD values for cylindrospermopsin by route-
 7    to-route extrapolation could not be considered due to a lack of inhalation data.
 8
 9    5.3.    INHALATION REFERENCE CONCENTRATION (RfC)
10
11           No information is available on  the toxicity of inhaled cylindrospermopsin.
12
13    5.4.    CANCER ASSESSMENT
14
15           No dose-response or other information is available regarding the carcinogenicity of pure
16    cylindrospermopsin.
                                               33        DRAFT: DO NOT CITE OR QUOTE

-------
 1               6.  MAJOR CONCLUSIONS IN THE CHARACTERIZATION OF
 2                              HAZARD AND DOSE RESPONSE
 3
 4
 5   6.1.    HUMAN HAZARD POTENTIAL
 6
 7          Cylindrospermopsin is a naturally occurring chemical produced by Cylindrospermopsis
 8   (particularly C. raciborskii) and at least four other genera of freshwater cyanobacteria.
 9   Toxicokinetic studies of Cylindrospermopsin have not been performed using natural routes of
10   exposure, but oral toxicity studies show that it is absorbed from the gastrointestinal tract, and i.p.
11   toxicokinetic studies indicate that it is mainly distributed to the liver and excreted in the urine as
12   unmetabolized compound. Main targets of Cylindrospermopsin toxicity include the liver and
13   kidneys, and possible modes of action include inhibition of protein synthesis, bioactivation to a
14   reactive intermediate and covalent binding of parent compound or a metabolite to DNA and/or
15   RNA.
16
17          The main information on the toxicity of Cylindrospermopsin in humans is from
18   qualitative reports of a hepatoenteritis-like illness that is attributed to the acute or short-term
19   consumption of drinking water containing C. raciborskii. The database on oral toxicity of
20   purified Cylindrospermopsin in animals is limited by a small number of studies and insufficient
21   reporting. No studies have been performed assessing the acute oral toxicity of pure
22   Cylindrospermopsin. Information on short-term oral toxicity is available from inadequately
23   reported 14- and 21-day studies in mice and rats.  Data on the subchronic oral toxicity of pure
24   Cylindrospermopsin are available from a comprehensive 11-week study that identified NOAELs
25   and LOAELs for kidney and liver effects in mice. No chronic toxicity, reproductive toxicity,
26   developmental toxicity or carcinogenicity studies of Cylindrospermopsin have been conducted.
27   Testing following inhalation has not been performed.
28
29   6.2.    DOSE RESPONSE
30
31          Kidney effects data in the 11-week toxicity study (Humpage and Falconer, 2003) provide
32   a suitable basis for deriving a subchronic oral RfD.  The most sensitive effect in this study was
33   increased relative kidney weight; decreased urinary protein and minor histopathological damage
34   to the liver occurred at the next highest dose. As discussed in Section 4.5.2,  increased kidney
35   weight and decreased urinary protein are consistent with suppressed protein synthesis, a known
36   mode of action of Cylindrospermopsin, and represent part of the progression  of effects leading to
37   toxicity. Based on a BMDL of 33.1  |j,g/kg-day  for increased relative kidney weight in mice, a
38   subchronic RfD of 3xlO"5 mg/kg-day was derived by dividing the BMDL by a UF of 1000.  The
39   UF comprises component factors of 10 for interspecies extrapolation, 10 for interindividual
40   variability and 10 for database deficiencies.  Acute, short-term and chronic oral RfDs could not
41   be derived due to inadequate data. Inhalation RfC derivation is precluded by the lack of data for
42   this route of exposure. There is inadequate information to evaluate the carcinogenicity  of
43   Cylindrospermopsin.
                                                34         DRAFT: DO NOT CITE OR QUOTE

-------
 1                                      7. REFERENCES
 2

 3   Akimov, S.S. and A.M. Belkin. 2001.  Cell surface tissue transglutaminase is involved in
 4   adhesion and migration of monocytic cells on fibronectin. Blood. 98(5): 1567-1576.

 5   Bachmann, S., A.B. Dawnay, N. Bouby and L. Bankir. 1991. Tamm-Horsfall protein excretion
 6   during chronic alterations in urinary concentration and protein intake in the rat. Renal Physiol.
 7   Biochem. 14(6):236-245.

 8   Bachmann, S., K. Mutig, J. Bates et al. 2005. Renal effects of Tamm-Horsfall protein
 9   (uromodulin) deficiency in mice. Am. J. Renal Physiol. 288(3):F559-F567.

10   Banker, R., B. Teltsch, A. Sukenik and S. Carmeli. 2000. 7-Epicylindrospermopsin, a toxic
11   minor metabolite of the cyanobacterium Aphanizomenon ovalisporum from Lake Kinneret,
12   Israel. J. Nat. Prod. 63(3):387-389.

13   Banker, R., S. Carmeli, M. Werman et al. 2001. Uracil moiety is required for toxicity of the
14   cyanobacterial hepatotoxin cylindrospermopsin. J. Toxicol. Environ. Health A. 62(4):281-288.

15   Bates, J.M., H.M. Raffi, K. Prasadan et al. 2004. Tamm-Horsfall protein knockout mice are
16   more prone to urinary tract infection: Rapid communication. Kidney Int.  65(3):791-797.

17   Blyth, S. 1980.  Palm Island mystery disease. Med. J. Aust. 2(l):40-42.

18   Carmichael, W.W., S.M.F.O. Azevedo, IS. An et al.  2001. Human fatalities from
19   cyanobacteria: Chemical  and biological evidence for cyanotoxins.  Environ. Health Perspect.
20   109(7):663-668.

21   Chiswell, R.K., G.R. Shaw, G. Eaglesham et al.  1999. Stability of cylindrospermopsin, the
22   toxin from cyanobacterium, Cylindrospermopsis raciborskii: Effect of pH, temperature, and
23   sunlight on decomposition. Environ. Toxicol. 14(1):155-161.

24   Falconer, I.R. and A.R. Humpage.  2001. Preliminary evidence for in vivo tumour initiation by
25   oral administration of extracts of the blue-green alga Cylindrospermopsis raciborski containing
26   the toxin cylindrospermopsin.  Environ. Toxicol.  16(2): 192-195.

27   Falconer, I.R., SJ. Hardy, A.R. Humpage et al. 1999. Hepatic and renal toxicity of the blue-
28   green alga (cyanobacterium) Cylindrospermopsis raciborski in male Swiss albino mice.
29   Environ. Toxicol. 14(1):143-150.

30   Fastner, J., R. Heinze, A.R. Humpage  et al.  2003.  Cylindrospermopsin occurrence in two
31   German lakes and preliminary assessment of toxicity and toxin production of
32   Cylindrospermopsis raciborskii (cyanobacteria) isolates.  Toxicon.  42(3):313-321.

33   Fessard, V. and C. Bernard. 2003.  Cell alterations but no DNA strand breaks induced in vitro
34   by cylindrospermopsin in CHO Kl cells. Environ. Toxicol. 18(5):353-359.
                                                35         DRAFT: DO NOT CITE OR QUOTE

-------
 1   Fesus, L., V. Thomazy and A. Falus. 1987. Induction and activation of tissue transglutaminase
 2   during programmed cell death.  FEES Lett. 224(1): 104-108.

 3   Froscio, S.M., A.R. Humpage, P.C. Burcham and I.R. Falconer. 2003. Cylindrospermopsin-
 4   induced protein synthesis inhibition and its dissociation from acute toxicity in mouse
 5   hepatocytes. Environ. Toxicol. 18(4):243-251.

 6   Grenard, P., S. Bresson-Hadni,  S. El Alaoui et al. 2001. Transglutaminase-mediated cross-
 7   linking is involved in the stabilization of extracellular matrix in human liver fibrosis. J. Hepatol.
 8   35(3):367-375.

 9   Griffiths, DJ. and M.L. Saker.  2003. The Palm Island mystery disease 20 years on: A review of
10   research on the cyanotoxin cylindrospermopsin.  Environ. Toxicol. 18(2):78-93.

11   Hawkins, P.R., M.T.C. Runnegar, A.R.B. Jackson and I.R. Falconer.  1985.  Severe
12   hepatotoxicity caused by the tropical cyanobacterium (blue-green alga) Cylindrospermopsis
13   raciborskii (Woloszynska) Seenaya and Subba Raju isolated from a domestic water supply
14   reservoir. Appl. Environ. Microbiol. 50(5): 1292-1295.

15   Hawkins, P.R., N.R. Chandrasena, GJ.  Jones et al.  1997. Isolation and toxicity of
16   Cylindrospermopsis raciborskii from an ornamental lake. Toxicon.  35(3):341-346.

17   Hayman, J. 1992. Beyond the  Barcoo - probable human tropical cyanobacterial poisoning in
18   outback Australia. Med. J. Aust.  157(11-12):794-796.

19   HSDB (Hazardous Substances Data Bank). 2006. Produced by the U.S. National Library of
20   Medicine (NLM), Toxicology Data Network (TOXNET), Bethesda, Maryland.  Accessed
21   January 2, 2006 at http://toxnet.nlm.nih.gov/cgi-bin/sis/htmlgen7HSDB.

22   Humpage, A.R. and I.R. Falconer. 2003.  Oral toxicity of the cyanobacterial toxin
23   cylindrospermopsin in male Swiss albino mice: Determination  of no observed adverse effect
24   level for deriving a drinking water guideline value.  Environ. Toxicol.  18(2):94-103.

25   Humpage, A.R., M.  Fenech, P.  Thomas and I.R. Falconer. 2000. Micronucleus induction and
26   chromosome loss in transformed human white cells indicate clastogenic and aneugenic action of
27   the cyanobacterial toxin, cylindrospermopsin.  Mutat. Res. 472:155-161.

28   Humpage, A.R., F. Fontaine,  S. Froscio, P. Burcham and I.R. Falconer. 2005.
29   Cylindrospermopsin genotoxicity and cytotoxicity:  Role of cytochrome P-450 and oxidative
30   stress. J. Toxicol. Environ. Health, Part A. 68(9):739-753.

31   Lewis, R.J. 2000. Sax's Dangerous Properties of Industrial Materials, Vol. 1-3, 10th ed. John
32   Wiley & Sons Inc., New York,  NY. p.  1061.

33   Li, R., W.W. Carmichael, S. Brittain et  al. 2001.  First report of the cyanotoxins
34   cylindrospermopsin and deoxycylindrospermopsin from Raphidiopsis curvata (cyanobacteria).
35   J. Phycol. 37(6): 1121-1126.
                                               36        DRAFT: DO NOT CITE OR QUOTE

-------
 1   Mirza, A., S-L. Liu, E. Frizell et al. 1997. A role for tissue transglutaminase in hepatic injury
 2   and fibrogenesis, and its regulation by NF-KB.  Am. J. Physiol.  272:G281-G288.

 3   Norris, R.L., G.K. Eaglesham, G. Pierens et al.  1999. Deoxycylindrospermopsin, an analog of
 4   cylindrospermopsin from Cylindrospermopsis raciborskii.  Environ. Toxicol.  14(1): 163-165.

 5   Norris, R.L.G., A. A. Seawright, G.R. Shaw et al. 2001. Distribution of 14C cylindrospermopsin
 6   in vivo in the mouse. Environ. Toxicol. 16(6):498-505.

 7   Norris, R.L.G., A.A. Seawright, G.R. Shaw et al. 2002. Hepatic xenobiotic metabolism of
 8   cylindrospermopsin in vivo in the mouse. Toxicon. 40(4):471-476.

 9   NRC (National Research Council). 1983.  Risk Assessment in the Federal Government:
10   Managing the Process.  National Academy Press, Washington, DC.

11   Ohtani, I, R.E. Moore andM.T.C. Runnegar. 1992.  Cylindrospermopsin: A potent hepatotoxin
12   from the blue-green alga Cylindrospermopsis raciborskii.  J. Am. Chem. Soc.
13   114(20):7941-7942.

14   O'Neil, M.J., Ed.  2001. The Merck Index - An Encyclopedia of Chemicals, Drugs, and
15   Biologicals, 13th ed. Merck and Co., Inc., Whitehouse Station, NJ.

16   Pesce, AJ. and M.R. First.  1979.  Proteinuria. An integrated review.  Marcell Dekker.
17   pp. 54-79

18   Piacentini, M., M.G. Farrace, L. Piredda et al. 2002.  Transglutaminase overexpression
19   sensitizes neuronal cell lines to apoptosis by increasing mitochondrial membrane potential and
20   cellular oxidative stress. J. Neurochem. 81(5):1061-1072.

21   Pilotto, L., P. Hobson, M.D. Burch, G.  Ranmuthugala, R. Attewell and W. Weightman.  2004.
22   Acute skin irritant effects of cyanobacteria (blue-green algae) in healthy volunteers. Aust. N. Z.
23   J. Public Health. 28(3):220-224.

24   Runnegar, M.T., S.M. Kong, Y-Z. Zhong et al.  1994. The role of glutathione in the toxicity of a
25   novel cyanobacterial alkaloid cylindrospermopsin in cultured rat hepatocytes.  Biochem.
26   Biophys. Res. Commun. 201(1):235-241.

27   Runnegar, M.T., S.M. Kong, Y-Z. Zhong and S.C. Lu. 1995. Inhibition of reduced glutathione
28   synthesis by cyanobacterial alkaloid cylindrospermopsin in cultured rat hepatocytes. Biochem.
29   Pharmacol. 49(2):219-225.

30   Runnegar, M.T., C.  Xie, B.B. Snider et al. 2002. In vitro hepatotoxicity of the cyanobacterial
31   alkaloid cylindrospermopsin and related synthetic analogues.  Toxicol. Sci. 67(l):81-87.

32   Schoel, B. and G. Pfleiderer. 1987. The amount of Tamm-Horsfall protein in the human kidney,
33   related to its daily excretion. J. Clin. Chem. Clin. Biochem. 25(10):681-682.
                                                37        DRAFT: DO NOT CITE OR QUOTE

-------
 1   Seawright, A.A., C.C. Nolan, G.R. Shaw et al. 1999. The oral toxicity for mice of the tropical
 2   cyanobacterium Cylindrospermopsin raciborskii (Woloszynska). Environ. Toxicol.
 3   14(1):135-142.

 4   Shaw, G.R., A. Sukenik, A. Livne et al. 1999. Blooms of the Cylindrospermopsin containing
 5   cyanobacterium, Aphanizomenon ovalisporum (Forti), in newly constructed lakes, Queensland,
 6   Australia.  Environ. Toxicol.  14(1): 167-177.

 7   Shaw, G.R., A.A. Seawright, M.R. Moore and P.K. Lam. 2000. Cylindrospermopsin, a
 8   cyanobacterial alkaloid: Evaluation of its toxicologic activity. Ther. Drug Monit.  22(l):89-92.

 9   Shaw, G.R., A.A. Seawright and M.R. Moore. 2001. Toxicology and human health implications
10   of the cyanobacterial toxin Cylindrospermopsin. In: Mycotoxins and Phycotoxins in Perspective
11   at the Turn of the Millennium, WJ. Dekoe, R.A. Samson, H.P. van Egmond et al., Ed. IUPAC
12   & AOAC International, Brazil,  p. 435-443.

13   Shen, X., P.K.S. Lam, G.R. Shaw and W. Wickramasinghe.  2002.  Genotoxicity investigation of
14   a cyanobacterial toxin, Cylindrospermopsin. Toxicon.  40(10):1499-1501.

15   Shen, X., G.R. Shaw, G.A. Codd et al. 2003.  DNA microarray analysis of gene expression in
16   mice treated with the cyanobacterial toxin, Cylindrospermopsin. In: Proceedings of the Eighth
17   Canadian Workshop on Harmful Marine Algae, S.S. Bates, Ed. Fisheries and Oceans Canada,
18   Monkton, New Brunswick, p. 49-51.  Available at
19   http://www.glf. dfo-mpo.gc.ca/sci-sci/cwhma-atcamn/8th_cwhma_proceedings.pdf

20   Sigma. 2006. Cylindrospermopsin from Cylindrospermopsis raciborskii. C9866. Sigma-
21   Aldrich, Inc.,  Saint Louis, MO.  Accessed January 23, 2006 at
22   http://www.sigmaaldrich.com/catalog/search/ProductDetail?ProdNo=C9866&Brand=SIGMA.

23   Terao, K., S. Ohmori, K. Igarashi et al.  1994. Electron microscopic studies  on experimental
24   poisoning in mice induced by Cylindrospermopsin isolated from blue-green alga Umezakia
25   natans.  Toxicon. 32(7):833-843.

26   U.S. EPA.  1986a. Guidelines for the Health Risk Assessment of Chemical Mixtures.  Fed. Reg.
27   51(185):34014-34025.

28   U.S. EPA.  1986b. Guidelines for Mutagenicity Risk Assessment. Fed. Reg.
29   51(185):34006-34012.

30   U.S. EPA.  1988. Recommendations  for and Documentation of Biological Values for Use in
31   Risk Assessment. Prepared by the Office of Health and Environmental Assessment,
32   Environmental Criteria and Assessment Office, Cincinnati, OH for the Office of Solid Waste and
33   Emergency Response, Washington, DC. EPA 600/6-87/008. NTIS PB88-179874/AS.

34   U.S. EPA.  1991. Guidelines for Developmental Toxicity Risk Assessment.  Fed. Reg.
35   56(234):63798-63826.
                                               38        DRAFT: DO NOT CITE OR QUOTE

-------
 1   U.S. EPA.  1994a. Interim Policy for Particle Size and Limit Concentration Issues in Inhalation
 2   Toxicity Studies: Notice of Availability. Fed. Reg.  59(206):53799.

 3   U.S. EPA.  1994b. Methods for Derivation of Inhalation Reference Concentrations and
 4   Application of Inhalation Dosimetry. U.S. Environmental Protection Agency, Office of
 5   Research and Development,Washington, DC. EPA/600/8-90/066F. NTIS PB2000-500023.
 6   Available at http://www.epa.gov/iris/backgr-d.htm.

 7   U.S. EPA.  1995. Use of the Benchmark Dose Approach in Health Risk Assessment. U.S.
 8   Environmental Protection Agency, Risk Assessment Forum, Washington, DC.
 9   EPA/630/R-94/007.  NTIS PB95-213765.  Available at http://www.epa.gov/iris/backgr-d.htm.

10   U.S. EPA.  1996. Guidelines for Reproductive Toxicity Risk Assessment. Fed. Reg.
11   61(212):56274-56322.

12   U.S. EPA.  1998a. Guidelines for Neurotoxicity Risk Assessment. Fed. Reg.
13   63(93):26926-26954.

14   U.S. EPA.  1998b. Science Policy Council Handbook: Peer Review. U.S. Environmental
15   Protection Agency, Office of Research and Development, Washington, DC. EPA/100/B-98/001.
16   NTIS PB98-140726.

17   U.S. EPA.  2000a. Science Policy Council Handbook: Peer Review, 2nd ed. U.S. Environmental
18   Protection Agency, Office of Research and Development, Washington, DC. EPA/100/B-00/001.
19   Available at http://www.epa.gov/iris/backgr-d.htm.

20   U.S. EPA.  2000b. Science Policy Council Handbook: Risk Characterization. U.S.
21   Environmental Protection Agency, Office of Research and Development, Washington, DC.
22   EPA/ 100/B-00/002.  Available at http://www.epa.gov/iris/backgr-d.htm.

23   U.S. EPA.  2000c. Benchmark Dose Technical Guidance Document [External Review Draft].
24   U.S. Environmental Protection Agency, Risk Assessment Forum, Washington, DC.
25   EPA/630/R-00/001.  Available at http://www.epa.gov/iris/backgr-d.htm.

26   U.S. EPA.  2000d. Supplementary Guidance for Conducting Health Risk Assessment of
27   Chemical Mixtures.  U.S. Environmental Protection Agency, Office of Research and
28   Development, Washington, DC.  EPA/630/R-00/002.  Available at
29   http://www.epa.gov/iris/backgr-d.htm.

30   U.S. EPA.  2002. A Review of the Reference Dose and Reference Concentration Processes.
31   U.S. Environmental Protection Agency, Risk Assessment Forum, Washington, DC.
32   EPA/630/P-02/002F. Available at http://www.epa.gov/iris/backgr-d.htm.

33   U.S. EPA.  2005a. Guidelines for Carcinogen Risk Assessment.  U.S. Environmental Protection
34   Agency, Risk Assessment Forum, Washington, DC. EPA/630/P-03/001B. Available at
35   http://www.epa.gov/iris/backgr-d.htm.
                                               39        DRAFT: DO NOT CITE OR QUOTE

-------
 1    U.S. EPA. 2005b.  Supplemental Guidance for Assessing Susceptibility from Early-Life
 2    Exposure to Carcinogens. U.S. Environmental Protection Agency, Risk Assessment Forum,
 3    Washington, DC. EPA/630/R-03/003F.  Available at http://www.epa.gov/iris/backgr-d.htm.

 4    WHO (World Health Organization).  1999. Toxic Cyanobacteria in Water: A Guide to Their
 5    Public Health Consequences, Monitoring, and Management, I. Chorus and J. Bartram, Ed. ISBN
 6    0-419-23930-8. World Health Organization,  Geneva, Switzerland.  Available at
 7    http://www.who.int/water_sanitation_health/resourcesquality/toxcy anobacteria.pdf

 8    Zhang, L-X., KJ. Mills, M.L. Dawson et al.  1995. Evidence for the involvement of retinoic
 9    acid receptor RARa-dependent signaling pathway in the induction of tissue transglutaminase and
10    apostosis by retinoids.  J. Biol. Chem. 270(11):6022-6029.
                                               40         DRAFT: DO NOT CITE OR QUOTE

-------
                      APPENDIX A




BENCHMARK DOSE MODELING RESULTS FOR CYLINDROSPERMOPSIN
                         A-l

-------
Part I.




Humpage and Falconer 2003




male mice treated with purified cylindrospermopsin




rel kidney wt
                                      A-2

-------
       Polynomial Model.  Revision:  2.2  Date: 9/12/2002
       Input Data File:  C:\BMDS\DATA\CYLINDRO.(d)
       Gnuplot Plotting File:   C:\BMDS\DATA\CYLINDRO.plt
                                         Thu May 12 22:04:04 2005
HMDS MODEL RUN


  The form of the response function is:

  Y[dose]  = beta 0 + beta l*dose + beta 2*doseA2 + ...
  Dependent variable = MEAN
  Independent variable = dose
  rho is set to 0
  Signs of the polynomial coefficients are not restricted
  A constant variance model is fit

  Total number of dose groups = 5
  Total number of records with missing values = 0
  Maximum number of iterations = 250
  Relative Function Convergence has been set to:  le-008
  Parameter Convergence has been set to: le-008
                 Default Initial Parameter Values
                         alpha =     0.018741
                           rho =            0   Specified
                        beta_0 =        1.551
                        beta 1 =   0.00123333
                                Parameter Estimates
   Variable
   alpha
   beta_0
   beta 1
                            95.0% Wald Confidence Interval
Estimate    Std.  Err.     Lower Conf.  Limit   Upper Conf. Limit
0.0215477   0.00459399   0.0125437           0.0305518
1.54205     0.0311925     1.48091             1.60318
0.00138389  0.000287871   0.000819671
                                0.00194811
          Asymptotic Correlation Matrix of Parameter Estimates
    alpha
   beta_0
   beta 1
   alpha
       1
2.1e-009
1.2e-010
  beta_0
2.1e-009
       1
    -0.7
  beta_l
1.2e-010
    -0.7
       1
    Table of Data and Estimated Values of Interest
                                    A-3

-------
Dose
Res .
0
30
60
120
240
N
10
10
9
9
6
Obs
1.
1.
1.
1.
1.
Mean
48
57
66
82
78
Obs

0
0
0
0
Std Dev
0.1
.14
.16
.12
.17
Est
1.
1.
1.
1.
1.
Mean
54
58
63
71
87
Est
0.
0.
0.
0.
0.
Std Dev
147
147
147
147
147
ChiA
-
-0
0

-
2
1.34
.292
.714
2.29
1.57
  Model Descriptions for likelihoods calculated
 Model Al:        Yij
           Var{e(ij)}

 Model A2:        Yij
           Var{e(ij) }

 Model  R:         Yi
            Var{e(i) }
             Mu (i) + e (ij )
             SigmaA2

             Mu (i) + e (ij )
             Sigma(i)A2

             Mu + e (i)
             SigmaA2
                       Likelihoods of Interest
Model
Al
A2
fitted
R
Log (likelihood)
68.148702
69.554943
62.424690
52.631671
DF
6
10
2
2
AIC
-124.297404
-119.109885
-120.849381
-101.263343
 Test 1:  Does response and/or variances differ among dose
levels
          (A2 vs. R)
 Test 2:  Are Variances Homogeneous  (Al vs A2)
 Test 3:  Does the Model for the Mean Fit  (Al vs.  fitted)
   Test

   Test 1
   Test 2
   Test 3
          Tests of Interest

-2*log(Likelihood Ratio)  Test df
            33.8465
            2.81248
             11.448
p-value

  <.0001
  0.5897
0.009534
The p-value for Test 1 is less than  .05.  There appears
to be a
difference between response and/or variances among the
dose levels.
It seems appropriate to model the data

The p-value for Test 2 is greater than  .05.  A
homogeneous variance
model appears to be appropriate here
                                      A-4

-------
The p-value for Test 3 is less  than  .05.   You may want

to try a

different model

 Benchmark Dose Computation

Specified effect =              1
Risk Type
                  Estimated  standard deviations from the control mean
Confidence level =



             BMD =
                        0.95



                     106.072
            BMDL =
                          76.5569
                           Linear Model with 0.95 Confidence Level
     1.9
  o>
  V)
  c
  o
  Q.
  v)
  CD
  o:

  c
  CD
1.7




1.6




1.5




1.4
           Linear
                            BMD
                                    BMP
                          50
                                 100          150


                                     dose
   22:0405/122005
200
250
                                      A-5

-------
        Polynomial Model. Revision: 2.2  Date: 9/12/2002
        Input Data File: C:\BMDS\DATA\CYLINDRO.(d)
        Gnuplot Plotting File:  C:\BMDS\DATA\CYLINDRO.plt
                                          Thu May 12 22:27:06 2005
 HMDS MODEL RUN


   The form of the response function is:

   Y[dose]  = beta 0 + beta l*dose + beta 2*doseA2 + ...
   Dependent variable = MEAN
   Independent variable = dose
   rho is set to 0
   The polynomial coefficients are restricted to be positive
   A constant variance model is fit

   Total number of dose groups = 5
   Total number of records with missing values = 0
   Maximum number of iterations = 250
   Relative Function Convergence has been set to: le-008
   Parameter Convergence has been set to: le-008
                  Default Initial Parameter Values
                          alpha =
                            rho =
                         beta_0 =
                         beta_l =
                         beta 2 =
                          0.018741
                                 0
                           1.46523
                                 0
                                 0
                     Specified
                                 Parameter Estimates
   Variable Estimate
   alpha    0.0215477
   beta 0   1.54205
   beta_l
   beta 2
            Std.  Err.
            0.00459399
            0.0311925
0.00138389  0.000287871
0
                        NA
               95.0% Wald Confidence Interval
          Lower Conf. Limit   Upper Conf. Limit
          0.0125437           0.0305518
          1.48091             1.60318
          0.000819671
                  0.00194811
NA - Indicates that this parameter has hit a bound
     implied by some inequality constraint and thus
     has no standard error.
           Asymptotic Correlation Matrix of Parameter Estimates
     alpha
    beta_0
    beta 1
      alpha
          1
   1.4e-010
   7.2e-011
  beta_0
1.4e-010
       1
    -0.7
  beta_l
7.2e-011
    -0.7
       1
                                     A-6

-------
The following parameter(s) have been estimated at a
boundary
point or have been specified.  Correlations are not
computed:

beta 2
     Table of Data and Estimated Values of Interest
Dose
Res .
0
30
60
120
240
N
10
10
9
9
6
Obs
1.
1.
1.
1.
1.
Mean
48
57
66
82
78
Obs

0
0
0
0
Std Dev
0.1
.14
.16
.12
.17
Est
1.
1.
1.
1.
1.
Mean
54
58
63
71
87
Est
0.
0.
0.
0.
0.
Std Dev
147
147
147
147
147
ChiA2
-1
-0.
0.
2
-1

.34
292
714
.29
.57
  Model Descriptions for likelihoods calculated
 Model Al:        Yij
           Var{e(ij)}

 Model A2:        Yij
           Var{e(ij) }

 Model  R:         Yi
            Var{e(i) }
             Mu (i)  + e (ij )
             S i gma A 2

             Mu (i)  + e (ij )
             Sigma(i)A2

             Mu + e (i)
             SigmaA2
                       Likelihoods of Interest
Model Log (likelihood)




Test 1:
levels

Test 2:
Test 3:
Al
A2
fitted
R
Does response

(A2 vs. R)
Are Variances
Does the Model
68.148702
69.554943
62.424690
52.631671
and/or variances


DF
6
10
2
2
differ


AIC
124.297404
119.109885
120.849381
101.263343
among dose


Homogeneous (Al vs A2 )
for the Mean Fit
(Al vs
. fitted)
                     Tests of Interest
   Test

   Test 1
   Test 2
   Test 3
-2*log(Likelihood Ratio)  Test df
            33.8465
            2.81248
             11.448
p-value

  <.0001
  0.5897
0.003267
                                      A-7

-------
The p-value for Test 1 is less than .05.  There appears
to be a
difference between response and/or variances among the
dose levels.
It seems appropriate to model the data

The p-value for Test 2 is greater than  .05.  A
homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is less than .05.  You may want
to try a
different model
 Benchmark Dose Computation
Specified effect =             1

Risk Type        =     Estimated standard deviations from the control mean
Confidence level =          0.95

             BMD =       106.072


            BMDL =       76.5569
                                      A-8

-------
                        Polynomial Model with 0.95 Confidence Level
     1.9
 o>
 V)
 c
 o
 Q.
 w
 CD
 o:
 c
 CD
1.7


1.6


1.5


1.4
           Polynomial
                           BMD
                                   BMP
                         50
                                100
150
200
250
                                         dose
   22:2705/122005
       Power Model. $Revision: 2.1 $ $Date: 2000/10/11 20:57:36  $
       Input Data File: C:\BMDS\DATA\CYLINDRO.(d)
       Gnuplot Plotting File:  C:\BMDS\DATA\CYLINDRO.plt
                                         Thu May 12 22:28:38 2005
HMDS MODEL RUN


  The form of the response function is:

  Y[dose]  = control + slope * doseApower
  Dependent variable = MEAN
  Independent variable = dose
  rho is set to 0
  The power is restricted to be greater than or equal to  1
  A constant variance model is fit

  Total number of dose groups = 5
  Total number of records with missing values = 0
  Maximum number of iterations = 250
  Relative Function Convergence has been set to: le-008
  Parameter Convergence has been set to: le-008
                                     A-9

-------
                 Default Initial Parameter Values
                         alpha =      0.018741

Asymptotic
alpha
alpha 1
rho -1
control NA
slope NA
power NA
NA - This parameter's

Variable
alpha
rho
control
slope
power
Table of Data and
Dose N Obs M
Res .
0 10 1.48
30 10 1.57
60 9 1.66
120 9 1.82
240 6 1.78
rho = 0 Specified
control = 1.48
slope = 0.0132605
power = 0. 612843
Correlation Matrix of Parameter Estimates
rho control slope
- 1 NA NA
1 NA NA
NA NA NA
NA NA NA
NA NA NA
variance has been estimated at zero.
Parameter Estimates
Estimate Std. Err.
0.0215477 0.0399798
0 3.81654
1.54205 0.0380886
0.00138389 0.00299534
1 0.337065
Estimated Values of Interest
ean Obs Std Dev Est Mean Est Std Dev
0.1 1.54 0.147
0.14 1.58 0.147
0.16 1.63 0.147
0.12 1.71 0.147
0.17 1.87 0.147


power
NA
NA
NA
NA
NA




ChiA2
-0.423
-0.0924
0.238
0.762
-0.642
     Model Descriptions for likelihoods  calculated


Model Al:        Yij = Mu(i) + e(ij)
          Var{e(ij)} = SigmaA2
                                    A-10

-------
 Model A2:        Yij = Mu(i) + e(ij)
           Var{e(ij)} = Sigma(i)A2

 Model  R:         Yi = Mu + e(i)
            Var{e(i)} = SigmaA2
                       Likelihoods of Interest

            Model      Log(likelihood)   DF        AIC
             Al           68.148702       6    -124.297404
             A2           69.554943      10    -119.109885
           fitted         62.424690       4    -116.849381
              R           52.631671       2    -101.263343

 Test 1:  Does response and/or variances differ among dose levels
          (A2 vs. R)
 Test 2:  Are Variances Homogeneous  (Al vs A2)
 Test 3:  Does the Model for the Mean Fit (Al vs. fitted)

                     Tests of Interest

   Test    -2*log(Likelihood Ratio)     df        p-value

   Test 1              33.8465          8         <.00001
   Test 2              2.81248          4          0.5897
   Test 3               11.448          2        0.003267

The p-value for Test 1 is less than  .05.  There appears to be a
difference between response and/or variances among the dose levels.
It seems appropriate to model the data

The p-value for Test 2 is greater than  .05.   A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is less than  .05.  You may want to try a
different model
 Benchmark Dose Computation
Specified effect =             1

Risk Type        =     Estimated standard deviations from the control mean

Confidence level =          0.95

             BMD =       106.072


            BMDL =       76.5569
                                     A-ll

-------
                          Power Model with 0.95 Confidence Level
     1.9
     1.8
 CD
 V)
 c
 o   17
 Q.  I-'
 CD
 o:
 ro   1.6
 0
     1.5
     1.4
         Power
                           BMD
   BMD
                         50
100         150
    dose
200
250
   22:2805/122005
       Hill Model. $Revision: 2.1 $ $Date: 2000/10/11 21:21:23 $
       Input Data File: C:\BMDS\DATA\CYLINDRO.(d)
       Gnuplot Plotting File:  C:\BMDS\DATA\CYLINDRO.plt
                                         Thu May 12 22:29:31 2005
HMDS MODEL RUN


  The form of the response function is:

  Y[dose]  = intercept + v*doseAn/(kAn + doseAn)
  Dependent variable = MEAN
  Independent variable = dose
  rho is set to 0
  Power parameter restricted to be greater than 1
  A constant variance model is fit

  Total number of dose groups = 5
  Total number of records with missing values = 0
  Maximum number of iterations = 250
  Relative Function Convergence has been set to: le-008
  Parameter Convergence has been set to: le-008
                                    A-12

-------
                 Default Initial  Parameter Values
alpha = 0.0183633
rho = 0 Specified
intercept = 1.48
v = 0.34
n = 0.959904
k = 63.3333
Asymptotic Correlation Matrix of Parameter Estimates
alpha rho intercept v n
alpha 10000
rho 01000
intercept 00100
v 0 0 0 1 0
n 0 0 0 0 1
k 0 0 0 0 0
Parameter Estimates
Variable Estimate Std. Err.
alpha 0.0172084 1
rho 0 1
intercept 1.4859 1
v 0.33478 1
n 2.46401 1
k 51.624 1
Table of Data and Estimated Values of Interest
Dose N Obs Mean Obs Std Dev Est Mean Est Std Dev
Res .
0 10 1.48 0.1 1.49 0.131
30 10 1.57 0.14 1.56 0.131
60 9 1.66 0.16 1.68 0.131
120 9 1.82 0.12 1.78 0.131
240 6 1.78 0.17 1.81 0.131




k
0
0
0
0
0
1







ChiA2

-0.0449
0.11
-0.183
0.279
-0.253
Model Descriptions for likelihoods  calculated

Model Al:        Yij = Mu(i) + e(ij)
          Var{e(ij)} = SigmaA2
                                    A-13

-------
 Model A2:        Yij = Mu(i) + e(ij)
           Var{e(ij)} = Sigma(i)A2

 Model  R:         Yi = Mu + e(i)
            Var{e (i) } = SigmaA2
                       Likelihoods of Interest

            Model      Log(likelihood)   DF        AIC
             Al           68.148702       6    -124.297404
             A2           69.554943      10    -119.109885
           fitted         67.371885       5    -124.743770
              R           52.631671       2    -101.263343

 Test 1:  Does response and/or variances differ among dose levels
          (A2 vs. R)
 Test 2:  Are Variances Homogeneous  (Al vs A2)
 Test 3:  Does the Model for the Mean Fit (Al vs. fitted)

                     Tests of Interest

   Test    -2*log(Likelihood Ratio)  Test df     p-value

   Test 1              33.8465          8          <.0001
   Test 2              2.81248          4          0.5897
   Test 3              1.55363          1          0.2126

The p-value for Test 1 is less than  .05.  There appears to be a
difference between response and/or variances among the dose levels.
It seems appropriate to model the data

The p-value for Test 2 is greater than  .05.   A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than  .05.  The model chosen appears
to adequately describe the data
 Benchmark Dose Computation
Specified effect =             1

Risk Type        =     Estimated standard deviations from the control mean

Confidence level =          0.95

             BMD =       43.1891

Warning:  optimum may not have been found.  Bad completion code in
Optimization routine.

BMDL computation failed.
                                     A-14

-------
     2




    1.9
                                        Hill Model
       Hill
o>
V)
c
o
Q.
w
CD
o:

c
CD
1.7




1.6




1.5




1.4
                         BMP
  22:2905/122005
                          50
                                  100          150


                                       dose
200
250
                                     A-15

-------
Part II.




Humpage and Falconer 2003




male mice treated with purified cylindrospermopsin




rel kidney wt




drop high dose group
                                     A-16

-------
       Polynomial Model. Revision: 2.2  Date: 9/12/2002
       Input Data File: C:\BMDS\DATA\CYLINDRO.(d)
       Gnuplot Plotting File:  C:\BMDS\DATA\CYLINDRO.plt
                                         Thu May 12 22:31:50 2005
HMDS MODEL RUN


  The form of the response function is:

  Y[dose]  = beta 0 + beta l*dose + beta 2*doseA2 + ...
  Dependent variable = MEAN
  Independent variable = dose
  rho is set to 0
  The polynomial coefficients are restricted to be positive
  A constant variance model is fit

  Total number of dose groups = 4
  Total number of records with missing values = 0
  Maximum number of iterations = 250
  Relative Function Convergence has been set to: le-008
  Parameter Convergence has been set to: le-008
                 Default Initial Parameter Values
                         alpha =    0.0172471
                           rho =            0   Specified
                        beta_0 =        1.484
                        beta 1 =   0.00282857
                                Parameter Estimates
  Variable  Estimate
  alpha     0.0154483
  beta_0    1.4838
  beta 1    0.00283099
                            95.0% Wald Confidence Interval
            Std. Err.     Lower Conf.  Limit   Upper Conf. Limit
            0.00354407        0.008502             0.0223945
            0.0306522         1.42373              1.54388
            0.000456935       0.00193541           0.00372656
          Asymptotic Correlation Matrix of Parameter Estimates
    alpha
   beta_0
   beta 1
    alpha
        1
-1.7e-010
 1.4e-010
   beta_0
-1.7e-010
        1
    -0.75
  beta_l
1.4e-010
   -0.75
       1
                                    A-17

-------
     Table of Data and Estimated Values of Interest
Dose
Res .
0
30
60
120
N
10
10
9
9
Obs
1.
1.
1.
1.
Mean
48
57
66
82
Obs

0
0
0
Std Dev
0.1
.14
.16
.12
Est
1.
1.
1.
1.
Mean
48
57
65
82
Est
0.
0.
0.
0.
Std Dev
124
124
124
124
Chi
-0
0

-
A2
.0968
.0323
0.153
0.085
  Model Descriptions for likelihoods calculated
 Model Al:        Yij
           Var{e(ij)}

 Model A2:        Yij
           Var{e(ij)}

 Model  R:         Yi
            Var{e(i)}
             Mu (i) + e (ij )
             SigmaA2

             Mu (i) + e (ij )
             Sigma(i)A2

             Mu + e (i)
             S i gma A 2
                       Likelihoods of Interest
Model
Al
A2
fitted
R
Log (likelihood)
60.255446
61.376237
60.234922
46.462327
DF
5
8
2
2
AIC
-110.510893
-106.752474
-116.469843
-88.924654
 Test 1:  Does response and/or variances differ among dose
levels
          (A2 vs. R)
 Test 2:  Are Variances Homogeneous  (Al vs A2)
 Test 3:  Does the Model for the Mean Fit  (Al vs. fitted)
   Test

   Test 1
   Test 2
   Test 3
          Tests of Interest

-2*log(Likelihood Ratio)  Test df
            29.8278
            2.24158
          0.0410496
6
3
2
p-value

  <.0001
  0.5238
  0.9797
The p-value for Test 1 is less than  .05.  There appears
to be a
difference between response and/or variances among the
dose levels.
It seems appropriate to model the data
                                     A-18

-------
The p-value for Test 2 is greater  than  .05.   A
homogeneous variance
model appears to be appropriate  here
The p-value for Test 3 is greater  than  .05.   The model
chosen appears
to adequately describe the data
 Benchmark Dose Computation
Specified effect =              1
Risk Type
                    Estimated standard deviations  from  the  control  mean
Confidence level =

             BMD =
                         0.95

                      43.9038
            BMDL =
                          33.0684
                            Linear Model with 0.95 Confidence Level
       1.9
       1.8
   CD
   V)
   c
   O
   Q.
   V)
   CD
c
CD
CD
       -Ig
       ' -°
       1.5
       1.4
             Linear
                           BMDL     iBM.D
     22:31 05/122005
20        40        60
                   dose
                                                      80
                                                            100
120
                                      A-19

-------
        Polynomial Model. Revision: 2.2  Date: 9/12/2002
        Input Data File: C:\BMDS\DATA\CYLINDRO.(d)
        Gnuplot Plotting File:  C:\BMDS\DATA\CYLINDRO.plt
                                          Thu May 12 22:33:17 2005
 HMDS MODEL RUN


   The form of the response function is:

   Y[dose]  = beta 0 + beta l*dose + beta 2*doseA2 + ...
   Dependent variable = MEAN
   Independent variable = dose
   rho is set to 0
   The polynomial coefficients are restricted to be positive
   A constant variance model is fit

   Total number of dose groups = 4
   Total number of records with missing values = 0
   Maximum number of iterations = 250
   Relative Function Convergence has been set to: le-008
   Parameter Convergence has been set to: le-008
                  Default Initial Parameter Values
                          alpha =
                            rho =
                         beta_0 =
                         beta_l =
                         beta 2 =
                      0.0172471
                              0
                        1.47945
                     0.00314242
                              0
                     Specified
                                 Parameter Estimates
   Variable
   alpha
   beta_0
   beta_l
   beta 2
                        95.0% Wald Confidence Interval
Estimate    Std.  Err.     Lower Conf. Limit  Upper Conf. Limit
0.0154483   0.00354407     0.008502         0.0223945
1.4838      0.0306522       1.42373          1.54388
0.00283099  0.000456935
              0.00193541
0
                           NA
                  0.00372656
NA - Indicates that this parameter has hit a bound
     implied by some inequality constraint and thus
     has no standard error.
           Asymptotic Correlation Matrix of Parameter Estimates
     alpha
    beta_0
    beta 1
   alpha
       1
4.9e-011
9.5e-011
  beta_0
4.9e-011
       1
   -0.75
  beta_l
9.5e-011
   -0.75
       1
                                     A-20

-------
The following parameter(s) have been estimated at a
boundary
point or have been specified.  Correlations are not
computed:

beta 2
 Dose
Res .
     Table of Data and Estimated Values of Interest

            N    Obs Mean    Obs Std Dev   Est Mean
    0
   30
   60
  120
10
10
 9
 9
                                             Est Std Dev
                                                  ChiA2
1.48
1.57
1.66
1.82

0
0
0
0.1
.14
.16
.12
1.
1.
1.
1.
48
57
65
82
0.
0.
0.
0.
124
124
124
124
-0
0

-

.
0
0
0968
0323
.153
.085
  Model Descriptions for likelihoods calculated
 Model Al:        Yij
           Var{e(ij)}

 Model A2:        Yij
           Var{e(ij)}

 Model  R:         Yi
            Var{e(i)}
               Mu (i)  + e (ij )
               SigmaA2

               Mu (i)  + e (ij )
               Sigma(i)A2

               Mu + e (i)
               S i gma A 2
                       Likelihoods of Interest
Model Log (likelihood)




Test 1:
levels

Test 2:
Test 3:
Al
A2
fitted
R
Does response

(A2 vs. R)
Are Variances
Does the Model
60.255446
61.376237
60.234922
46.462327
and/or variances


DF
5
8
2
2
differ


AIC
110.510893
106.752474
116.469843
-88.924654
among dose


Homogeneous (Al vs A2 )
for the Mean Fit
(Al vs
. fitted)
                                     A-21

-------
                     Tests of Interest

   Test    -2*log(Likelihood Ratio)  Test df     p-value

   Test 1              29.8278          6          <.0001
   Test 2              2.24158          3          0.5238
   Test 3            0.0410496          1          0.8394

The p-value for Test 1 is less than .05.  There appears
to be a
difference between response and/or variances among the
dose levels.
It seems appropriate to model the data

The p-value for Test 2 is greater than  .05.  A
homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .05.  The model
chosen appears
to adequately describe the data
 Benchmark Dose Computation
Specified effect =             1

Risk Type        =     Estimated standard deviations from the control mean
Confidence level =          0.95

             BMD =       43.9038


            BMDL =       33.0684
                                     A-22

-------
                        Polynomial Model with 0.95 Confidence Level

1.9
1.8
n Response
^j
eg 1-6
^
1.5
1.4


22:33 (
- ' ' _ ' 	 .' ' 	 -
polynomial
: ^~^-
\ " / 	 1
; T ..>--^ :
: ^T\^ 1 ;
^ 1 1 -L -
i ^ i ;
; 1 ;
'-- 	 , 	 , BMDL 	 iBM.P. , , , 	 , 	 , 	 , 	 :
0 20 40 60 80 100 120
dose
D5/122005
       Power Model. $Revision: 2.1 $ $Date: 2000/10/11 20:57:36 $
       Input Data File: C:\BMDS\DATA\CYLINDRO.(d)
       Gnuplot Plotting File:  C:\BMDS\DATA\CYLINDRO.plt
                                         Thu May 12 22:34:08 2005
HMDS MODEL RUN


  The form of the response function is:

  Y[dose]  = control + slope * doseApower
  Dependent variable = MEAN
  Independent variable = dose
  rho is set to 0
  The power is restricted to be greater than or equal to 1
  A constant variance model is fit

  Total number of dose groups = 4
  Total number of records with missing values = 0
  Maximum number of iterations = 250
  Relative Function Convergence has been set to: le-008
  Parameter Convergence has been set to: le-008
                                    A-23

-------
                 Default  Initial  Parameter Values
                          alpha  =     0.0172471
rho = 0 Specified
control = 1.48
slope = 0.00345163
power = 0. 958769
Asymptotic Correlation Matrix of Parameter Estimates
alpha rho control slope
alpha 1 -0.99 0.25 -0.29
rho -0.99 1 -0.25 0.29
control 0.25 -0.25 1 -0.71
slope -0.29 0.29 -0.71 1
power 0.29 -0.28 0.66 -1
Parameter Estimates
Variable Estimate Std. Err.
alpha 0.0154483 0.0279175
rho 0 3.65432
control 1.4838 0.0410106
slope 0.00283099 0.00517499
power 1 0.372848
Table of Data and Estimated Values of Interest
Dose N Obs Mean Obs Std Dev Est Mean Est Std Dev
Res .
0 10 1.48 0.1 1.48 0.124
30 10 1.57 0.14 1.57 0.124
60 9 1.66 0.16 1.65 0.124
120 9 1.82 0.12 1.82 0.124


power
0.29
-0.28
0.66
-1
1



ChiA2
-0.0306
0.0102
0.051
-0.0283
     Model Descriptions  for  likelihoods  calculated
Model Al:        Yij
          Var{e(ij)}

Model A2:        Yij
          Var{e(ij)}
Mu (i) + e (ij )
SigmaA2

Mu (i) + e (ij )
Sigma(i)A2
                                     A-24

-------
 Model  R:         Yi = Mu + e(i)
            Var{e(i)} = SigmaA2
                       Likelihoods of Interest
            Model
             Al
             A2
           fitted
              R
             Log(likelihood)   DF
                60.255446       5
                61.376237       8
                60.234922       4
                46.462327       2
    AIC
-110.510893
-106.752474
-112.469843
 -88.924654
 Test 1:

 Test 2:
 Test 3:
Does response and/or variances differ among dose levels
(A2 vs. R)
Are Variances Homogeneous  (Al vs A2)
Does the Model for the Mean Fit  (Al vs. fitted)
                     Tests of Interest

   Test    -2*log(Likelihood Ratio)     df

   Test 1              29.8278          6
   Test 2              2.24158          3
   Test 3            0.0410496          1
                                        p-value

                                        <.00001
                                         0.5238
                                         0.8394
The p-value for Test 1 is less than .05.  There appears to be a
difference between response and/or variances among the dose levels.
It seems appropriate to model the data

The p-value for Test 2 is greater than  .05.  A homogeneous variance
model appears to be appropriate here
The p-value for Test 3 is greater than .05.  The model chosen appears
to adequately describe the data
 Benchmark Dose Computation
Specified effect =
Risk Type

Confidence level =

             BMD =
             Estimated standard deviations from the control mean

                  0.95

               43.9038
            BMDL =
                         33.0684
                                     A-25

-------
                          Power Model with 0.95 Confidence Level

1.9
1.8
CD
V)
c
o 17
Q. I-'
CD
o:
C! . —
«5 1.6
CD
S
1.5
1.4



22:34 (
: _ 	 :
rower
[ ~r j
i ^^1
; T ^^-^ 1 ;
; T ^-^ \
'-- ^ /-^r "" ^ ^
; ^^ 1 ;
\ 1 ^
; 	 , 	 , BMDL 	 IBMD. , , , 	 , 	 , 	 , 	 ;
0 20 40 60 80 100 120
dose
D5/122005
       Hill Model. $Revision: 2.1 $ $Date: 2000/10/11 21:21:23 $
       Input Data File: C:\BMDS\DATA\CYLINDRO.(d)
       Gnuplot Plotting File:  C:\BMDS\DATA\CYLINDRO.plt
                                         Thu May 12 22:34:43 2005
HMDS MODEL RUN


  The form of the response function is:

  Y[dose]  = intercept + v*doseAn/(kAn + doseAn)
  Dependent variable = MEAN
  Independent variable = dose
  rho is set to 0
  Power parameter restricted to be greater than 1
  A constant variance model is fit

  Total number of dose groups = 4
  Total number of records with missing values = 0
  Maximum number of iterations = 250
  Relative Function Convergence has been set to: le-008
  Parameter Convergence has been set to: le-008
                                    A-26

-------
                 Default Initial Parameter Values
                         alpha =
                           rho =
                     intercept =
                             v =
                             n =
                             k =
0.0163951
        0
     1.48
     0.34
  0.68364
  63.3333
Specified
          Asymptotic Correlation Matrix of Parameter Estimates

alpha
rho
intercept
V
n
k
alpha
1
0
0
0
0
0
rho
0
1
0
0
0
0
intercept
0
0
1
0
0
0
V
0
0
0
1
0
0
n
0
0
0
0
1
0
k
0
0
0
0
0
1
                         Parameter Estimates
Variable
alpha
rho
intercept
V
n
k
Estimate
0.0154316
0
1.48
1.62004
1.08746
406.089
Std. Err
1
1
1
1
1
1
    Table of Data and Estimated Values of Interest
R
-



Dose
es .
0
30
60
120
N
10
10
9
9
Obs
1.
1.
1.
1.
Mean
48
57
66
82
Obs

0
0
0
Std Dev
0.1
.14
.16
.12
Est
1.
1.
1.
1.
Mean
48
57
66
82
Est
0.
0.
0.
0.
Std
124
124
124
124
Dev ChiA2
-2.33e-007
-4.42e-007
8.2e-007
-1.25e-008
Model Descriptions for likelihoods calculated

Model Al:        Yij = Mu(i) + e(ij)
          Var{e(ij)} = SigmaA2
                                    A-27

-------
 Model A2:        Yij = Mu(i) + e(ij)
           Var{e(ij)} = Sigma(i)A2

 Model  R:         Yi = Mu + e(i)
            Var{e (i) } = SigmaA2
 Warning: Likelihood for fitted model larger than the Likelihood for model
Al.

                       Likelihoods of Interest

            Model      Log(likelihood)   DF        AIC
             Al           60.255446       5    -110.510893
             A2           61.376237       8    -106.752474
           fitted         60.255447       5    -110.510893
              R           46.462327       2     -88.924654

 Test 1:  Does response and/or variances differ among dose levels
          (A2 vs. R)
 Test 2:  Are Variances Homogeneous  (Al vs A2)
 Test 3:  Does the Model for the Mean Fit (Al vs. fitted)

                     Tests of Interest

   Test    -2*log(Likelihood Ratio)  Test df     p-value

   Test 1              29.8278          6          <.0001
   Test 2              2.24158          3          0.5238
   Test 3        -4.58279e-007          0              NA

The p-value for Test 1 is less than  .05.  There appears to be a
difference between response and/or variances among the dose levels.
It seems appropriate to model the data

The p-value for Test 2 is greater than  .05.   A homogeneous variance
model appears to be appropriate here


NA - Degrees of freedom for Test 3 are  less than or equal to 0.  The Chi-
Square
     test for fit is not valid
 Benchmark Dose Computation
Specified effect =             1

Risk Type        =     Estimated standard deviations from the control mean

Confidence level =          0.95

             BMD =       41.1966

            BMDL =        21.722
                                     A-28

-------
                            Hill Model with 0.95 Confidence Level
           Hill
CD
V)
c
CD
o:
    1.9
    1.8
    * -,
    1.7
    1
    ' -
    1.5
    1.4
                    BMD
  BMD
                       20
  22:3405/122005
40
 60

dose
80
100
120
                                     A-29

-------
Part III.




Humpage and Falconer 2003




male mice treated with purified cylindrospermopsin




urinary protein levels
                                     A-30

-------
       Polynomial Model.  Revision:  2.2  Date:  9/12/2002
       Input Data File:  C:\BMDS\DATA\CYLINDRO.(d)
       Gnuplot Plotting File:   C:\BMDS\DATA\CYLINDRO.plt
                                         Thu May 12 23:15:09 2005
HMDS MODEL RUN


  The form of the response function is:

  Y[dose]  = beta 0 + beta l*dose + beta 2*doseA2 + ...
  Dependent variable = MEAN
  Independent variable = dose
  rho is set to 0
  Signs of the polynomial coefficients are not restricted
  A constant variance model is fit

  Total number of dose groups = 5
  Total number of records with missing values = 0
  Maximum number of iterations = 250
  Relative Function Convergence has been set to:  le-008
  Parameter Convergence has been set to: le-008
                 Default Initial Parameter Values
                         alpha =    0.0467949
                           rho =            0   Specified
                        beta_0 =      4.00125
                        beta 1 =   -0.0114583
                                Parameter Estimates
  Variable  Estimate
  alpha     0.109717
  beta 0    4.03293
  beta 1
            Std.  Err.
            0.0233918
            0.070386
             95.0% Wald Confidence Interval
            Lower Conf.  Limit    Upper Conf.  Limit
                 0.0638701             0.155564
                 3.89498               4.17088
-0.0120086  0.000649584
                -0.0132818
                        -0.0107354
          Asymptotic Correlation Matrix of Parameter Estimates
    alpha
   beta_0
   beta 1
     alpha
         1
  2.5e-009
  9.1e-010
  beta_0
2.5e-009
       1
    -0.7
  beta_l
9.1e-010
    -0.7
       1
                                    A-31

-------
     Table of Data and Estimated Values of Interest
Dose
Res .
0
30
60
120
240
N
10
10
9
9
6
Obs
4.
3
3.
2.
1
Mean
25
.7
25
15
.5
Obs

0
0

0
Std Dev
0.3
.25
.05
0.2
.15
Est
4.
3.
3.
2.
1.
Mean
03
67
31
59
15
Est
0.
0.
0.
0.
0.
Std Dev
331
331
331
331
331
ChiA2
2
0.
-0.

2

.07
261
565
-4
.58
  Model Descriptions for likelihoods calculated
 Model Al:        Yij = Mu(i) + e(ij)
           Var{e(ij)} = SigmaA2

 Model A2:        Yij = Mu(i) + e(ij)
           Var{e(ij)} = Sigma(i)A2

 Model  R:         Yi = Mu + e(i)
            Var{e(i)} = SigmaA2
                       Likelihoods of Interest

            Model      Log(likelihood)   DF        AIC
             Al           48.017412       6     -84.034824
             A2           59.392540      10     -98.785081
           fitted         26.616696       2     -49.233392
              R          -21.651321       2      47.302642

 Test 1:  Does response and/or variances differ among dose
levels
          (A2 vs. R)
 Test 2:  Are Variances Homogeneous  (Al vs A2)
 Test 3:  Does the Model for the Mean Fit (Al vs. fitted)
   Test

   Test 1
   Test 2
   Test 3
          Tests of Interest

-2*log(Likelihood Ratio)  Test df
            162.1
            22.7503
            42.8014
p-value

  <.0001
0.000142
  <.0001
The p-value for Test 1 is less than  .05.  There appears
to be a
difference between response and/or variances among the
dose levels.
It seems appropriate to model the data
                                     A-32

-------
The p-value for Test 2 is less than  .05.
running a
non-homogeneous variance model

The p-value for Test 3 is less than  .05.
to try a
different model
                                      Consider
                                      You may  want
 Benchmark Dose Computation
Specified effect =             1
Risk Type
                  Estimated standard  deviations  from the control mean
Confidence level =

             BMD =
                       0.95

                    27.5832
            BMDL =
                         22.9758
                           Linear Model with 0.95 Confidence Level
     4.5
  CD
  V)
  c
  o
  Q.
  V)
  CD
  c
  CD
  CD
3.5
2.5
     1.5
           Linear
               BMDL
                BMD
              0           50

   23:1505/122005
                                 100          150         200
                                     dose
250
                                     A-33

-------
       Polynomial Model.  Revision:  2.2  Date: 9/12/2002
       Input Data File:  C:\BMDS\DATA\CYLINDRO.(d)
       Gnuplot Plotting File:   C:\BMDS\DATA\CYLINDRO.plt
                                         Thu May 12 23:16:16 2005
HMDS MODEL RUN


  The form of the response function is:

  Y[dose]  = beta 0 + beta l*dose + beta 2*doseA2 + ...
  Dependent variable = MEAN
  Independent variable = dose
  Signs of the polynomial coefficients are not restricted
  The variance is to be modeled as Var(i)  = alpha*mean(i)Arho

  Total number of dose groups = 5
  Total number of records with missing values = 0
  Maximum number of iterations = 250
  Relative Function Convergence has been set to:  le-008
  Parameter Convergence has been set to: le-008
                 Default Initial Parameter Values
                         alpha =    0.0467949
                           rho =            0
                        beta_0 =      4.00125
                        beta 1 =   -0.0114583
                                Parameter Estimates

                                       95.0% Wald Confidence Interval
  Variable   Estimate      Std.  Err.    Lower  Conf.  Limit   Upper Conf. Limit
  alpha      0.0612302     0.0136561       0.0344646        0.0879957
  rho        -0.214753     0.0359694       -0.285252        -0.144255
  beta_0     4.26301       0.0535233       4.15811          4.36791
  beta 1     -0.0177625    0.00084855      -0.0194257       -0.0160994
          Asymptotic Correlation Matrix of Parameter Estimates

                 alpha          rho       beta_0       beta_l
    alpha            1         0.12         O.Il        -0.18
      rho         0.12            1        -0.18         0.29
   beta_0         0.11        -0.18            1        -0.75
   beta 1        -0.18         0.29        -0.75            1
                                    A-34

-------
 Dose
Res .
     Table of Data and Estimated Values of Interest

            N    Obs Mean    Obs Std Dev   Est Mean   Est Std Dev
                                                           ChiA2
    0
   30
   60
  120
  240
10
10
 9
 9
 6
4.25
 3.7
3.25
2.15
 1.5
 0.3
0.25
0.05
 0.2
0.15
     4.26
     3.73
      3.2
     2.13
4.95e-008
0.212
0.215
0.218
0.228
 1.51
-0.194
-0.444
 0.724
 0.243
  2.44
 Model Descriptions for likelihoods calculated
 Model Al:        Yij
           Var{e(ij)}

 Model A2:        Yij
           Var{e(ij)}

 Model A3:        Yij
           Var{e(ij)}

 Model  R:         Yi
            Var{e(i)}
               Mu (i)  + e (ij )
               S i gma A 2

               Mu (i)  + e (ij )
               Sigma(i)A2

               Mu (i)  + e (ij )
               alpha*(Mu(i))Arho

               Mu + e(i)
               SigmaA2
                       Likelihoods of Interest

            Model      Log(likelihood)   DF        AIC
             Al           48.017412       6     -84.034824
             A2           59.392540      10     -98.785081
             A3           49.799962       7     -85.599924
           fitted         33.618701       4     -59.237401
              R          -21.651321       2      47.302642
 Test 1:
levels?

 Test 2:
 Test 3:
 Test 4:
          Explanation of Tests

 Does response and/or variances differ among Dose

 (A2 vs.  R)
 Are Variances Homogeneous? (Al vs A2)
 Are variances adequately modeled?  (A2 vs. A3)
 Does the Model for the Mean Fit?  (A3 vs. fitted)
            Tests of Interest
   Test

   Test 1
   Test 2
   Test 3
   Test 4
  -2*log(Likelihood Ratio)   Test df        p-value
              162.1
              22.7503
              19.1852
              32.3625
                      8          <.0001
                      4        0.000142
                      3       0.0002503
                      3          <.0001
                                     A-35

-------
The p-value for Test 1 is less than .05.  There appears
to be a
difference between response and/or variances among the
dose levels
It seems appropriate to model the data

The p-value for Test 2 is less than .05.  A
non-homogeneous variance
model appears to be appropriate

The p-value for Test 3 is less than .05.  You may want
to consider a
different variance model

The p-value for Test 4 is less than .05.  You may want
to try a different
model
 Benchmark Dose Computation
Specified effect =             1

Risk Type        =     Estimated standard deviations from the control mean
Confidence level =          0.95

             BMD =       11. 9223


BMDL computation failed.
                                     A-36

-------
                          Linear Model with 0.95 Confidence Level
    4.5
 CD
 V)
 c
 o
 Q.
 
-------
                  Default Initial Parameter Values
                          alpha
                            rho
                         beta_0
                         beta_l
                         beta_2
                         beta 3
                            0.0467949
                                    0
                              4.23439
                           -0.0148331
                         -5.51541e-005
                         2.89474e-007
                               Specified
                                 Parameter Estimates
 Variable  Estimate
 alpha
 beta_0
 beta_l
 beta_2
 beta 3
  0.0424211
  4.23509
  -0.0149779
  -5.33342e-005
  2.84322e-007
                             95.0% Wald Confidence Interval
                  Std.  Err.     Lower Conf. Limit   Upper Conf. Limit
         0.0090442
         0.0633648
         0.00335279
         4.06738e-005
         1.17673e-007
              0.0246948
              4.1109
              -0.0215492
              -0.000133053
              5.36875e-008
                     0.0601474
                     4.35929
                     -0.00840654
                     2.63851e-005
                     5.14956e-007
           Asymptotic Correlation Matrix of Parameter Estimates

alpha
beta 0
beta 1
beta 2
beta 3
alpha
1
-1.8e-008
-4.3e-009
1.3e-009
1.5e-009
beta 0
-1.8e-008
1
-0.71
0.55
-0.48
beta 1
-4.3e-009
-0.71
1
-0.96
0.91
beta 2
1.3e-009
0.55
-0.96
1
-0.99
beta 3
1.5e-009
-0.48
0.91
-0.99
1
 Dose
Res .
     Table of Data and Estimated Values of Interest

            N    Obs Mean    Obs Std Dev   Est Mean   Est Std Dev   ChiA2
    0
   30
   60
  120
  240
10
10
 9
 9
 6
4.25
 3.7
3.25
2.15
 1.5
 0.3
0.25
0.05
 0.2
0.15
4.24
3.75
3.21
2.16
 1.5
0.206
0.206
0.206
0.206
0.206
 0.229
-0.698
 0.643
-0.161
0.0141
  Model Descriptions for likelihoods calculated

 Model Al:         Yij = Mu(i)  + e(ij)
           Var{e(ij)} = SigmaA2
 Model A2:         Yij = Mu(i)  + e(ij)
           Var{e(ij)} = Sigma(i)A2

 Model  R:          Yi = Mu + e(i)
            Var{e (i) } = SigmaA2
                                     A-38

-------
                       Likelihoods of Interest

            Model      Log(likelihood)   DF        AIC
             Al           48.017412       6     -84.034824
             A2           59.392540      10     -98.785081
           fitted         47.522415       4     -87.044831
              R          -21.651321       2      47.302642

 Test 1:  Does response and/or variances differ among dose
levels
          (A2 vs. R)
 Test 2:  Are Variances Homogeneous  (Al vs A2)
 Test 3:  Does the Model for the Mean Fit (Al vs. fitted)

                     Tests of Interest

   Test    -2*log(Likelihood Ratio)  Test df     p-value

   Test 1              162.088          8          <.0001
   Test 2              22.7503          4        0.000142
   Test 3             0.989994          1          0.3197

The p-value for Test 1 is less than  .05.  There appears
to be a
difference between response and/or variances among the
dose levels.
It seems appropriate to model the data

The p-value for Test 2 is less than  .05.  Consider
running a
non-homogeneous variance model

The p-value for Test 3 is greater than  .05.   The model
chosen appears
to adequately describe the data
 Benchmark Dose Computation
Specified effect =             1

Risk Type        =     Estimated standard deviations from the control mean
Confidence level =          0.95

             BMD =       13.1764


            BMDL =        9.7619
                                     A-39

-------
                        Polynomial Model with 0.95 Confidence Level
    4.5
 o
 Q.
 W
 CD
 o:
 c
 CD
 CD
     1.5
           Polynomial
           BMD
BMD
                         50
                    100          150
                        dose
200
250
   23:1705/122005
       Polynomial Model. Revision: 2.2  Date: 9/12/2002
       Input Data File: C:\BMDS\DATA\CYLINDRO.(d)
       Gnuplot Plotting File:  C:\BMDS\DATA\CYLINDRO.plt
                                         Thu May 12 23:18:39 2005
HMDS MODEL RUN


  The form of the response function is:

  Y[dose]  = beta 0 + beta l*dose + beta 2*doseA2 +  ...
  Dependent variable = MEAN
  Independent variable = dose
  rho is set to 0
  The polynomial coefficients are restricted to be negative
  A constant variance model is fit

  Total number of dose groups = 5
  Total number of records with missing values = 0
  Maximum number of iterations = 250
  Relative Function Convergence has been set to: le-008
  Parameter Convergence has been set to: le-008
                                    A-40

-------
                  Default Initial Parameter Values
                          alpha
                            rho
                         beta_0
                         beta_l
                         beta_2
                         beta 3
                            0.0467949
                                    0
                              4.23439
                           -0.0148331
                         -5.51541e-005
                                    0
                               Specified
                                 Parameter Estimates
Variable
alpha
beta 0
beta 1
beta 2
beta 3
Estimate
0.109717
4.03293
-0.0120086
0
0
Std. Err.
0.0233918
0.070386
0.000649584
NA
NA
                                          95.0% Wald Confidence Interval
                                        Lower Conf.  Limit  Upper Conf. Limit
                                             0.0638701           0.155564
                                             3.89498             4.17088
                                             -0.0132818          -0.0107354
NA - Indicates that this parameter has hit a bound
     implied by some inequality constraint and thus
     has no standard error.
           Asymptotic Correlation Matrix of Parameter Estimates

                  alpha       beta 0       beta 1
     alpha            1    -1.2e-(XT7     8.2e-OC>8
    beta_0    -1.2e-007            1         -0.7
    beta_l     8.2e-008         -0.7            1

The following parameter(s) have been estimated at a
boundary
point or have been specified.  Correlations are not
computed:

beta 2  beta 3
 Dose
Res .
     Table of Data and Estimated Values of Interest

            N    Obs Mean    Obs Std Dev   Est Mean   Est Std Dev   ChiA2
    0
   30
   60
  120
  240
10
10
 9
 9
 6
4.25
 3.7
3.25
2.15
 1.5
 0.3
0.25
0.05
 0.2
0.15
4.03
3.67
3.31
2.59
1.15
  Model Descriptions for likelihoods calculated


 Model Al:        Yij = Mu(i) + e(ij)
           Var{e(ij)} = SigmaA2
0.331
0.331
0.331
0.331
0.331
  2.07
 0.261
-0.565
    -4
  2.58
                                     A-41

-------
 Model A2:        Yij = Mu(i) + e(ij)
           Var{e(ij)} = Sigma(i)A2

 Model  R:         Yi = Mu + e(i)
            Var{e(i)} = SigmaA2
                       Likelihoods of Interest

            Model      Log(likelihood)   DF        AIC
             Al           48.017412       6     -84.034824
             A2           59.392540      10     -98.785081
           fitted         26.616696       2     -49.233392
              R          -21.651321       2      47.302642
 Test 1:  Does response and/or variances differ among dose
levels
          (A2 vs. R)
 Test 2:  Are Variances Homogeneous  (Al vs A2)
 Test 3:  Does the Model for the Mean Fit  (Al vs. fitted)
   Test

   Test 1
   Test 2
   Test 3
          Tests of Interest

-2*log(Likelihood Ratio)   Test df
            162.1
            22.7503
            42.8014
p-value

  <.0001
0.000142
  <.0001
The p-value for Test 1 is less than .05.  There appears
to be a
difference between response and/or variances among the
dose levels.
It seems appropriate to model the data

The p-value for Test 2 is less than .05.  Consider
running a
non-homogeneous variance model

The p-value for Test 3 is less than .05.  You may want
to try a
different model
 Benchmark Dose Computation
Specified effect =             1
Risk Type
            Estimated standard deviations from the control mean
Confidence level =

             BMD =
                 0.95

              27.5832
            BMDL =
                         22.9758
                                     A-42

-------
                        Polynomial Model with 0.95 Confidence Level
    4.5
 CD
 V)
 c
 o
 Q.
 
-------
                  Default Initial Parameter Values
                          alpha
                            rho
                         beta_0
                         beta_l
                         beta 2
                            0.0467949
                                    0
                              4.31231
                           -0.0224959
                         4.45961e-005
                               Specified
                                 Parameter Estimates
 Variable  Estimate
 alpha     0.0480497
 beta_0    4.30818
 beta_l    -0.0223839
 beta 2    4.40376e-005
                  Std.  Err.
                  0.0102442
                  0.0592565
                  0.00144605
                  5.86023e-006
                   95.0% Wald Confidence Interval
                        Lower Conf. Limit  Upper Conf.  Limit
                           0.0279713           0.068128
                           4.19204             4.42432
                           -0.0252181          -0.0195497
                           3.25518e-005        5.55235e-005
           Asymptotic Correlation Matrix of Parameter Estimates

alpha
beta 0
beta 1
beta 2
alpha
1
3.3e-009
-6.2e-010
-7.7e-009
beta 0
3.3e-009
1
-0.75
0.62
beta 1
-6.2e-010
-0.75
1
-0.95
beta 2
-7.7e-009
0.62
-0.95
1
 Dose
Res .
     Table of Data and Estimated Values of Interest

            N    Obs Mean    Obs Std Dev   Est Mean   Est Std Dev
                                                           ChiA2
    0
   30
   60
  120
  240
10
10
 9
 9
 6
4.25
 3.7
3.25
2.15
 1.5
 0.3
0.25
0.05
 0.2
0.15
4.31
3.68
3.12
2.26
1.47
0.219
0.219
0.219
0.219
0.219
-0.839
 0.342
  1.73
 -1.45
 0.306
  Model Descriptions for likelihoods calculated
 Model Al:        Yij
           Var{e(ij)}

 Model A2:        Yij
           Var{e(ij) }

 Model  R:         Yi
            Var{e(i) }
               Mu (i)  + e (ij )
               S i gma A 2

               Mu (i)  + e (ij )
               Sigma(i)A2

               Mu + e (i)
               SigmaA2
                                     A-44

-------
                       Likelihoods of Interest

            Model      Log(likelihood)   DF        AIC
             Al           48.017412       6     -84.034824
             A2           59.392540      10     -98.785081
           fitted         44.781449       3     -83.562899
              R          -21.651321       2      47.302642
 Test 1:  Does response and/or variances differ among dose
levels
          (A2 vs. R)
 Test 2:  Are Variances Homogeneous (Al vs A2)
 Test 3:  Does the Model for the Mean Fit  (Al vs. fitted)
   Test

   Test 1
   Test 2
   Test 3
          Tests of Interest

-2*log(Likelihood Ratio)   Test df
            162.1
            22.7503
            6.47193
p-value

  <.0001
0.000142
 0.03932
The p-value for Test 1 is less than .05.  There appears
to be a
difference between response and/or variances among the
dose levels.
It seems appropriate to model the data

The p-value for Test 2 is less than .05.  Consider
running a
non-homogeneous variance model

The p-value for Test 3 is less than .05.  You may want
to try a
different model
 Benchmark Dose Computation
Specified effect =             1
Risk Type
            Estimated standard deviations from the control mean
Confidence level =

             BMD =
                 0.95

              9.98916
            BMDL =
                          !.24167
                                     A-45

-------
                        Polynomial Model with 0.95 Confidence Level
    4.5
 CD
 V)
 c
 o
 Q.
 
-------
                  Default Initial Parameter Values
                          alpha =
                            rho =
                         beta_0 =
                         beta_l =
                         beta 2 =
                            0.0467949
                                    0
                              4.31231
                           -0.0224959
                                    0
                               Specified
                                 Parameter Estimates
   Variable
   alpha
   beta_0
   beta_l
   beta 2
     Estimate
     0.109717
     4.03293
     -0.0120086
     0
          Std.  Err.
          0.0233918
          0.070386
          0.000649584
          NA
          95.0% Wald Confidence Interval
         Lower Conf. Limit    Upper  Conf.  Limit
            0.0638701           0.155564
            3.89498             4.17088
            -0.0132818          -0.0107354
NA - Indicates that this parameter has hit a bound
     implied by some inequality constraint and thus
     has no standard error.
           Asymptotic Correlation Matrix of Parameter Estimates

alpha
beta 0
beta 1
alpha
1
1.4e-008
2.1e-009
beta 0
1.4e-008
1
-0.7
beta 1
2.1e-009
-0.7
1
The following parameter(s)  have been estimated at a
boundary
point or have been specified.  Correlations are not
computed:

beta 2
 Dose
Res .
     Table of Data and Estimated Values of Interest

            N    Obs Mean    Obs Std Dev   Est Mean   Est Std Dev   ChiA2
    0
   30
   60
  120
  240
10
10
 9
 9
 6
4.25
 3.7
3.25
2.15
 1.5
 0.3
0.25
0.05
 0.2
0.15
4.03
3.67
3.31
2.59
1.15
0.331
0.331
0.331
0.331
0.331
  2.07
 0.261
-0.565
    -4
  2.58
  Model Descriptions for likelihoods calculated

 Model Al:        Yij = Mu(i)  + e(ij)
           Var{e(ij)} = SigmaA2
                                     A-47

-------
 Model A2:        Yij = Mu(i) + e(ij)
           Var{e(ij)} = Sigma(i)A2

 Model  R:         Yi = Mu + e(i)
            Var{e(i)} = SigmaA2
                       Likelihoods of Interest

            Model      Log(likelihood)   DF        AIC
             Al           48.017412       6     -84.034824
             A2           59.392540      10     -98.785081
           fitted         26.616696       2     -49.233392
              R          -21.651321       2      47.302642

 Test 1:  Does response and/or variances differ among dose
levels
          (A2 vs. R)
 Test 2:  Are Variances Homogeneous  (Al vs A2)
 Test 3:  Does the Model for the Mean Fit (Al vs. fitted)
   Test
                     Tests of Interest

           -2*log(Likelihood Ratio)  Test df
p-value
   Test 1              162.088          8          <.0001
   Test 2              22.7503          4        0.000142
   Test 3              42.8014          2          <.0001
The p-value for Test 1 is less than .05.  There appears
to be a
difference between response and/or variances among the
dose levels.
It seems appropriate to model the data
The p-value for Test 2 is less than .05.
running a
non-homogeneous variance model
                                          Consider
The p-value for Test 3 is less than .05.  You may want
to try a
different model
 Benchmark Dose Computation
Specified effect =             1
Risk Type
                       Estimated standard deviations from the control mean
Confidence level =

             BMD =
                            0.95

                         27.5832
            BMDL =
                         22.9758
                                     A-48

-------
                        Polynomial Model with 0.95 Confidence Level
    4.5
 CD
 V)
 c
 o
 Q.
 
-------
                 Default Initial Parameter Values
alpha = 0.0467949
rho = 0 Specified
control = 4.25
slope = -340.972
power = -0. 879496
Asymptotic Correlation Matrix of Parameter Estimates
alpha rho control slope
alpha NA NA NA NA
rho NA NA NA NA
control NA NA NA NA
slope NA NA NA NA
power NA NA NA NA
NA - This parameter's variance has been estimated at zero.
Parameter Estimates
Variable Estimate Std. Err.
alpha 0.109717 0.114352
rho 0 0.935809
control 4.03293 0.129954
slope -0.0120086 0.0133934
power 1 0.202697
Table of Data and Estimated Values of Interest
Dose N Obs Mean Obs Std Dev Est Mean Est Std Dev
Res .
0 10 4.25 0.3 4.03 0.331
30 10 3.7 0.25 3.67 0.331
60 9 3.25 0.05 3.31 0.331
120 9 2.15 0.2 2.59 0.331
240 6 1.5 0.15 1.15 0.331


power
NA
NA
NA
NA
NA




ChiA2
0.655
0.0825
-0.188
-1.33
1.05
     Model Descriptions for likelihoods  calculated


Model Al:        Yij = Mu(i) + e(ij)
          Var{e(ij)} = SigmaA2
                                    A-50

-------
 Model A2:        Yij = Mu(i) + e(ij)
           Var{e(ij)} = Sigma(i)A2

 Model  R:         Yi = Mu + e(i)
            Var{e (i) } = SigmaA2
                       Likelihoods of Interest

            Model      Log(likelihood)   DF        AIC
             Al           48.017412       6     -84.034824
             A2           59.392540      10     -98.785081
           fitted         26.616696       4     -45.233392
              R          -21.651321       2      47.302642

 Test 1:  Does response and/or variances differ among dose levels
          (A2 vs. R)
 Test 2:  Are Variances Homogeneous  (Al vs A2)
 Test 3:  Does the Model for the Mean Fit (Al vs. fitted)

                     Tests of Interest

   Test    -2*log(Likelihood Ratio)     df        p-value

   Test 1              162.088          8         <.00001
   Test 2              22.7503          4        0.000142
   Test 3              42.8014          2         <.00001

The p-value for Test 1 is less than  .05.  There appears to be a
difference between response and/or variances among the dose levels.
It seems appropriate to model the data

The p-value for Test 2 is less than  .05.  Consider running a
non-homogeneous variance model

The p-value for Test 3 is less than  .05.  You may want to try a
different model
 Benchmark Dose Computation
Specified effect =             1

Risk Type        =     Estimated standard deviations from the control mean

Confidence level =          0.95

             BMD =       27.5832


            BMDL =       22.9758
                                     A-51

-------
                          Power Model with 0.95 Confidence Level
    4.5
    3.5
 CD
 V)
 c
 o
 Q.   o
 03    -J
 CD
 o:
 ro  2.5
 CD
     1.5
         Power
              BMDU  BMD
                         50
100         150
    dose
200
250
   23:21 05/122005
       Power Model. $Revision: 2.1 $ $Date: 2000/10/11 20:57:36 $
       Input Data File: C:\BMDS\DATA\CYLINDRO.(d)
       Gnuplot Plotting File:  C:\BMDS\DATA\CYLINDRO.plt
                                         Thu May 12 23:21:44 2005
HMDS MODEL RUN


  The form of the response function is:

  Y[dose]  = control + slope * doseApower
  Dependent variable = MEAN
  Independent variable = dose
  rho is set to 0
 The power is not restricted
  A constant variance model is fit

  Total number of dose groups = 5
  Total number of records with missing values = 0
  Maximum number of iterations = 250
  Relative Function Convergence has been set to: le-008
  Parameter Convergence has been set to: le-008
                                    A-52

-------
                 Default  Initial  Parameter Values
alpha = 0.0467949
rho = 0 Specified
control = 4.25
slope = -0.0336166
power = 0. 803617
Asymptotic Correlation Matrix of Parameter Estimates
alpha rho control slope
alpha 1 -0.97 -0.22 0.57
rho -0.97 1 0.22 -0.58
control -0.22 0.22 1 -0.67
slope 0.57 -0.58 -0.67 1
power 0.6 -0.62 -0.6 0.99
Parameter Estimates
Variable Estimate Std. Err.
alpha 0.0696661 0.0661961
rho 0 0.846718
control 4.30267 0.0817494
slope -0.0726613 0.0255026
power 0.676841 0.0658079
Table of Data and Estimated Values of Interest
Dose N Obs Mean Obs Std Dev Est Mean Est Std Dev
Res .
0 10 4.25 0.3 4.3 0.264
30 10 3.7 0.25 3.58 0.264
60 9 3.25 0.05 3.14 0.264
120 9 2.15 0.2 2.45 0.264
240 6 1.5 0.15 1.34 0.264


power
0.6
-0.62
-0.6
0.99
1



ChiA2
-0.2
0.468
0.41
-1.12
0.623
     Model Descriptions  for  likelihoods  calculated

Model Al:        Yij = Mu(i)  +  e(ij)
          Var{e(ij)} = SigmaA2

Model A2:        Yij = Mu(i)  +  e(ij)
          Var{e(ij)} = Sigma(i)A2
                                     A-53

-------
 Model  R:         Yi = Mu + e(i)
            Var{e(i)} = SigmaA2
                       Likelihoods of Interest

            Model      Log(likelihood)   DF        AIC
             Al           48.017412       6     -84.034824
             A2           59.392540      10     -98.785081
           fitted         36.608912       4     -65.217824
              R          -21.651321       2      47.302642

 Test 1:  Does response and/or variances differ among dose levels
          (A2 vs. R)
 Test 2:  Are Variances Homogeneous  (Al vs A2)
 Test 3:  Does the Model for the Mean Fit (Al vs. fitted)

                     Tests of Interest

   Test    -2*log(Likelihood Ratio)     df        p-value

   Test 1              162.088          8         <.00001
   Test 2              22.7503          4        0.000142
   Test 3               22.817          2       l.lle-005

The p-value for Test 1 is less than  .05.  There appears to be a
difference between response and/or variances among the dose levels.
It seems appropriate to model the data
The p-value for Test 2 is less than .05.
non-homogeneous variance model

The p-value for Test 3 is less than .05.
different model
                   Consider running a
                   You may want to try a
 Benchmark Dose Computation
Specified effect =
Risk Type

Confidence level =

             BMD =
Estimated standard deviations from the control mean

     0.95

  6.72481
            BMDL =
                         3.97902
                                     A-54

-------
                          Power Model with 0.95 Confidence Level
    4.5


      4


 o>  3-5
 V)
 c
 o
 Q.   o
 03    -J
 CD
 o:
 ro  2.5
 CD
     1.5
         Power
          BMD
L BMP
                         50
                       100         150
                           dose
200
250
   23:21 05/122005
       Hill Model. $Revision: 2.1 $ $Date: 2000/10/11 21:21:23 $
       Input Data File: C:\BMDS\DATA\CYLINDRO.(d)
       Gnuplot Plotting File:  C:\BMDS\DATA\CYLINDRO.plt
                                         Thu May 12 23:22:53 2005
HMDS MODEL RUN


  The form of the response function is:

  Y[dose]  = intercept + v*doseAn/(kAn + doseAn)
  Dependent variable = MEAN
  Independent variable = dose
  rho is set to 0
  Power parameter restricted to be greater than 1
  A constant variance model is fit

  Total number of dose groups = 5
  Total number of records with missing values = 0
  Maximum number of iterations = 250
  Relative Function Convergence has been set to: le-008
  Parameter Convergence has been set to: le-008
                                    A-55

-------
                  Default Initial Parameter Values
                          alpha =
                            rho =
                      intercept =
                              v =
                              n =
                              k =
                            0.0263306
                                    0
                                 4.25
                                -2.75
                              2.08208
                              80.4545
                               Specified
           Asymptotic Correlation Matrix of Parameter Estimates

                  alpha      rho        intercept
alpha
rho
intercept
V
n
k
1
0
0
0
0
0
0
1
0
0
0
0
0
0
1
0
0
0
V
0
0
0
1
0
0
n
0
0
0
0
1
0
k
0
0
0
0
0
1
       Variable
          alpha
            rho
      intercept
              v
              n
              k
                 Parameter Estimates

                 Estimate
                 0.0466116
                         0
                   4.22518
                  -3.35427
                   1.63976
                   94.2147
                             Std.
                                  Err.
                                    1
                                    1
                                    1
                                    1
                                    1
                                    1
 Dose
Res .
     Table of Data and Estimated Values of Interest

            N    Obs Mean    Obs Std Dev   Est Mean
    0
   30
   60
  120
  240
10
10
 9
 9
 6
                                             Est Std Dev
4.25
 3.7
3.25
2.15
 1.5
                                                  ChiA2
0
0.
0.
0
0.
.3
25
05
.2
15
4
3
3
2
1
.23
.78
.14
.22
.47
0
0
0
0
0
.216
.216
.216
.216
.216
0
-0
0
-0
0
.115
.369
.502
.323
.156
                                     A-56

-------
 Model Descriptions for likelihoods calculated
 Model Al:        Yij = Mu(i) + e(ij)
           Var{e(ij)} = SigmaA2

 Model A2:        Yij = Mu(i) + e(ij)
           Var{e(ij)} = Sigma(i)A2

 Model  R:         Yi = Mu + e(i)
            Var{e(i)} = SigmaA2
                       Likelihoods of Interest

            Model      Log(likelihood)   DF        AIC
             Al           48.017412       6     -84.034824
             A2           59.392540      10     -98.785081
           fitted         45.449933       5     -80.899866
              R          -21.651321       2      47.302642

 Test 1:  Does response and/or variances differ among dose levels
          (A2 vs. R)
 Test 2:  Are Variances Homogeneous  (Al vs A2)
 Test 3:  Does the Model for the Mean Fit (Al vs. fitted)

                     Tests of Interest

   Test    -2*log(Likelihood Ratio)  Test df     p-value

   Test 1              162.088          8          <.0001
   Test 2              22.7503          4        0.000142
   Test 3              5.13496          1         0.02345

The p-value for Test 1 is less than  .05.  There appears to be a
difference between response and/or variances among the dose levels.
It seems appropriate to model the data

The p-value for Test 2 is less than  .05.  Consider running a
non-homogeneous variance model

The p-value for Test 3 is less than  .05.  You may want to try a
different model
 Benchmark Dose Computation
Specified effect =             1

Risk Type        =     Estimated standard deviations from the control mean

Confidence level =          0.95

             BMD =       18.4161

            BMDL =       12.8257
                                     A-57

-------
                           Hill Model with 0.95 Confidence Level
   4.5



     4
o
Q.
W
CD
o:

c
CD
    1.5
          Hill
           BMDL
BMD
                         50
  23:2205/122005
                   100         150

                       dose
200
250
                                    A-58

-------
Part IV.




Humpage and Falconer 2003




male mice treated with purified cylindrospermopsin




urinary protein levels




drop high dose
                                     A-59

-------
       Polynomial Model. Revision: 2.2  Date: 9/12/2002
       Input Data File: C:\BMDS\DATA\CYLINDRO.(d)
       Gnuplot Plotting File:  C:\BMDS\DATA\CYLINDRO.plt
                                         Thu May 12 23:24:32 2005
HMDS MODEL RUN


  The form of the response function is:

  Y[dose]  = beta 0 + beta l*dose + beta 2*doseA2 + ...
  Dependent variable = MEAN
  Independent variable = dose
  rho is set to 0
  Signs of the polynomial coefficients are not restricted
  A constant variance model is fit

  Total number of dose groups = 4
  Total number of records with missing values = 0
  Maximum number of iterations = 250
  Relative Function Convergence has been set to:  le-008
  Parameter Convergence has been set to: le-008
                 Default Initial Parameter Values
                         alpha =    0.0503676
                           rho =            0   Specified
                        beta_0 =         4.25
                        beta 1 =    -0.017381
                                Parameter Estimates

                                         95.0% Wald Confidence Interval
  Variable    Estimate     Std.  Err.     Lower Conf. Limit  Upper Conf. Limit
  alpha       0.0457633    0.0104988        0.025186            0.0663406
  beta_0      4.24885      0.052757         4.14545             4.35225
  beta 1      -0.017373    0.000786454      -0.0189144          -0.0158316
          Asymptotic Correlation Matrix of Parameter Estimates

                 alpha       beta_0       beta_l
    alpha            1     2.2e-OC>9       le-OC>9
   beta_0     2.2e-009            1        -0.75
   beta 1       le-009        -0.75            1
                                    A-60

-------
     Table of Data and Estimated Values of Interest
Dose
Res .
0
30
60
120
N
10
10
9
9
Obs
4.
3
3.
2.
Mean
25
.7
25
15
Obs Std Dev
0.3
0.25
0.05
0.2
Est
4.
3.
3.
2.
Mean
25
73
21
16
Est
0
0
0
0
Std Dev
.214
.214
.214
.214
ChiA2
0.
-0.
0.
-0.

017
409
611
198
  Model Descriptions for likelihoods calculated
 Model Al:        Yij
           Var{e(ij)}

 Model A2:        Yij
           Var{e(ij)}

 Model  R:         Yi
            Var{e(i)}
             Mu (i) + e (ij )
             SigmaA2

             Mu (i) + e (ij )
             Sigma(i)A2

             Mu + e (i)
             S i gma A 2
                       Likelihoods of Interest
Model
Al
A2
fitted
R
Log (likelihood)
39.893006
50.462856
39.601194
-10.831438
DF
5
8
2
2
AIC
-69.786011
-84.925712
-75.202387
25.662876
 Test 1:  Does response and/or variances differ among dose
levels
          (A2 vs. R)
 Test 2:  Are Variances Homogeneous  (Al vs A2)
 Test 3:  Does the Model for the Mean Fit  (Al vs. fitted)
   Test

   Test 1
   Test 2
   Test 3
          Tests of Interest

-2*log(Likelihood Ratio)  Test df
            122.589
            21.1397
           0.583624
6
3
2
p-value

  <.0001
  <.0001
  0.7469
The p-value for Test 1 is less than  .05.  There appears
to be a
difference between response and/or variances among the
dose levels.
It seems appropriate to model the data
                                     A-61

-------
The p-value for Test 2 is less than  .05.  Consider
running a non-homogeneous variance model

The p-value for Test 3 is greater than  .05.  The model
chosen appears to adequately describe the data
 Benchmark Dose Computation
Specified effect =
Risk Type
 Estimated  standard  deviations  from the control mean
Confidence level =

             BMD =
       0.95

    12.3135
            BMDL =
                         10.1957
                            Linear Model with 0.95 Confidence Level
      4.5
   CD
   V)
   o  3.5
   Q.
   V)
   CD
   o:
   CD
   CD
      2.5
            Linear
               BMDL!
BMP
                        20
            40
 60
dose
80
100
120
    23:2405/122005
                                     A-62

-------
       Polynomial Model.  Revision:  2.2  Date:  9/12/2002
       Input Data File:  C:\BMDS\DATA\CYLINDRO.(d)
       Gnuplot Plotting File:   C:\BMDS\DATA\CYLINDRO.plt
                                         Thu May 12 23:26:08 2005
HMDS MODEL RUN


  The form of the response function is:

  Y[dose]  = beta 0 + beta l*dose + beta 2*doseA2 + ...
  Dependent variable = MEAN
  Independent variable = dose
  rho is set to 0
  Signs of the polynomial coefficients are not restricted
  A constant variance model is fit

  Total number of dose groups = 4
  Total number of records with missing values = 0
  Maximum number of iterations = 250
  Relative Function Convergence has been set to:  le-008
  Parameter Convergence has been set to: le-008
                 Default Initial Parameter Values
                         alpha =    0.0503676
                           rho =            0   Specified
                        beta_0 =      4.23636
                        beta_l =   -0.0164394
                        beta 2 = -7.57576e-006
                                Parameter Estimates
  Variable  Estimate
  alpha     0.0456423
  beta_0    4.23685
  beta 1    -0.016519
  beta 2
                           95.0% Wald  Confidence  Interval
                Std.  Err.    Lower  Conf.  Limit Upper  Conf.  Limit
                0.0104711       0.0251194        0.0661652
                0.0648402       4.10977          4.36394
                0.00280362      -0.022014        -0.011024
-6.92416e-006   2.18203e-005
                -4.96911e-005
                      3.58428e-005
          Asymptotic Correlation Matrix of Parameter Estimates
    alpha
   beta_0
   beta_l
   beta 2
     alpha
         1
  1.2e-010
 -1.4e-013
  l.le-010
  beta_0
1.2e-010
       1
   -0.73
    0.58
   beta_l
-1.4e-013
    -0.73
        1
    -0.96
  beta_2
l.le-010
    0.58
   -0.96
       1
                                    A-63

-------
 Dose
Res .
     Table of Data and Estimated Values of Interest

            N    Obs Mean    Obs Std Dev   Est Mean   Est Std Dev
                                                           ChiA2
    0
   30
   60
  120
10
10
 9
 9
4.25
 3.7
3.25
2.15
 0.3
0.25
0.05
 0.2
4.24
3.74
3.22
2.15
0.214
0.214
0.214
0.214
  0.195
 -0.519
   0.41
-0.0684
  Model Descriptions for likelihoods calculated
 Model Al:        Yij
           Var{e(ij)}

 Model A2:        Yij
           Var{e(ij)}

 Model  R:         Yi
            Var{e(i)}
               Mu (i)  + e (ij )
               SigmaA2

               Mu (i)  + e (ij )
               Sigma(i)A2

               Mu + e (i)
               S i gma A 2
                       Likelihoods of Interest
 Test 1:
levels

 Test 2:
 Test 3:
   Model      Log(likelihood)    DF        AIC
    Al           39.893006       5     -69.786011
    A2           50.462856       8     -84.925712
  fitted         39.651475       3     -73.302950
     R          -10.831438       2      25.662876

 Does response and/or variances differ among dose

 (A2 vs. R)
 Are Variances Homogeneous (Al vs A2)
 Does the Model for the Mean Fit (Al vs. fitted)

            Tests of Interest
   Test

   Test 1
   Test 2
   Test 3
  -2*log(Likelihood Ratio)   Test df
              122.589
              21.1397
             0.483061
                               p-value

                                 <.0001
                                 <.0001
                                  0.487
The p-value for Test 1 is less than .05.  There appears
to be adifference between response and/or variances among the
dose levels.It seems appropriate to model the data

The p-value for Test 2 is less than .05.  Consider
running a non-homogeneous variance model
The p-value for Test 3 is greater than  .05.  The model
chosen appears to adequately describe the data
                                     A-64

-------
 Benchmark Dose Computation

Specified effect =              1
Risk Type
  Estimated standard deviations from the  control  mean
Confidence level =



             BMD =
       0.95



    12.8637
            BMDL =
                          9.74645
                         Polynomial Model with 0.95 Confidence Level
     4.5
  CD
  V)

  o  3.5
  Q.
  V)
  CD
  c
  CD
  CD
     2.5
            Polynomial
               ftMDL
BMP
                        20
    23:2605/122005
            40
 60

dose
80
100
120
                                      A-65

-------
       Polynomial Model.  Revision:  2.2  Date:  9/12/2002
       Input Data File:  C:\BMDS\DATA\CYLINDRO.(d)
       Gnuplot Plotting File:   C:\BMDS\DATA\CYLINDRO.plt
                                         Thu May 12 23:27:07 2005
HMDS MODEL RUN


  The form of the response function is:

  Y[dose]  = beta 0 + beta l*dose + beta 2*doseA2 + ...
  Dependent variable = MEAN
  Independent variable = dose
  rho is set to 0
  The polynomial coefficients are restricted to be negative
  A constant variance model is fit

  Total number of dose groups = 4
  Total number of records with missing values = 0
  Maximum number of iterations = 250
  Relative Function Convergence has been set to: le-008
  Parameter Convergence has been set to: le-008
                 Default Initial Parameter Values
                         alpha =    0.0503676
                           rho =            0   Specified
                        beta_0 =      4.23636
                        beta_l =   -0.0164394
                        beta 2 = -7.57576e-006
                                Parameter Estimates

                                       95.0% Wald Confidence Interval
  Variable  Estimate        Std. Err.    Lower Conf. Limit  Upper Conf.  Limit
  alpha     0.0456423       0.0104711       0.0251194          0.0661652
  beta_0    4.23685         0.0648402       4.10977            4.36394
  beta_l    -0.016519       0.00280362      -0.022014          -0.011024
  beta 2    -6.92416e-006   2.18203e-005    -4.96911e-005      3.58428e-005
          Asymptotic Correlation Matrix of Parameter Estimates

                 alpha       beta_0       beta_l       beta_2
    alpha            1     1.2e-010    -1.4e-013     l.le-010
   beta_0     1.2e-010            1        -0.73         0.58
   beta_l    -1.4e-013        -0.73            1        -0.96
   beta 2     l.le-010         0.58        -0.96            1
                                    A-66

-------
 Dose
Res .
    0
   30
   60
  120
     Table of Data and Estimated Values of Interest

            N    Obs Mean    Obs Std Dev   Est Mean   Est Std Dev   ChiA2
10
10
 9
 9
4.25
 3.7
3.25
2.15

0
0

0
.
.
0
.3
25
05
.2
4
3
3
2
.24
.74
.22
.15
0
0
0
0
.214
.214
.214
.214
0
-0

-0.
.195
.519
0.41
0684
  Model Descriptions for likelihoods calculated
 Model Al:        Yij
           Var{e(ij)}

 Model A2:        Yij
           Var{e(ij)}

 Model  R:         Yi
            Var{e(i)}
               Mu (i)  + e (ij )
               SigmaA2

               Mu (i)  + e (ij )
               Sigma(i)A2

               Mu + e (i)
               S i gma A 2
                       Likelihoods of Interest





Test 1:
levels

Test 2:
Test 3:
Model
Al
A2
fitted
R
Log (likelihood)




Does response

(A2 vs .

R)
Are Variances
Does the
39.893006
50.462856
39.651475
-10.831438




and/or variances


Homogeneous


DF
5
8
3
2
differ


AIC
-69.786011
-84.925712
-73.302950
25.662876
among dose


(Al vs A2)
Model for the Mean Fit
(Al vs
. fitted)
   Test

   Test 1
   Test 2
   Test 3
                     Tests of Interest
  -2*log(Likelihood Ratio)   Test df
              122.589
              21.1397
             0.483061
                               p-value

                                 <.0001
                                 <.0001
                                  0.487
The p-value for Test 1 is less than .05.  There appears to be a difference
between response and/or variances among the dose levels.
It seems appropriate to model the data

The p-value for Test 2 is less than .05.  Consider running a non-homogeneous
variance model
The p-value for Test 3 is greater than  .05.  The model
chosen appears to adequately describe the data
                                     A-67

-------
 Benchmark Dose Computation

Specified effect =              1
Risk Type
  Estimated standard deviations  from the  control mean
Confidence level =



             BMD =
       0.95



    12.8637
            BMDL =
                          10.2295
                         Polynomial Model with 0.95 Confidence Level
     4.5
  CD
  V)

  o  3.5
  Q.
  V)
  CD
  o:

  c    q
  CD    -J
  CD
     2.5
            Polynomial
               BMDL
BMP
                        20
    23:2705/122005
40
 60

dose
                                80
100
120
                                      A-68

-------
       Power Model.  $Revision: 2.1 $ $Date: 2000/10/11 20:57:36 $
       Input Data File:  C:\BMDS\DATA\CYLINDRO.(d)
       Gnuplot Plotting File:  C:\BMDS\DATA\CYLINDRO.plt
                                         Thu May 12 23:27:43 2005
HMDS MODEL RUN


  The form of the response function is:

  Y[dose]  = control + slope * doseApower
  Dependent variable = MEAN
  Independent variable = dose
  rho is set to 0
  The power is restricted to be greater than or equal to 1
  A constant variance model is fit

  Total number of dose groups = 4
  Total number of records with missing values = 0
  Maximum number of iterations = 250
  Relative Function Convergence has been set to: le-008
  Parameter Convergence has been set to: le-008
                 Default Initial Parameter Values
                         alpha =    0.0503676
                           rho =            0   Specified
                       control =         4.25
                         slope =     -22.4349
                         power =    -0.494765
          Asymptotic Correlation Matrix of Parameter Estimates

alpha
rho
control
slope
power
alpha
1
-0.98
-0.041
-0.0042
-0.014
rho
-0.98
1
0.042
0.0043
0.014
control
-0.041
0.042
1
-0.67
-0.62
slope
-0.0042
0.0043
-0.67
1
1
power
-0.014
0.014
-0.62
1
1
                                    A-69

-------
                         Parameter Estimates
Variable
alpha
rho
control
slope
power
Estimate
0.0457244
0
4.24139
-0.0159238
1.01795
Std. Err
0.0552045
0.931145
0.0671298
0.00783392
0.100882
    Table of Data and Estimated Values of Interest
Dose
Res .
0
30
60
120
N
10
10
9
9
Obs
4.
3
3.
2.
Mean
25
.7
25
15
Obs

0
0

Std Dev
0.3
.25
.05
0.2
Est
4.
3.
3.
2.
Mean
24
73
21
16
Est
0
0
0
0
Std Dev
.214
.214
.214
.214
Chi
0
-

-0
A2
.0403
0.157
0.172
.0426
     Model Descriptions for likelihoods calculated
Model Al:        Yij = Mu(i) + e(ij)
          Var{e(ij)} = SigmaA2

Model A2:        Yij = Mu(i) + e(ij)
          Var{e(ij)} = Sigma(i)A2

Model  R:         Yi = Mu + e(i)
           Var{e(i)} = SigmaA2

                      Likelihoods of Interest

           Model      Log(likelihood)   DF        AIC
            Al           39.893006       5     -69.786011
            A2           50.462856       8     -84.925712
          fitted         39.617353       4     -71.234706
             R          -10.831438       2      25.662876
Test 1:

Test 2:
Test 3:
Does response and/or variances differ among dose levels
(A2 vs. R)
Are Variances Homogeneous  (Al vs A2)
Does the Model for the Mean Fit  (Al vs. fitted)
  Test

  Test 1
  Test 2
  Test 3
           Tests of Interest

 -2*log(Likelihood Ratio)

             122.589
             21.1397
            0.551305
df

6
3
1
   p-value

   <.00001
9.847e-005
    0.4578
                                    A-70

-------
The p-value for Test 1 is less than .05.  There appears to be a
difference between response and/or variances among the dose levels.
It seems appropriate to model the data

The p-value for Test 2 is less than .05.  Consider running a
non-homogeneous variance model

The p-value for Test 3 is greater than  .05.  The model chosen appears
to adequately describe the data
 Benchmark Dose Computation
Specified effect =
Risk Type

Confidence level =

             BMD =
Estimated standard deviations from the control mean

     0.95

  12.8275
            BMDL =
                         10.2065
                            Power Model with 0.95 Confidence Level
      4.5
   CD
   I  3.5
   Q.
   V)
   CD
   o:
   CD
      2.5
            Power
                BMDL BMP
                         20
           40
 60
dose
80
100
120
     23:2705/122005
                                     A-71

-------
       Hill Model. $Revision: 2.1 $ $Date: 2000/10/11 21:21:23 $
       Input Data File: C:\BMDS\DATA\CYLINDRO.(d)
       Gnuplot Plotting File:  C:\BMDS\DATA\CYLINDRO.plt
                                         Thu May 12 23:28:26 2005
HMDS MODEL RUN


  The form of the response function is:

  Y[dose]  = intercept + v*doseAn/(kAn + doseAn)
  Dependent variable = MEAN
  Independent variable = dose
  rho is set to 0
  Power parameter restricted to be greater than 1
  A constant variance model is fit

  Total number of dose groups = 4
  Total number of records with missing values = 0
  Maximum number of iterations = 250
  Relative Function Convergence has been set to: le-008
  Parameter Convergence has been set to: le-008
                 Default Initial Parameter Values
                         alpha =    0.0273861
                           rho =            0   Specified
                     intercept =         4.25
                             v =         -2.1
                             n =       1.7744
                             k =      62.7273
          Asymptotic Correlation Matrix of Parameter Estimates

                alpha     rho       intercept

                                      0

                                      0

                                      1

                                      0

                                      0

                                      0
alpha
rho
intercept
V
n
k
1
0
0
0
0
0
0
1
0
0
0
0
V
0
0
0
1
0
0
n
0
0
0
0
1
0
k
0
0
0
0
0
1
                                    A-72

-------
                          Parameter Estimates
       Variable
          alpha
            rho
      intercept
              v
              n
              k
                Estimate
                0.0457348
                        0
                  4.24125
                 -461.591
                  1.02098
                  23703.4
Std.
     Err.
       1
       1
       1
       1
       1
       1
     Table of Data and Estimated Values of Interest
R
-



Dose
es .
0
30
60
120
N
10
10
9
9
Obs
4.
3
3.
2.
Mean
25
.7
25
15
Obs

0
0

Std Dev
0.3
.25
.05
0.2
Est
4.
3.
3.
2.
Mean
24
73
21
16
Est
0
0
0
0
Std Dev
.214
.214
.214
.214
ChiA2
0.0409
-0.159
0.174
-0.0428
 Model Descriptions for likelihoods calculated
 Model Al:        Yij
           Var{e(ij)}

 Model A2:        Yij
           Var{e(ij) }

 Model  R:         Yi
            Var{e(i) }
              Mu (i) + e (ij )
              SigmaA2

              Mu (i) + e (ij )
              Sigma(i)A2

              Mu + e (i)
              SigmaA2
Degrees of freedom for Test Al vs fitted <= 0

                       Likelihoods of Interest

            Model      Log(likelihood)   DF
             Al           39.893006       5
             A2           50.462856       8
           fitted         39.613009       5
              R          -10.831438       2
 Test 1:

 Test 2:
 Test 3:
   Test
                                         AIC
                                      -69.786011
                                      -84.925712
                                      -69.226018
                                       25.662876
Does response and/or variances differ among dose levels
(A2 vs. R)
Are Variances Homogeneous  (Al vs A2)
Does the Model for the Mean Fit  (Al vs. fitted)
           Tests of Interest

 -2*log(Likelihood Ratio)  Test df
  p-value
                                     A-73

-------
   Test 1              122.589          6          <.0001
   Test 2              21.1397          3          <.0001
   Test 3             0.559993          0              NA

The p-value for Test 1 is less than .05.  There appears to be a
difference between response and/or variances among the dose levels.
It seems appropriate to model the data

The p-value for Test 2 is less than .05.  Consider running a
non-homogeneous variance model

NA - Degrees of freedom for Test 3 are less than or equal to 0.  The Chi-
Square test for fit is not valid
 Benchmark Dose Computation
Specified effect =             1

Risk Type        =     Estimated standard deviations from the control mean

Confidence level =          0.95

             BMD =       12.8644

Warning:  optimum may not have been found.  Bad completion code in
Optimization routine.

BMDL computation failed.
                                     A-74

-------
                                        Hill Model
    4.5
CD
V)

o   3.5
Q.

CD
o:

c    q
CD    °
CD
    2.5
           Hill
                     BMP
                       20
40
 60


dose
80
100
120
  23:2805/122005
                                     A-75

-------