PB81-187239
State-of-the-Art Report  on
Structure-Activity  Methods  Development
(U.S.) Environmental Research Lab.
Duluth, MN
Apr 81
                     U.S. DEPARTMENT OF COMMERCE
                  National Technical Information Service
                                   NTIS

-------
United States                         EPA 600/3-81-029
Environmental Protection                    April 1981
Agency
Research  and
   eveiopment
STATE-OF-THE-ART REPORT ON STRUCTURE-ACTIVITY
METHODS DEVELOPMENT
OFFICE OF TOXIC SUBSTANCES
Prepared by

Environmental Research
Laboratory
Duluth MN 55804

-------
                                   TECHNICAL REPORT DATA
                            (Please read Instructions on the reverse before completing)
1. REPORT NO.
    EPA-6QQ/3-81-02q
Prnhlpm-flHpntpH Ropnrt
!•> neriPIENT'S ACCESSION NO.
 F881    187239
4. TITLE AND SUBTITLE
 State-of-the-Art of Structure Activity

 Methods Development
                          5. REPORT DATE
                               April  1981
                          6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
Oilman  D.  Veith
                                                           8. PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
U.  S.  Environmental Protection Agency
Environmental Research Laboratory-Duluth
6201 Congdon Boulevard
Duluth,  Minnesota  55804
                                                           10. PROGRAM ELEMENT NO.
                           11. CONTRACT/GRANT NO.
12. SPONSORING AGENCY NAME AND ADDRESS
U.  S.  Environmental Protection  Agency
Environmental Research Laboratory-Duluth
6201 Congdon Boulevard
Duluth,  Minnesota  55804
                                                           13. TYPE OF REPORT AND PERIOD COVERED
                           14. SPONSORING AGENCY CODE
IS. SUPPLEMENTARY NOTES
16. ABSTRACT T;he  overall objective of  this  project  is to provide the Agency  with the tech-
nical basis  for estimating the toxicity  and  environmental behavior of  organic chemicals
from molecular structure.  The project  is  directed toward the evaluation  of  quantitative
structure-activity relationships (QSAR)  for  use  by EPA Program Offices  and  toward the
development  of new data and QSAR methods to  extend the technique to meet  Agency needs.
     Specifically, the objective of  the  Structure-Activity Project is  to  develop meth-
ods to  predict the toxicity, persisitence,  and treatability of large  numbers of untest-
ed chemicals using QSAR based on structural,  chemical, and biological  properties of
representative reference data bases.  Development of QSAR is being tailored  for use in
the (1)  initial screening of chemicals  under TSCA, (2) development of  risk  assessment
strategies,  (3) prioritization of chemicals  for  Water Quality Criteria development, and
(4) the  optimization of national monitoring  programs for toxic chemicals.
     This  report summarizes the progress during  the first six months  of the  project.
The report provides a literature review  and  perspective for applying  structure-activity
methods  to aquatic toxicity of industrial  chemicals.  Experimental work centered on
developing methods for estimating molecular  descriptors such as  log P  and connectivity
indexes  and  on the development of a systematic structure-activity data base  for aquatic
toxicity.   A new program for entering structures into a computer and  calculating con-
nectivity  indexes is discussed.  A  general  model for predicting 96 hour LC50 for
narcotic chemicals is presented.	
17.
                                KEY WORDS AND DOCUMENT ANALYSIS
                  DESCRIPTORS
             b.lOENTIFIERS/OPEN ENDED TERMS  C.  COS ATI Field/Group
18. DISTRIBUTION STATEMENT

Release  to Public
              19. SECURITY CLASS (ThisReport)
                Unclassified
                                                                         21.
                                              20. SECURITY CLASS (Thispage)
                                                Unclassified
                                                                         22. PRICE
EPA Form 2220-1 (Rซv. 4-77)   PREVIOUS EDITION is OBSOLETE

-------
                  NOTICE






THIS  DOCUMENT  HAS  BEEN REPRODUCED



FROM TH'E  BEST  COPY  FURNISHED  US BY



THE  SPONSORING AGENCY.  ALTHOUGH IT



IS RECOGNIZED THAT CERTAIN PORTIONS



ARE  ILLEGIBLE,  IT I'S  BEING  RELEASED



IN THE ''INTEREST OF  MAKING  AVAILA-BLE



AS  MUCH INFORMATION AS POSSIBLE.

-------
                                                                  EPA 600/3-81-029
                                                                  April  1981
 STRUCTURE-ACTIVITY RESEARCH AT THE ENVIRONMENTAL RESEARCH LABORATORY-DULUTH
      State-of-the-Art Report on Structure-Activity Methods Development

                                October, 1980
                                 Prepared by

                               Oilman D. Veith

                  Environmental Research Laboratory-Duluth

                           6201 Congdon'Boulevard   •

                          Duluth, Minnesota  55804
                            List of Contributors:
ERL-Duluth

Rosemarie Russo

David Weininger

Brad Greenwood

Gary Phipps

Gary Holcombe

Dave DeFoe
University of Wisconsin-
	Superior	

     Daniel Call

     Larry Brooke

     Nasim Ahmad

     Carolanne Curtis  '•;"

     Anne Lima     •

     Steven Lozano

     Joseph Richter

     Karen Studders
University of Minnesota-
	Duluth	

    Vincent. Magnuson

    Donald Harriss

    Wayne Maanum

 • ::-. Marc ia:-Fulton 	 •;•-

-------
                              Table .of Contents
Section
Title
Page
   I.       Project Objectives




  II.       Project Overview




 III.       Project Organization         -




  IV.       Estimation of .Acute Toxicity of Organic Chemicals




              to Fish




   V.       Mathematical Methods  for QSAR




  VI.       Molecular Descriptor Generation




 VII.       QSAR Information System




VIII.       Activity Reports




  IX.       Evaluation of Rapid-Screening Tests for QSAR




   X.       Literature Review on Narcosis




  XI.       References
                                           1




                                           2




                                           6




                                           8








                                          26




                                          36




                                          54




                                          57




                                          63




                                          88




                                         101

-------
                       ERL-D Structure-Activity Research Project









I.    PROJECT OBJECTIVES




          The overall objective of this project is to provide the Agency with the




     technical basis for estimating the toxicity and environmental behavior of




     organic chemicals from molecular structure.  The project is directed toward




     the evaluation of quantitative structure-activity relationships (QSAR) for




     use by EPA Program Offices and toward the development of new data and QSAR




     methods to extend the technique to meet Agency needs.




          Specifically, the objective of the Structure-Activity Project is to




     develop methods to predict the toxicity, persistence, and treatability of




     large numbers of untested chemicals using QSAR based on structural, chemical,




     and biological properties- of representative reference data bases.




     Development of QSAR is being tailored for use in the (1) initial screening of




     chemicals under TSCA, (2) development of risk assessment strategies, (3)




     prioritization of chemicals for Water Quality Criteria development, and (4)




     the optimization of national monitoring programs for toxic chemicals.

-------
II.  PROJECT OVERVIEW




          The broad objectives of  predicting  tioxicity,  persistence,  and




     treatibility of chemicals require considerable definition  at  the  outset.   The




     use of QSAR carries with  it the assumption  that  a  well-defined  chemical  or




     biological activity can be related to a  combination  of  free-energy,




     'electronic, and steric properties of the molecule.   This  assumption  dictates




     th'at the measured activity (hereafter called endpoints) be well-defined  and




     dependent on the chemical rather than the environment  into which  the chemical




     is placed.  While aquatic toxicity may be estimated  directly  by predicting




     'the acute LC50 or chronic MATC  from reference data bases,  estimations  of       ',:.'.:




     persistence or treatability are likely to be as  dependent  on  the  system        • ••'-.




     studied as on the properties  of the chemical.  Therefore,  QSAR  in these  areas.




     'can only be used in predicting  rate constants and  equilibria  which  are needed




     inputs to process models  for  persistence and treatability. QSAR  can be




     developed to estimate hydrolysis, adsorption, volatilization, and




     biodegradability - but the full use of this approach will  only  be realized in




     the development of transport  and transformation  process models  such  as the




     EXAMS model development at ERL-Athens.




          The overall QSAR development process is outlined  in  Figure 1.   Most




     'often, a data base is compiled  for the endpoints of  interest  (acute  toxicity,




     volatility, etc.) and selected  molecular descriptors which are  thought to be




     important controlling factors in the relative activities  of the chemicals.




     The data base is often classified into homologous  or semi-homologous series




     of chemicals to increase  the  likelihood  of  common  modes of action or reaction




     mechanisms.  Where critical data; are lacking, a  data generation program  is




     needed to provide the data.   The compiled molecular  descriptors and  endpoints




     are analyzed by a variety of  mathematical techniques such  as  multiple




                       • ' '    ••      '        2          •    '

-------
                                QSAR  PROCESS


Classify

Reference
Base







p* 	 .-^..|

~* 	 ^~



Select
Molecular
Descriptors



Conduct
Sv^tpmntir
Testina









Compile
Molecular
Descriptors

J
?
1ป- Dpuplnn

QSAR
                                                                          : 8.
                                                                          nฐ
                                                                          o 3
 Untested
Chemicals
                                    Generate
                                    Molecular
                                   Descriptors
   Estimate,^.
 Toxicity an9
Fate End points
  OUTPUT

PREDICTIONS
                          Figure 1.  Flow-Chart of the QSAR Process

-------
regression and pattern recognition to derive a  quantitative




structure-activity relationship  for the selected  class  of  chemicals  and  the




specified endpoint.




     In making predictions beyond the tested chemicals,  the  chemical must  be




classified to select the proper  QSAR model.  The  molecular descriptors which




are the variables in the QSAR  for the chemical  must  be  generated,  and the




endpoint is computed as a QSAR prediction.  Obviously,  only  endpoints for




which there are QSAR can be predicted.




     Figure 1 shows that QSAR  must be developed along many research  fronts if




it is;to be useful for large numbers of diverse chemicals.   In  the initial




phase of this project, it was  immediately  apparent that  there were few data




of'use-in QSAR development.  The aquatic toxicity literature contained little




systematically derived data due  to past emphasis  on  testing  relatively few




highly'hazardous chemicals extensively.  Moreover, important molecular




descriptors were nearly non-existent•for chemicals outside the  interests of'




the''drug and pesticide industry.  It would-be of•little  value to develop QSAR




models which required parameters not available  for individual chemicals.




     Limitations with respect  to the level and  mechanism of  funding  for  this




project required that the first  year'objectives^be redefined as  follows.  It




was apparent that numerous testing laboratories could develop QSAR models




independent of this project if the necessary molecular  descriptors were  made




av'ailable to them.  Therefore, approximately 25 percent  of the  resources are




assigned to the evaluation, calculation, and compilation of  molecular




descriptors.  An additional 25 percent of  the resources  are  assigned to  the




development of a user oriented,  computeriz-ed data dissemination  system called




the Information System for"Hazardous Organic^ in  Water  (ISHOW).  The

-------
remaining 50 percent of the resources are  assigned  to  the  data  generation




effort.




     Because of the expertise of  this laboratory  in aquatic  toxicity  testing,




data generation is being focused  on developing a  systematic  aquatic toxicity




data base for industrial chemicals.  In anticipation that  the data base  will




have to contain approximately 20  chemicals of each  chemical  class, the use  of




chronic tests is precluded by the cost and time requirements.   Therefore,  the




initial data generation consists  primarily of the determination of the 96




hour LC50 of industrial chemicals for fish.  Lesser efforts  are being made  on




expanding the model developed to  predict the bioaccumulation potential of




chemicals in aquatic organisms, and comparing the relative sensitivity of




different aquatic species.




  '  .'tt.-'is-anticipated that testing will expand to  include additional




toxicity endpoints after the usefulness of the QSAR technique is




demonstrated.  Furthermore, coordination of this  project with other ORD




laboratories which have the expertise in studies  of persistence and




treatability of organic chemicals is underway.  Joint  efforts are being




discussed for the generation of data bases of key endpoints  needed in the




modeling efforts in the area commonly called "fate".

-------
III. PROJECT ORGANIZATION                                .



          The organization of ..the ERL-Duluth QSAR project  is presented  in general



     terms in Figure 2.  The project is centered with a core group of ERL-Duluth



     staff which is supported by the ORD Decision Unit for Toxic - Ecological



     Effects and Transport and Fate.  The ERL-Duluth staff provides project



     planning and coordination, tailoring' of QSAR methods  to meet EPA needs, and a



     nucleus for toxicity testing.  A cooperative agreement between ERL-Duluth and



     the University of Wisconsin-Superior and funded jointly by ORD and OTS



     provides the project with data compilation, toxicity  data generation,



     toxicological interpretation of test results and the  development of rapid



     screening methods for the structure activity project.  A cooperative



     agreement with the University of Minnesota-Duluth funded by ORD and



     EPA-Region V is1 focused on the'dissemination 'of QSAR  predictive technology
                                                         •j*          ...


     for large numbers of chemicals, and the development of QSAR models to  be



     tested.  The progress of each of the teams listed in  Figure 2 will be



     presented in this report.

-------
                   Figure 2.  Organization of QSAR Project
University of Wisconsin-
       Superior
 Cooperative Agreement
   ERL-Duluth
 Literature Review an<
 Data Compilation
    Lima
    Studders
    2 Technicians
 (Screening Test Methods}

    Richter
    Curtis

  {Toxicology!

    Ahmad
    Joilymore

 JToxicity Test Teama[

  Test Operations

    Shubat
    Poirer
    Anderson
    Call
    Brooke

  Analysis and Methods

    Knuth
    Hammermeister
    Hoglund
[Project Coordination}

     Veith
Modeling and Computer
  	Programs	
University of Minnesota-
         Dululth
 Cooperative Agreement
   [ISHOW Development)

        Magnuson
        Maanum
        Fulton
        LaLiberte
     Weiniger
     Greenwood

  (Biostat 1st ics)

     Lozano

[Toxic ity  Test TeamS}

  Test Operations

     DeFoe

  Analysis and  Methods

     Huot

JToxicity  Test Teamfl

  Test Operations

     Phipps
     Holcombe

  Analysis and  Mehtods

     Fiandt
     Felhaber
     Walbridge
      IQSAR Models!

        Harriss
        2 Associate
          Scientists

-------
IV.    ESTIMATION OF ACUTE TOXICITY OF ORGANIC CHEMICALS TO FISH




      A.    Definition of Boundaries




           The literature concerning the application of QSAR to environmental




      sciences is growing rapidly.  As with the drug and pharmacology  literature,




      many of the QSAR equations are isolated bits and pieces of  information which




      fail to put any individual study into a larger perspective.  This




      fragmentation becomes especially frustrating when the literature is  searched




      for possible solutions to the problems facing EPA.  In considering the




      problem of estimating the behavior of 30,000-40,000 industrial chemicals  in




      the aquatic environment, it seems necessary to approach the  literature with




      the belief that such things as toxicity and persistence follow very 'simple




      relationships among chemicals and that there are some boundary conditions to




      the problem.  Without:this belief; our future generations will be continuing




      our testing in the "hit and miss" fashion of the past 30 years.




           In 1968, Corwin Hansch (Hansch, et al. 1968) provided  an illustrative




      example which demonstrated that organic chemicals in water  behaved in an




      orderly fashion.  Due to the relationship between factors controlling dis-




      solution of organic chemicals in water and that of partitioning  the  chemical




      between a water phase and a lipid phase, Hansch demonstrated quantitative




      relationships between the water' solubility of chemicals and  their




      n-octanol/water partition coefficients.  The 'relationships  were  simple:-




                               log 1 = a log P +i b




      where s is the solubility in moles/1, P is the partition coefficient, and a




      and b are constants.  Using data for 156 chemicals of 9 chemical classes, the




      equation




          log 1 = 1.339 (+_ 0.07) log P - 0.978 (^ 0.15)       r -  0.935




      held over a range of about 100,000 in water solubility.  When the data were




      separated into individual classes, values of r were as great as  0.990.




                                       .8

-------
Therefore, it is apparent that  log P  is  a  free-energy  term which  is  a

manifestation of much of the structural  information  which  affects  the

behavior of chemicals in aqueous  systems.

     Implicit in these relationships  is  the  likelihood that  the  log  P will

also be important  in the partitioning of chemicals  from water into aquatic

organisms and onto organic-coated surfaces.   Therefore,  it is almost anti-

climactic to have  published papers  like  that  of Veith  et al.  (1978)  showing

linear relationships between the  bioconcentration  factor in  aquatic  organisms

and log P, or that of Karickhoff  et al.  (1978) showing relationships between

the sediment/water partition coefficients  and log  P  as follows:

                           Kp = 0.6 x P  x  (O.C.)

where K_ is the sediment/water  partition coefficient and O.C. is  the

fractional mass of organic carbon with sediment.   Lopez-Avila and Kites

(1980) reported that the concentration of  numerous  organic chemicals in river

sediments downstream from discharges  can be modeled  with the  empirically

derived equation

                   log CQ = bQ  +  b2 (distance/log  P)
                       C


where CQ is the concentration in  the  discharge and  bQ  and  b2  are

constants for the  river system.

     The work presented by Hansch is  important to  aquatic  toxicity studies

because it helps to define the  boundaries  of  the "testing  space".  For

example, Figure 3  presents a plot of  the "testing  space" for  chemicals.over

the log P range from -2 to 8.   This is 10  orders of  magnitude of  lipo-

phillicity and  includes the vast  majority  of. industrial organic  chemicals on

the TSCA inventory.  The vertical axis in  Figure 3  is  the.  logarithm  of the

concentration of chemicals expressed  as  moles/liter  which  varies  from 2 to -8,

                                      9

-------
  0
-2
-4
  0
-2
-4   J.
                                                                                           O
                                                                                           o
                                                                                           -J
                         Figure 3.  Aquatic Toxicity Testing Space Boundaries

-------
which is approximately Che range of concentrations  of  interest  in testing for




most toxic effects in the aquatic system.  The  right-hand  scale  is  the  water




solubility of chemical expressed as the  logarithm of  the molar  concentration




at saturation.  The  line represents the  equation  presented by Hansch




reanalyzed to include the 95 percent  prediction limits  of  the relationship




between water solubility and log P.   This  "testing  space"  should enclose most




of the organic chemicals encountered  in  this  study.  Quite obviously,  if the




estimated LC50 for a chemical of known log P  lies above the water solubility




line, it should not be possible to measure the  LC50 in  aquatic  systems.




B.   Model for Physical Toxicity in Acute  Tests




     Little has been done in the area of predicting the toxicity of organic




chemicals to aquatic organisms from structure since the work of  Meyer  and




Overton in 1899-1901. ..The literature contains  a  'few  isolated examples




demonstrating toxicity-structure relationships; however, a perspective




suitable for shaping research to address the  problems  facing EPA, and OTS in




particular, is absent.




     There is a thread running through the literature which may  give the




necessary guidance in tackling the problem.   Richet's Rule published in 1893




proposed that the toxicity of a chemical is  roughly inversely proportional to




its water solubility.  Implicit in this  rule  is the fact  that comparable




modes of action are  considered.  Ferguson  (1939)  proposed  that  when a physio-




logical effect is reversible, when an equilibrium exists between the organism




and the external phase, and when the  physiological  effect  is a  function of




the external concentration, the toxic effect  must be  physical in character.




He suggested that all substances at some concentration  will exert toxic
                                      11

-------
 action by physical mechanisms and even nitrogen gas at high enough pressure


 can produce effects.  The physical effect may be masked by irreversible


 chemical effects; but in the absence of chemical effects, the primary process


 is  reversible and physical in nature.


      The physical effect of Ferguson was expressed as

                                   1                             •
                             C = kSn


 where S is the water solubility expressed in moles/liter.  This physical


 effect is called narcosis which is any reversible decrease in physiological


 function induced by physical or chemical agents (Mullin, 1954).  Ferguson


 proposed that the chemical activity for chemicals at concentrations producing


 narcosis is constant.  Since the thetmodynamic activity of a 'chemical


 producing narcosis (f^a^ is calculated by
                     .
   ;.; •• ..-.-,-.   /.   .-. .  _.    -;•  cs •'_".:•:. ;••  :•::;••    '  -  .   -.'     ..•-.-•



'where  Cnar ฃs tne concentration:producing'harcosis and Cg is the

"solubility of the chemical in water expressed as moles/liter, if the


 activities of chemicals for the same physiological effect are constant  for


 series  of  chemicals, the' observed effects should" be a constant ratio with


 respect to the water solubility as expressed by Richet's Rule.


     Mullin (1954) presented data consistent with this idea and  further


 elaborated on the nature of physical and chemical toxicity.  He  defined


 'narcosis as any reversible decrease in physiological function where the


 effects can be divided into narcosis of cell division and narcosis of  the


 central nervous system (general anesthesia).  Most chemicals can cause  both


 physical and chemical toxicity; however, if the chemical toxicity is  small


 under  the  exposure time studied, the predominant effect will be  physical


                                      U

-------
toxicity or narcosis as  the exposure concentration  approaches  the water




solubility of the chemical.  Thus,  in a  simplistic  way,  the  prediction of




toxicity of chemicals to  fish may parallel  the  zero based  budgeting (ZBB)




method in that, unless specific  functional  groups that  produce chemical




toxicity are present in  a chemical  of interest,  the chemical is likely to




cause physical toxicity, or narcosis, at  some fraction  of  its  water




solubility.  If chemical toxicity occurs, the physical  toxicity is masked and




the observed toxic effects are likely to  be  at  lower concentrations than the




concentration needed for narcosis.




     The model immediately posed a  problem  for  this project.  If the




thermodynamic activity at equipotency for narcosis  is constant within a




series, it would be impossible for  higher members of the series (i.e. higher




log P) to. be non-lethal  at water solubility.  However,  it  is a common




occurrence in toxicity tests with fish to find  chemicals non-lethal at




solubility, with hexachlorobenzene  and tridecanbl being  prime  examples.




Glave and Hansch (1972)  provide  the answer  to this  apparent  discrepancy.




They presented evidence  that the linear  relationship between narcosis and




lipophillicity as measured by log P (inversely  proportional  to water




solubility) was only linear for  the more water  soluble  chemicals.   The




relationship between toxicity and log P was  actually parabolic in nature and




was termed the "parabolic effect".  The  biological  effect  is actually modeled




by:




                  logฃ = -a(log P)2 + b  log  P•+  c




Thus, at small values of log P,  the equation may appear  linear.  However,.as




the range of log P is extended,  the relationship between biological activity




and log P reverses and'the t'dxicity actual.ly'decreases with  increasing log




P.




                                     13

-------
Since the log P and water solubility  are  inversely  but  linearly related,  the




expected physical toxicity for chemicals  with-a  large  log  P- could  be  above




the expected water solubility and  no  effects  would  be measured.




     The "parabolic effect" has been  interpreted  in many ways.   The  fact  that




toxicity decreases with  increasing log P  at  the  high end of the log  P scale




seems to indicate that the chemicals  either  diffuse slower through membranes




or that the fatty tissue act as a  storage media  and remove the  chemical from




the blood and key receptors.  Since the bioconcentration process is  linear




with log P, chemicals of high log  P may .produce  large  residues  in  fish at




water concentrations which are not toxic  to  the  fish.   This is  classically




illustrated with PCBs where the hazards to consumers of fish due to  the




accumulated residues far outweight the hazards to fish  at  environmental




concentrations.




     The "parabolic effect" may be more accurately  described as a  bilinear




effect resulting from competing membrane  transport  phenomena.   These  models




have been described by Yalkowsky and  Flynn (1973) and Kubinyi (1979).  The




bilinear model is expressed as:




                 log 1 = a log P - b  log(3P  + 1)  +  c




This is more consistent with the understanding of membrane transport  which




suggests that water-soluble-chemical  transport may  be  controlled by  the




membrane while lipid-soluble-chemical transport  may be  controlled  by  the




diffusion layer.                          j j




C.   Experimental Procedures




     The chemicals initially selected for -this study  include a wide  variety




of alcohols, ketones, aldehydes, ethers,  phenols, and  chlorinated  aliphatic




and aromatic hydrocarbons.  Figure 4  illustrates  the procedure  for selecting
                                      14

-------
              Figure 4.  Flow-Chart for Selecting Test Chemicals
SELECTION OF
TEST CHEMICALS
AVAILABILITY OF
I  PURE  CHEMICAL.
HEALTH AND SAFETY
     APPROVAL
                                                          PROCUREMENT REQUESI
                                                               (100 gm)
                                                          ANALYTICAL METHODS
                                                              DEVELOPMENT
                                                          PROCUREMENT REQUEST
                                                            TEST QUANTITIES
                                      15

-------
chemicals for QSAR modeling.  The QSAR modeling  group  selects  appropriate'




classes of chemicals  for consideration.  The  classes of  chemicals  are  mapped




onto a subset of  the  TSCA-  inventory  to select  chemicals  which  meet  the QSAR




criteria and are  in actual production in industry.  The  list  is  further




reduced to those  which can be purchased as  reasonable  cost  in  a  pure  state.




The list is then  screened by the ERL-D Health  and  Safety Officer to  insure




worker safety.  Small quantities of  the chemicals  are  purchased  to  develop




suitable analytical methods and conduct range  finding  tests  so that  an




appropriate amount of chemical can be purchased.   After  the  test is




conducted, evaluated  and entered into the data base, the data  are  used to




determine if additional chemicals in the class should  be tested  to  extend the




predictive capabilities of the QSAR  models.   The entire  process  in  Figure 4




requires approximately 4 to 5 months.  Therefore,  there  is  not the  rapid




feedback needed for a cost-effective testing  program and this  project  is




optimizing the testing resources by  testing several classes  of chemicals  in a




4 to 5 month "leap-frog" manner.  This permits the testing  teams to  test




.initial chemicals of  one class while the modeling  team is  selecting




additional chemicals  of another class.




     Exposure Systems.  Tests were conducted,  in  proportional  diluters  (Mount




and.Brungs, 1967) each with a dilution factor  of 0.6.  Each  diluter  delivered




12 L every 20 min cycle to six flow  splitting  chambers which  delivered 0.5  L




to duplicate test chambers.  Tests chambers were glass aquaria measuring  20 x




35 x 25 cm with a 9 cm standpipe, providing a 6.3  L volume.   Fluourescent




bulbs provided 28-34  ft candles of light at the  water  surface.  A  16  hr




light, 8 hr dark  photo period was used.             .   .  •
                                      16

-------
     Water Characteristics.  The water  source was Lake Superior  water




maintained at 25ฐC in the test chambers.  Hardness  and alkalinity were  either




measured in the control tanks during  the  test or  from a  separate line weekly.




pH was measured once during each test at  three concentrations  and  in  the




control.  Dissolved oxygen was also measured at three concentrations  and  the




control at 0, 48, and 96 hours.  Ten  tests yielded  a mean hardness  of 56.3




mg/L CaCC>3, a mean alkalinity of 42.2 mg/L CaCO}  and an  arithmetic  mean




pH of 7.5.  Dissolved oxygen was maintained at 60%  saturation  or higher.




Other chemical characteristics of Lake  Superior water are presented in




Table 1.




     Toxicant Additions.  Toxicants were  usually  introduced  directly  into the




diluter by a metering pump (FMI Corp.).   Some tests required delivery of  a




saturated solution of the toxicant in lake water.   The saturator systems




consisted of either a 20 L glass tank or  20 L glass jar  which  contained




constant volume of water and excess of  the toxicant.  The water  was stirred




continuously to maintain saturation in  the incoming water as the saturated




solution was pumped to the diluter.   Two  or three tanks  and/or jars,




connected with siphons were installed when toxicity was  very close  to




saturation.




     Test Organisms.  All tests were  conducted with fathead  minnows




(Pimephales promelas) from the Environmental Research Laboratory-Duluth




culture units.  Fish were hatched and reared in Lake Superior  water and fed




live brine shrimp at least twice daily.   Typical  fish tested were  30  days old




and weighed 0.12 g.  Twenty-five fish were randomly assigned to  each  of  .




twelve tanks in lots of five.  Fish were  not fed  during  the  96 hour tests.




Deaths were recorded at 1, 3, 6, 12,  and  24 hr -and  every 24  hrs  thereafter
                                      17

-------
   Table 1.  Chemical characteristics of dilution water




     from Lake Superior (all values in ug/liter except




                  where noted otherwise)
pH (pH units)
Hardness
Alkalinity
Chloride
Sulfate
Total phosphate
Ammonia-nitrogen
Nitrite-nitrogen .
Nitrate-nitrogen
Silica
Specific conductance
(ymho, 25ฐ C)
Aluminum
Arsenic
Cadmium
Calcium . .
7.7
46,000
40,000
1,300
3,800
2
N.D.a
1
220 :
2,100
93
8
0.7
0.04 '
13,000
Chromium
Cobalt
Copper
Iron
Lead
Magnesium
Manganese
.Mercury
Nickel
Potassium
Silver
Selenium
Sodium
Zinc

<0.1
<0.4
1.0
28.0
<0.2
1,400
5.6
; <0.01
<0.5
600
<0.02
0.6
1,300
0.9

N.D.
not detected
                            18

-------
for the remainder of the test.  Other observations  such  as  changes  in




equilibrium, schooling, and distribution of the  fish  in  the  tank were




recorded.




     Data Analysis.  Lethal effects of  the toxicants  were determined by




calculating LC50 values at several time intervals  for each  test.  All




calculations were made on the ERL-D1s PDP-11 computer using  the trimmed




Spearman-Karber Method (Hamilton et al., 1977).




     Analytical Techniques for Toxicity Test Chemicals.  Concentrations  of




toxicity test chemicals are regularly determined  in the  exposure chambers




during each test.  Water samples for the polar chemicals were  collected  and




analyzed either by direct aqueous injection on a  gas-liquid  chromatograph




(GLC) or by UV-spectroscopy.  GLC analysis was performed on  a  Hewlett  Packard




Model 5730A gas chromatograph with dual flame ionization detectors.  The




instrument was equipped with a 122 cm x 2 mm I.D.   column packed with  the




porous polymer, Tenax-GC (Applied Science, Riviera  Beach, Fla.).  The  flame




detector was operated at 300ฐC and inlet temperatures were  either 200  or




250ฐC depending on the compound being analyzed.   Nitrogen was  used  as  the




carrier gas and hydrogen and air were used for flame  operation.  Nitrogen,




hydrogen, and air flow rates were 15, 25, and 240 ml-min"'-,




respectively.




     Peak area calculations were performed by a Hewlett-Packard Laboratory




Automation Data System.  Gas chromatographic parameters  for  each compound




analyzed are listed  in Table 2.




     Standards for alcohols, ketones and aldehydes  were  prepared in Lake




Superior water or distilled water just  prior to  each  test.   No differences




were noticed in standards prepared in either lake  water  or  distilled water.
                                      19

-------
                         Table 2.   GLC Parameters and Analytical Quality Control Results for Test Compounds
Test Compound
Ethanol
2-Propanol
l-Butanol
1-Hexanol
1-Octanol
1 Decanol
1 -Methy 1 - 1 -propane 1
2-Chloroethanol
2,2,2.-Trlchloroethanol
2 ,3-0 1 bromopr opano 1
2-Mathy 1 -2 , 4-pentaned lo 1
Cyclohexanol
2-Phenoxyethanol
Acetone
2-Butanone
2-Octanone
2-Decanone
N> 4 -Methy l-2-pentanone
5-Methy 1 -2-hexanona
6-Methy l-hepten-2-one
2 , 4-Pentaned 1 one
Cyc 1 ohexanone
Cyclohexanone
Ethanol
Buta-nol
1 sotherma 1
GLC
Temp (C)
100
100
130
180
185
205
130
140
180
205
200
180
225
100
130
185
205
165
165
200
165
180
180
80
120
Retention'
Time
(Win)
0.95
2.11
2.80
1.13
3.52
3.9
1.1
1.7
1.6
2.7
1.0
1.7
2.1
1.95
1.6
2.1
2.9
1.1
2.0
1.54
1.8
2.0
2.0
1.6
2.3
Attenuation
5,120
1,600
640
40
3 ,:
2
160
8
16
8
12,800
40
10
3,200
640 .
8
2
160
40
16
8 •'."
80
80
4
•' 2
Working2
Range .
(ug/mT1)
2,000-25,000
1,800-30,000
300-10,000
15-400
30-32
0.6-10
60-1,900
12-300
48-480
5-130
2,500-25,000
70-1,200
25-900
1,500-25,000
600-5,000
5-80
0.4-12
220-4,400
120-2,000
18-200
7-390
80-1,300
80-1,300
10-200
10-120
Mean % Recovery
X + s.d.
n = 4-5
99.3 j* 1.2
-
-
103.6 +_ 9. 1
94.6 jf 4.9
104.6 +_6.2
95.8 _f 4.6
104.1 +_ 18.3
94.8 _+ 9.0
99.4 +_ 10.6
100.3 _+ 1.3
100.7 +_ 1.02
102.8 _+ 3;9
105.7 +_ 2.2
110.8 _t 4.1
121.4 i 4.4
89.7jH 9.6
104.4 +_ 3.0
122.4 _+ 2.8
112.2^5-2
108.5 _+ 13.8
101.6 +.2.3
100.3 _+ 3.9
104.4 +_ 1.7
99.6 + 3.3
% Agreement
of Dupl Icates
n = 4-5
98.1 +_ 1.4
97.6 +_ 2.2 ,
97.4 jf 2.2
95.6^3.6
97.8 _+ 2.2
97. 1 +. 3.3
94.8 _+ 3.8
95.0 ฑ_ 5.3
97.3 jt 2.9
94.2^4.3
98.9 _+ 0.9
98.7^0.4
94.8^ 4.4
99.0 +_ 0.3
98.4 jf 1.0
90.0 i 11.3
90.1 _+ 8.8
97.2 +_ 1.4
92.5 _+ 6.6
98.8 *_ 0.8
93.3 ฑ 4.5
98.6 +_ 1.1
98.0 _+ 1.0
99.3^ 0.8
96.9 + 1.9
2 working range  Is dependent on  toxlclty  exposure levels and not on GLC capabilities.

-------
For water soluble compounds (>100 ppm) standards were  prepared  by.direct




addition of the compounds into water.  Stock  solutions  of  compounds  with




lower water solubility ซ100 ppm) were made in  acetone  or  metHanoi.   Working




standards were made by diluting the stock  solutions with water.   The solvent




concentration in the aqueous working  standards  was kept below 10% (V/V).




Quality control consisted of a duplicate and  a  spiked water  sample with every




set of 6 to 12 samples.  Spikes were  prepared in the same  manner  as  the




standards.  The spikes provided information on  the accuracy  of  the standards,




the stability of standards with time, and  the reproducibility of  spike




preparations.




D.   Experimental Results




     During the first phase of this project,  in excess  of  80 chemicals have  .




been tested.  The absence of molecular descriptors such as log  P  in  the




literature precludes a detailed analysis of all data at this time.   However,




to test the hypothesis described in the previous sections, a subset  of 37




alcohols, ketones, ethers and alkyl halides which appeared to cause  death  by




physical toxicity and for which log P estimates could  be made were




considered.  The 96-hr LC50 for these chemicals are presented in  Table 3.




The data show that the LC50 (expressed as  mg/l) ranged  from 28,200 for




methano1 to 1.53 for hexachloroethane.  The log P for  the  chemicals  ranged




from approximately -0.85 for methanol to 5.6  -for tridecanol. The 96 hr LC50




for tridecanol is greater than the water..solubility of  this  chemical.




     The data in Table 3 are plotted  (converted to moles/liter) vs.  the




respective log P values in. Figure 5 which  is  similar to the  general




description of the aquatic "testing space" described in Figure  3.  The water




solubility line is calculated from the- equations presented by :Hansch,      '




excluding alkanes.  With.in experimental error,  the 96  hr LC50 decreases




                                      21

-------
        Table 3.  Summary of Acute Tbxicity Tests




           •96 Hour Tests with Fathead Minnows









Chemical Name	LC50 (mg/L)




N-Butyl ether               ..             32.5




Phenyl ether         .                     4.0




Tetrachloroethylene                      13.5




1,1,2,2,-Tetrachloroethane               20.3




Pentachloroethane                         7.3




1,1.,2-Trichloroethane            .        81.7




1,1,2-Trichloroethylene                  44.1




1,2-Dichloropropane                     139




1,2-Dichloroethane                      118




Hexachloroethane                          1.5




1,3-Dichloropropane                     131




Methanol                        .     28,200




Ethanol                              14,800




2-Propanol                            9,640




1-Butanol                             1,740




1-Hexanol                                97.2




1-Octanol                          •      13.4    '




2,2,2-Trichloroethanol          •  ••      298




2,3-Dibromopropanol                      71




2-Methyl-l-propanol                   1,460




2,4-Pentanedione                         96.0




2-Methyl-2,4-pentanediol             10,700
                           22

-------
Table 3.  (Continued)
          Chemical Name
    LC50 (mg/L)
          Cyclohexanol

          2-Phenoxyethanol

          1-Decanol

          Tridecanol


          Acetone

          2-Butanone

          2-Octanone

          5-Methyl-2-Hexanone

          4-Methyl-2-Pentanone

          Cyclohexanone

          2-Decanone

          3,3-Dimethyl-2-butanone

          5-Nonanone

          Acetophenone

          3-Pentanone
    705

    346

      2.3

No toxicity at
   solubility

  8,140

  3,200

     37.0

    158

    509

    630

      5.7

     86.0

     31.0

    161

  1,530
                                     23

-------
Linearly with  increasing  log  P  (i.e.  toxicity increases linearly with log P)




at least until  log P ฐ  3  is reached.   In  this region of the testing space for




these chemicals  producing narcosis,  the 96  hr LC50 appears to be a constant




proportion of  the expected water  solubility as suggested by the previous




discussion.  Although,  it is  premature to propose a constant factor to




estimate toxicity of narcotics,  the  LC50  for the group of chemicals appears




to be approximately 0.017 of  the  water solubility.




     The LC50  values for  chemicals'which  have log P values near 4 appear to




be much closer to the water solubility.   I-t is proposed that this is the




•region where the "parabolic effect"  is evident.   The data point plotted just




above the solubility line is  that of tridecahol  which is non-lethal to fish




in a saturated solution of the  chemical.   The LC50 for decanol is below water




solubility (log  P • 4.07) which suggests  the "toxicity line" and the "water




so-lubilty line"  cross between log P  of 4  and 5.5.  Chemicals with log P




values in this  range are  scheduled  for testing next month to better define




the toxicity relationship in  this range.   At the same time, a computer




program incorporating the bilinear model  is being developed specifically for




these data.




     In preparing Figure  5, seven chemicals appeared to have anomalous




behaviour.  Four of these were  found to have incorrect log P values reported




including an apparent typographical  error of log P-.= -1.70 for decanol.




Corrections of these errors -removed  apparent discrepancies.  The apparent




outlying point  in Figure  5 with an LC50 of -3.0  is that of 2,4-pentanediol.




The log P value  for this  chemical was estimated  using the substituent




constant approach.  Since 2-methyl-2,4-pentanediol is not anomolous, it is




reasonable to  expect these discrepancies  can be  evaluated when more reliable




log P values are available.




                                      24

-------
                0
                -D
N3
Ol
             -3


|moles/liter{ .  -,


             -4





     '  '  '   ,-5
                -7
                -8
                   -2
dฉ   ฉ
?
                                      j   'i'.l.'J   '    1   '   i
                                                                                  I	1	I
                                                  LOG    P

                                Figure 5.  Relationship oiTLog LC50 and  Log P for Narcotics
                                                     8
                                                                                              -5
                                                                                              -6
                                                                                           -7
                                                       -8
                                                                                                    jmoJes/8itef|

-------
     Two chemicals which were clearly below  the  "physical  toxicity"  line  were




2-chloroethanol and ir-amino-2-propanol, both of  which were much more  toxic




than would be predicted from this model.  'These  chemicals  (not plotted) both




produced symptoms different from narcosis  including bursting  of the  eyes  from




the fish. ' These chemicals  are  somewhat similar  in that  a  single  chemical




substituent is located on the carbon adjacent to the carbon with  the  hydroxyl




group.  Tests are underway  to evaluate the relationship  between substitution




patterns, metabolism, and lethal effects.  This  may be a good example  of  how




physical toxicity is masked by  an irreversible chemical  toxicity  which is




observed in concentrations well below concentrations needed to produce




narcosis.                                        •
                                      26

-------
V.   MATHEMATICAL METHODS FOR QSAR




          The development of Quantitative Structure Activity Relationships  (QSAR)




     and pattern recognition techniques for drug design have grown  rapidly  since




     the Hansch and Free-Wilson methods and discriminant  analysis (Jurs  et  al.




     1969) were introduced.  After computer time and  statistical expertise  became




     available, refinements and new methods were added to the organic  chemist's




     analytical techniques.  Free-Wilson, linear free-energy and bilinear models




     are separated from the other techniques only  for convenience since  all models




     and analyses share many common features.  In  discussion of each approach,.the




     development of the mathematical model rather  than chemical application should




     be stressed.




          No matter how elegant the data treatment, the results will only be




     useful if the chemical and biological thinking is sound.  Statistical




     significance will never replace intuition and logical  thinking.   Within a




     homologous series, regression analysis is the simplest method  to  show




     dependence of one variable on another.  Usually  the  coefficients  of




     regression analysis can be understood in terms of the  independent parameters




     which allows the chemist to relate these parameters  to biological




     mechanisms(s) and to make new predictions from the regression  equations.   If




     there is a choice of multiple regression methods, all  possible subsets gives




     the best results.  If nonlinear regression is available, the bilinear  model




     best describes the apparent parabolic effect  of  biological activity on




     hydrophilic parameters.




          When pattern recognition is used to describe large, diverse  data  sets,




     the first choice of methods should be factor  and cluster analysis.'  Martin.




     and Panas (1979) propose useful criteria and  steps for. using these  methods  in




     series design.  Discriminant analysis can be  useful  when seeking  broad




                                          27

-------
classification, but consideration  should  be  given  to  the  limitation and




interpretation of results.   Finally,  all  statistical  analyses  should include




correlation tables of -descriptor variables  and -development  of  equations




(Ogina et al., 1980; Wu  et:al;, 1980).




     Free-Wilson Model.  The Free-Wilson :model  assumes .that each substituent




makes an additive contribution to  the biological response (BA) .   The model is




expressed in equation  1,
where u is an overall  average, and  Gฑ:  is  the  contribution of the




substituent X^  in  position  j.   With  the  additive  assumption, each




substituent effect  at- each  binding. .site.  can;be' estimated  by multiple




regression analysis.   The biological  and -mathematical  assumptions and.




implications as well  as.- modifications , of  the  Free-Wilson  model  are summarized




by"Kubinyi and Kehrhahn- (1976a,b) .  They: have shown, that  the FujitarBan model




has'; several advantages . over the .classical.! Free-Wilson  and Cammarata, models.




In summary, the Fujita-Ban  model  require.s. fewer restrictions on the-:. data




matrix, gives estimates of  the theoretical predictive  activity  value of the




parent compound, shows lack of linear -dependence  and has  a close mathematical




relationship to the Hansch  linear- free energy model.  Schaad and Hess (1977.)




reviewed the theory of linear equations  and suggest ways  to detect .and




circumvent linear  dependence difficulties for all variations of the




Free-Wilson model.  Two general- problems  with the classical or  modified




Free-Wilson model  is  the  inability to  generalize  biological mechanisms for a




class of chemical  compounds and' the  difficulty of preparing the data matrix




before analysis.




     Hansch Linear  Free-Energy Model.  The linear free-energy model has been




used and modified • numerous  times .   The model  assumes a linear and additive




                                      28     :             : '-

-------
contribution from the hydrophobia, electronic, and steric  properties  of  a




drug or chemical compound on biological activity.  Usually the  log  of the




inverse of the biological response (log ฃ) is correlated to various




physical-chemical parameters.  The general form, equation  2,




            log J =•  a^i +  bjEj +  ckSk + d




where the hydrophobic (H^), electronic (Ej) and  steric  (Sk)  effects are




estimated by multiple linear regression coefficients  a^, bj,  and  c^




respectively.  The ease of calculation, availability  of canned  programs, and




potential for describing biological mechanisms led to numerous  publications




of correlation studies.




     Most problems that arise  from using multiple  linear regression are




related to the choice and number of variables that are  used  to  describe  the




biological response.  The ratio of observations  to variables  should be




greater than five.  Topliss and Edwards (1979) have developed control charts




for determining the probability of a chance correlation'based on  the  number




of observations and variables  used in regression.  These charts are based on




the total number of variables  that were tried as well as taking into  account




the collinearity of the descriptor variables.  Unger  and Hansch (1973) and




Otto et al. (1979) have stressed the point that  the investigator  should  not




maximize the correlation coefficient at the cost of prediction  by including




more variables than can be supported by the data.  Draper  and Smith (1966)




offer numerous statistical methods for evaluating  the usefulness  of




regression coefficients and equations.  The interdependence  between physical




parameters must also be kept to a minimum if correlations  are going to




describe different biological  mechanisms of action.   Craig (1971) and Hansch




et al. (1973) have tested the  interdependence of numerous  substituent




parameters by correlation analysis.




                                     29

-------
      To' further improve  the interpret at ion  of  correlation s-tudles,  linger and


Hans'ch (1973) have presented criteria' for' selecting the' "best equation".


They  have also listed  the"minimal information  that needs to be  included in


all multiple linear  regression analyses".  Otto et  al.  (1979) have  described a


method that validates  the final regression  model.   By applying  the


"leave-L-out" technique,  they were 'ab'le to 'show' remarkable stability  in 'their



regression equations by  keeping'' the nuiab"er Jof  variables to a minimum.


      Recent papers by  Iwam'ur a (1980) and Wu "et !al." '(1980) have  included


useful tables with a comple'te" cross 'correlation of 'variables and  the


development of equat'ions*."  Heymans">et al. '(1980) 'fdundHa global stefic  effect


in MR'(molar refraction)' but "coii'd have improved their interpretation 'by--


using 'an indicator variable' for steric ef feet's "atM* an independent  value for


es't'imating'-'the hydropKob'ic 'ef'fe^ct'XB'^ag'i"^^ 'aft1, "198b!;'Wu'et al . ,  1980) .
Botli 'Biagi et al. anS Wu' et*"atV 'fdu'nd'1 tn^a't^a^stericr'indicitor' variable 'was"



superior than MR since 'MR waV'higty •Xorre''^^^' with log" P (Hejrmaris ''et al'. ,



1980  also reported  this  result) .  Thev"in? t'.n^iiH "t'"i    '' ' •   '  .        •••••;.
authors  to test for  specif ic  "steric1 "effects Jthat  would 'otherwise  be  averaged'



out by MR.  These analyses 'are Tsim]p"l>if ie"dr"verssi%ns>'of the mixed model



suggested by Kubinyi and" Kehrh''^^!!'^!^^^; "wh'ere "a Fujita-Ban arid Harisch model



are combined.  Biagi et' al. (l^St))' 'als'o  fou'nd' "that the regression technique
of  all possible subsets "'was''superiorr"to'"forwar'd "stepwTse regression.  • One


final  problem with  variable select ion''is" the'^sp'read'^or'standard  deviation of


the variable.  Unless  there is "an adeqiia'te'range' for each variable  used in


the "regression model  then 'there is little chance"of detecting  the  biological


response to 'that variable (Martin arid" Panas^ 1979).


      Bilinear Model.   The'linear refatioo'snip between biological activity and


•the hydrophobic feature  (log^P) of a  compound will  not continue  indefinitely.


                         •:             30              "  .-,:.;

-------
Hansch and Fujita (1964) suggested a parabolic regression model to explain




the nonlinear biological response to the increasing hydrophobic effect of a




compound.  To account for the strictly  linear response  for  lower  log P




values, Kubinyi (1977) has proposed a generalized bilinear model, adapted




from the McFarland (1970) probability model,




                log ฃ = a log P - b log (BP + 1) + C




This generalized model generates curves with unsymmetrical,  linear ascending




and decending sides and a parabolic portion around the  optimal value for  log




P.  Through residual and statistical comparisons of 57  data  sets, Kubinyi has




shown the bilinear model superior to the parabolic model proposed by Hansch




and Fujita.  Kubinyi found situations when the parabolic model gives a better




approximation of the biological response (few data points or  small range  for




log P).  This result can be attributed  to the number of variables (4 versus




3) and calculation procedure (nonlinear regression).




     Drug Design and Classification.  The demonstrated  relationship between




biological activity and physical-chemical properties of chemical compounds




has led to the search for methods that  identify promising new leads.  These




methods can be separated into two categories, pattern recognition and




non-mathematical searches.




     (a)  Pattern Recognition - Within  the general field of  pattern




recognition, three categories can be found; linear discriminant analysis,




cluster analysis and factor analysis.   One of the first uses  of discriminant




analysis was to classify mass spectra (Jurs et al., 1969).  Methods used  for




discriminant analysis include positive  feedback functions (linear learning




machines), linear discriminant functions, K-nearest neighbor,  and threshold




logical units.  There are numerous examples that use discriminant analysis to
                                     31

-------
classify biological activity  into two or three categories  (Chu  et  al.,  1975;




Rasmussen et al., 1979; Henry and Block, 1979).  All method  of  discriminant




analysis, except K-nearest  neighbor, attempt  to  divide  a multidimensional




data array by a hyperplane.  K-nearest neighbor  measures the distance between




two or more positions  in hyperspace  (Albano et al., 1978).   To  improve  the




linear separation and  reduce chance  correlations, various  data  treatments




were devised that verify the usefulness of the linear discriminant  function




(Kowalski and Bender,  1974; Weisel and Fasching, 1977).  Before analysis data




should be autoscaled,  i.e.  normalized to a zero  mean and variance  of one.




After scaling, a subset can be used  as a training set and  another  subset for




verification of the discriminant  function.  Another common way  to  verify the




discriminant analysis  is by the .Jackknife (leaye-one-put)  iterative procedure




(Weisel and Fasching,  1977).




 "  •  There are numerous factors  that need to  be-considered before'doing a




discriminant analysis.  Frank et  al. (1965) and  Morrison (1969) have




summarized the difficulties in interpreting results from a "canned"




statistical program.   These considerations include equality  of  numbers  and




covariance matrices of observations  between classes and the  upward  bias that




results from verifying results with  the same  measurements  used  for




classification.  It is also true  that the discriminant  function may not be




the "best" equation, since  many  solutions- are possible  for any  data set.




Gray (1976) found that the  "noise" feature will  influence  the  final




discriminant function.  Usually  the  ratio of  observations  to variables  should




be greater than three  to avoid unstable (data dependent) discriminant




functions.




     A good example of how  to properly use and discuss  a discriminant




analyses is given by Ogino  et al. (1980).  They  have included  tables of




                  • .    ,              32

-------
  variable correlations and the development of the  linear discriminant




  function.   These tables, which are similar to the tables used by Biagi  et  al.




  (1980),  and Wu et al. (1980), are absolutely necessary  for  comparison and




  interpretation of the discriminant analysis.




       Hansch and linger (1973) were the first to use cluster  analysis  as  a  tool




  to find  the most different analogue for a lead molecule.  They  used




  extrathermal substituent parameters as their factors  and chose  one compound




  from each  cluster for synthesis.  Cluster analysis uses the  correlation




.  matrix to  find patterns of similar structure.  There.are numerous methods




  that group variables by the size and interdependence  of their correlation




  coefficients.  The outcome of any cluster analysis is dependent on the  choice




  of information that goes into the analysis.  Hopfinger  et al. (1979) were  not




  able to  explain toxicity except in terms of molecular weight.   This




  illustrates the need to choose factors that are orthogonal  (not highly




  correlated) and descriptive of the chemical factors  influencing toxicity.




  The indiscriminant inclusion of meaningless factors  (unrelated  to the




  biological property) will lower the predictive capability of the analysis  and




  increase the possibility of getting a useless but successful classification.




       The branch of statistical science called factor  analysis has been  used




  by psychologists since 1904.  It wasn't until Cammarata and  Menon (1976)




  suggested  a classification scheme for handling large  bodies  of  information




  that factor analysis was used to describe toxicity data.  Their flowchart




  included factor analysis, discriminant anlaysis,  cluster analysis, and




  finally  correlation analysis to follow up a new lead  compound.  Several




  authors  (Saxberg et al., 1978; Albano et al., 1978;  Dunn and Wold, 1978) have




  compared factor analysis with discriminant, analysis  and. h-ave-'proposed a




  hierachy of classification.ability with discriminant  analysis'on the lowest




                                       33

-------
levei and factor analysis  on  Che highest.- Martin  and Panas  (1979)  have.




proposed another strategy  for exploring new "leads"  in  drug  design.   Based  on




a four part criteria, Martin  and Fanas suggest  that  factor  analysis and




cluster analysis will measure independence and  uniqueness of each  chemical




compound.




     Just as  in any  pattern recognition method,  selection of variables  is




important.  Martin and Panas  (1979)  state that  the standard  deviation of each




variable be greater  than one.  They  also  suggest that the variables be  tested




for independence by  factor analysis.  Malinowski (1977a,b)  has  developed




several methods for  evaluating the usefulness of each factor by examining  the




error in the  data matrix.  Martin and Panas (1979) selected  the number  of




factors used  for classification by examining eigenvalues.




     (b)  Non-Mathematical Searches  - Two non-mathematical  approaches to




optimize a  lead chemical are  the Topliss manual method  (Topliss, 1977)  and  a




Fibonacci search (Santara  and Augary, 1975).  Both methods  select  and rank  a




set of analogues by  molecular properties  usually found  in a Hansch




correlation study.   The most  active  analogues are  selected  for  synthesis




based on the  relationship  between biological activity and physical




parameters.   The. advantage of both methods is that neither  requires a




computer for  analysis.  Martin and Panas  (1979) evaluated the Topliss manual




method by factor and cluster  analyses.  They'found that the  .strategy w.as




useful but  not  ideal for finding the least!number  of compounds  that covered




the substituent space.




     Perspective.  The  literature is replete with  computer  programs with




intriguing  names which  are reported  by their developers as  QSAR methods.   In




general, the  programs consist of one or more of the  techniques  found in




systems such  as BMDP which are available  in many computing  centers.  In some




                                      34

-------
cases, such "canned" programs are tailored wich respect  to  input/output




routines to aid the analyst.  The specific program used  is  dependent  on  the




types of questions being asked.




     This project will be evaluating many of the techniques  discussed above




using systematic data sets to demonstrate the relative contribution of each




in addressing EPA needs.
                                     35

-------
VI.  MOLECULAR DESCRIPTOR  GENERATION        ,             .




          Attempts  to  predict  biological" or chemical'act'ivities  from structural




     characteristics have  involved  well  over 30 different  characteristics  which




     are loosely called molecular descriptors.   Common'descriptors,  or predictors,




     of activity include log P.,  Hammet constants,  Taft steric  constants, molar




     volume, molar  refractivity,  CIDS keys,  molecular  orbital  parameters,  and




     connectivity  indices.  In many cases,  each type of  descriptor  includes  a




     family of parameters.  Some descriptors can be  selected  to  model certain




     activities because the descriptor  is an important factor  mechanistically in




     the process being modeled.   However, many  times  the descriptors are  selected




     on a statistical  basis and  their  importance are rationalized  as the model




     develops.




         -To keep  this QSAR project directed toward  the  Agency's need for




     "screening thousands of industrial chemicals,  several  constraints were placed




     'on the selection  of molecular  descriptors.  The  first is  to minimize  the




     number of molecular descriptors which  are  used  as independent  variables in




     the QSAR models.  The literature demonstrates clearly that  there are  few




     •descriptors readily available  for  large numbers of  industrial  chemicals, and




     building models which require  several  descriptors further reduces the number




     of chemicals  for  which estimates of activity could  be made  from the models.




     the second is  that molecular descriptors be selected  where  possible,  which




     can be calculated directly  from  the structure of  the  chemicals.  This would




     permit extensive  computerized  evaluations  of the  QSAR models  on large numbers




     of chemicals.
                                           36

-------
A.   Calculation of Log P




     The discussion in previous sections of  this  report  has  demonstrated  the




usefulness of log P in estimating toxicity and  BCF.   Log P has  been  selected




because it is a measure of the hydrophobic properties of the  chemicals  which




is directly related to transport and binding  in membranes.  Moreover,  the  log




P for thousands of chemicals can be calculated with  adequate  accuracy  from




structure.  A project was developed with Dr.  Al Leo  and  Peter Jurs to




computerize the log P calculations.  When a  lack  of  funds prevented  the




timely initiation of the project, the Office  of Toxic Substances  funded the




project.  It is anticipated the first version of  the  program will be




available to EPA in November, 1980.  This project  stands ready  to evaluate




the calculations on a subset of 15,000  industrial  chemicals  when  the program




is received.




B.   Calculation of Molecular Connectivity Indices




     Molecular connectivity indices are a family  of  constants which  result




from the application of graph theory to molecular  topology.   A  review  of  the




calculation and examples from use of these indices in physical  and medicinal




chemistry has been presented by Kier and Hall (1976)  and is  beyond the  scope




of this report.  Briefly, the connectivity indices are calculated by




subdividing each molecule into a series of subgraphs  with the atoms  as




"vertices" and the bonds between them as "edges".  The subgraphs  are




categorized as paths, clusters, path-clusters,  and chains which are




self-explanatory.  The subgraphs are further  divided  into connectivity  orders




which are the number of edges.  In general only the  first four  orders  of




indices have to be used in QSAR due, in part, to  the  difficulty in




calculating the higher order indices for multicyc.lic'molecules.   For




example,




                                     37

-------
 endrin has in excess of .1,000 sixth order subgraphs, each of which muse"be - .




 analyzed individually and summed to give the individual sixth order, path,.




 cluster, path-cluster, and chain indices.




      Kier and Hall further divided the indices into simple and valence-




 corrected indices.  The simple indices are calculated as  if all  the  atoms




 were carbon atoms but all bonding was the same as  in the  original molecule.




 The valence-corrected indices apply empirically derived correction factors




 for each type of heteroatom present in the molecule.  The valence correction




 is intended to describe the variations in electron density around the




 heteroatom which may contribute to the reactivity of the molecule.   The




 terminology employed is X to represent the simple  indices and Xv to




'represent the valence-corrected indices.  An additional superscript  denotes




 the order such as ^Xv being the fourth-order valence-corrected indices.




 If no subscripts are present, it is implied that the index is the sum  of all




 the paths, clusters, etc. of that order.  If individual indices  are  used, the




 subscripts "p", "c", "pc", and "Ch" denote the path, cluster, path-cluster,




 and chain indices, respectively.  As an example,  Xvpc represents a




 sixth-order valence-corrected path-cluster index.




      A program has been developed by this project  to calculate indices




 through the tenth-order indices for molecules including the multicyclic




^molecules such as mirex and endrin.  The program calculates the




 valence-corrected indices, the simple indices which is the molecule  reduced




 to a hydrocarbon of  identical topology, and framework indices which  is the




 molecule reduced to  a saturated hydrocarbon of the same topology.  An  example




 of the data for molecular connectivity indices from this  program is  presented




 in Figure 6. The molecule considered is 2,2-dichloro-l-(2,4,5-trichloro-




 phenyl) ethanone which has the structure indicated in the upper  left.   The




                                      38

-------
 !- i-( >, •! .s-iv iciii.'ii.iii'HK'm. ji-'i
                                                                         K'lu UN 24-SF:f-bO
                                    ฃIIIIGKปHIIS:
a ง CI
0
0 1
PATii 14 14
CI.
CM
TllTAl. 14 14
tt Mi ••':•' ft CliKKLCrKI) OIllNiiCTlV ITY IMllTlKS:
0 1




•
MitPi.K <'UMnKCn VI TV 1 Mil 1 C 1 KS> :





m 2. i I'LiTM.:. 	 W.'lj')^ b.20lf.
51ฐ"! ri'AMc.piH'ri i:n,,Ni:.rri iMTY ninirK.S:


' /'fill



234



4.<ป7f>y 4.122S 3.6B13
~i 3 4
4.4313 i.USjti I.6l7b


.
4.4313 3.f"ซS 3.3bOO
2 3 4
h.lVO-/ 4.Mln7 3.1147




•-lIHDt.K---
2 3
4
2n 24 2u
b 0
0 0
20
20 30 4o
S
1.3073
0.27H3
o.oooo
i.87yo
3.^Sy7
5
l.lluO
O.OOuO
1.7040
J.07IS
S
0 . b 3 3 9
•u. 00(10
3.7001

b
o.fiyon
o.oooo
2.84lu
3.4b?a
6
O.S09J
0.0000
0.020o
2.<>b5-*
t>
0.0000
O.OSSo
1 1 A*SII
b 6 7
1 u 24 1 H
•1 0 i
u 1 4
3ป H7 133
73 112 1-57
7 a
0.^8B4 O.Obfly
O.U365 O.OOOi)
o.oC57 O.I38S
2.7798 7.5ttOb
3. l'ป04 2.HUBO
fKEDUCRu
7 U
0.233b 0.0703
0.046S O.OOOO
O.U721 0.1071
b y
10
920
010
u 13 iy
184 2CH 179
201 224 19U
0
0.01/4
0.0171
O.l3b7
2.2046
10
0.0000
0.0000
0.13B6
1.6133
2.37bO t.7biy
•in imiHOCArtHuNJ
4
0.0120
0.0170
O.lOoS
2.b707 2.2784 I.b3b1
(i(Fi)MCKI/ TO SATUI'ATtn lllDhf
7 b
0.i>?'*S 0.2091
0.1341 0.0000
0.19t>7 0.313b
o.2f>63 6.2bft3

q
0 . 0 3 V 0
0.0556
0.35S/I
b.4fiyi
K U ') tt fi
10
o.oooo
0.0000
o. i i 3y
1.1b42
1 .2681
10
0.0000
o.oooo
0.40M
3 . 9 '/ 7 y
4 Ixdli
Figure  6.   Connectivity  Indices for  2,2-dichloro-l-(2,4,5-trichlorophenyl)  ethanone

-------
table in Che upper right  summarizes' the number of  subgraphs  in  the  molecule.

The molecule has 14 atoms as evidenced.by. the 14 zero-order  subgraphs.  .The 6

third-order clusters are  located at the six carbon atoms which  are  substitut-

ed.  The phenyl ring obviously accounts for the sixth-order  chain.  The  lower

half of Figure 7 presents the individual and summed indices  for this

morecule.

     Dr. David Weininger  and Brad Greenwood at ERL-D have  a  complete  package

of programs called INDEX  which allow structures of chemicals  to be  entered

either manually on a remote TTY or with a graphics terminal,  to check the

molecule for proper valencies to calculate a complete  set  of up to  tenth

order indices, file the files along with other data on chemicals for  input  to

programs such as BMDP or  other statistical packages.   Approximately 800  chem-

icals which are part- of one.of the QSAR'data;bases have-been-entered'into  the

system.

     The INDEX system is  a  group of programs, command  files,  and structured

data files which currently  allow:

     -interactive graphic molecular structure specification,

     -verification of "foreign" files  of molecular structures,

     -hard- or CRT-graphic  display of  chemical structures  on files,

     -computation of connectivity  indices of the Ocn to lO'*1 orders',

     -storage of connectivity indices  for up to 20,000 chemicals in
                                                         i  •
      compressed form (the  master  file), ' , .            ^

     -a variety of options  for retrieval from the  master file,-

     -generation of master  file table  of contents  ordered  by location or by

      empirical formula,

     -automatic detection of duplicate entries on  master file,

     -encoding an additional parameter along with  the  connectivity  indices  of

      selected compounds  for subsequent BMDP analysis,

                                ,   . 40      .  . ..  •

-------
                           Figure 7.    Flow-Chart  of  Program INDEX 5
        FLOWCHART OF PROGRAM
      Zero out  all  arrays.
               T
Attempt to read  in molecular structure
     from .[IF-type input file.
  Set up topologica! matricies.
        Check  validity.
    Make valence  corrections.
   Calculate  empirical formula.
    Generate all  exclusive
    combinations  of atoms.
              Is
    NO  /suograph  \ YES
          connected'
                                   Write results  to output
                                      file INDXOUT.DAT
                                       Output  results to line
                                       printer 1 molecule/page
Evaluate the entire
subgraph formed  by this
combination of atoms.
                                                                        Break all  exclusive
                                                                        combinations of bonds,
                                                                        creating disconnected or
                                                                        less cyclic subgraph.
                                                  Evaluate  the hidden
                                                  paths  in  the chain.
                                                                       YES
                                                                        Evaluate the suograph
                                                                            which was hidden.
                                                        41

-------
     -locating the nearest  neighbor  (and most distant) of  the  master  file




     compounds in connectivity  factor space,




     -and several other  utility  operations.




Additional capabilities  expected to  be  implemented by  11/80  include automatic




master file backup and a routine for determining  if  a  given  molecular




structure falls inside or outside  the hypersolid  defined by  a  group of




molecules in  factor  space.  A general description of the INDEX programs  and




files follows.




     Creating and Handling  INDEX Input  Files.   The  first step  in the  process




of molecular  structure description is unambiguously  specifying a molecular




structure.  The format used in  INDEX is an extended, linearized connectivity




matrix; files containing lists  of  molecular  structures  in  this format can be




identified by the file'extension ".IIF".(Index  Input, Format).   Although  not




yet tested, the translation of  "foreign" topological matrixes  to IIF  format  •




should be straightforward.




     Program MOSS (MOlecular Structure  Specification)  is intended to  be  the




primary means of initially  specifying molecular structures.  MOSS is  an




interactive graphics  program designed to be  run on  a Tektronics 4025  Graphics




Terminal.  The program outputs  the structures of  graphically displayed




molecules in  a verified  IIF format file.  Verification includes checking atom




types and valence validity  and  extended topological  matrix symmetry.   The




user may specify the  entire structure or modify any  structure  stored  on  an




IIF file.  It is up  to the  MOSS  user to verify  that  the  correct name  is




associated with a molecular structure.




     Creating and Filing Connectivity Indices.  The  program INDEX5  reads




molecular structures  from an IIF file,  computes molecular  connectivity




indices, prints tabulated results, and  writes  the results  in the file




                                     42

-------
INDXOUT.DAT.  The program uses a general algorithm  for  connectivity  index




computation; the following limitations are imposed  by available valence




corrections, computation time, and storage effectiveness  considerations:




     -allowable non-H atoms: C, N, 0, S, Br, Cl, I,




     -maximum number of non-H atoms per molecule =  50,




     -highest order connectivity index computed = 10, and




     -highest allowable cyclic order of 11-atom-subgraph  =  5.




     The INDXOUT.DAT file contains the 100 character name,  the coded




empirical formula, and the following information for each of. the  34  possible




subgraph types of orders 0-10: the number of unique subgraphs, the




valence-corrected CI, the simple CI, and the framework  CI.




     The formatted sequential nature of the INDXOUT.DAT file  makes  it




inefficient with respect to disk storage space.  Therefore,  the information




contained in INDXOUT.DAT is transferred to a compressed master file




(MINDEX.DAT) by the program LOADEM.  A successful INDEX5  run  is normally




followed by a LOADEM ruii.  This two step process is used  to ensure  the




integrity of MINDEX.DAT in the event of a fatal error during  INDEX5




execution.




     Accessing Connectivity Indices on the Master File.   All  user access of




the master file must be done via program control because  the  file contents




are written in binary.  The master file, MINDEX.DAT, is set  up for  "direct




access" which allows any record to be directly retrieved  by  specifying  a




location.




     Three programs are specifically designed to retrieve.master  file




contents for specified chemicals.  Programs GETEM and FILEM  are essentially




identical programs except for the output file: GETEM prints  the retrieved'




entries in their original tabular form, FILEM writes entries  into the  file,




                                     43

-------
 SHMIND.DAT.  Boch programs  require  that  the  master file locations be


 specified and provide a means of  locating  an entry-by empirical formula.  Any


 number of entries can be  retrieved  during  a  single GETEM or FILEM run.


      A third retrieval program, PREPEM,  is similar to FILEM but also has


 these additional features:  • .. ,


      -The input file can  be  the master  file, the file INDXOUT.DAT, and/or any


       other file in the INDXOUT.DAT format.    ,


      -An additional molecular descriptor can be  specified (eg., "LOG P") for


       each compound  included in  the  output  file.


      -The molecular descriptor name and  values are stored in the  1'ast 20


       characters of  the  100 character  name.


      -Any output file.name  may be . specified.  This file can be directly used

?f     '                 - ~  •      ..  ~   '     •      .,-...     _               '- •
       by any BMDP program as 'a data file.  •'•.   '•-'.:'• .-.V.'-,..-:';..'•'.'•       •'   -


      The program INTDIST  is  .used  to find which.compounds in .the master file


 .are'most and least distant  from a given  chemical in the 34-dimensional


 Balance corrected connectivity space.   The user  can select the "given"


 compound from the master  file on  any  INDXOUT.DAT-type file.  INTDIST will


 •list the nearest and farthest compounds  :in the master file, automatically

  .   "**          '  .      ' '              '  .•     •       •:•.'•••      :
 discounting duplicates.                      :


      Operation of the INDEX5 Program.   INDEX5 is the fifth version of a


 series of programs which  compute  molecular connectivity indices for organic


 imoleculesi  As discussed  above,  INDEX5  us'es  a general algorithm for

         .-.'•'.'"•'.•        •: ,    •'   •  '      ' •         .'• •            '   ซ

 connectivity index computation.   A flowchart of  INDEX5 is presented in Figure


 .8.   Each molecular structure in  an INDEX Input Format (IIF) file  is


 rtransformed initially  into  two topological matrices: the simple^,conhection\,


 matrix which contains only  1's  and  O's  (T),  and  an extended matrix which
                                       44

-------
contains the topoLogical valence for each connection  (PT).  Valences  are  then




calculated based on both topological valence (number  of connections  to  a




vertex) and elemental valence corrections.




     The INDEX5 algorithm is vertex oriented.  All exclusive  combinations  of




up to 11 atoms are generated and the connectivity  index contributions  are




evaluated for the subgraphs which include all atoms in-the  combination  only.




The decision to evaluate up to 10th order connectivity indices  requires that




up to 11 atoms must be considered at a time.  An efficient  pseudo-recursive




(stack based) algorithm is used to generate the combinations.   CONECT's high




efficiency is obtained by automatically suppressing generation  of




disconnected combinations.




     Each connected combination generated by CONECT is evaluated by  the first




level routine EVAL.  The entire subgraph (represented by  all  atoms and  bonds




in the.combination) is evaluated.  The term "evaluate" is used  here  to  mean




"determining subgraph type and order, incrementing the ennumerative  counters,




computing the subgraph index contributions, and adding these  contributions  to




the appropriate index accumulators".  If the subgraph is  cyclic, it  will




contain a series of hidden subgraphs which must be evaluated.   In  the  case  of




a simple chain of N atoms (all topological valences = 2), EVAL  evaluates  the




N hidden paths of N-l order.  In more complex cases,  EVAL determines  the




number of cycles in the subgraph and passes the hidden subgraph evaluation  to




the routine CYCLIC.




The subroutine CYCLIC evaluates hidden subgraphs by using the following




theorem:  "If an edge is removed from a graph which is K  cyclic, the




resultant graph is either (K-l) cyclic or disconnected".  Combinations  of




bonds (edges) are broken such that not only are .the broken  bond combinations
                                     45

-------
exclusive, but so are  the  combinations  of  remaining  bonds.   Subroutine CYCLIC


is capable of breaking up  to  five edges  of a  subgraph  simultaneously enabling


analysis of 5-cycle, 11-atom  subgraphs.  The  above algorithm circumvents the


difficulties associated with  multiple cyclic  subgraphs;  by  breaking up to K


bonds of a K cyclic subgraph  one  is  assured that  all possible hidden


subgraphs will be uncovered (whether or  not they  are in  themselves  cyclic).


Responsibility for evaluation of  the generated  subgraphs is passed  to EV-AL2.


     Cyclic subgraphs  generated by edge  removal present  a further difficulty.


It is not possible to  directly determine from the topological valences


whether the subgraph is connected or whether  it represents  two connected


subgraphs which are disconnected  from each other. A special routine, .ICHECK,


is designed to cope with this problem.   ICHECK  uses  a  "route-finding"


algorithm to ensure .that thefe~;is a  route'from'every'vertex-to every other.


     . Subroutine EVAL2  is similar  to  EVAL except that EVAL2  does not need to


determine cyclic order and does need to  use ICHECK to'  determine


connectivity.                                  .


     After all evaluations have been made  for the combination of atoms


generated by CONECT, control  is passed  back to  CONECT  and the next


combination is generated.  After  all exclusive  connected combinations of


atoms have been evaluated, CONECT passes control  back  to the main program for


output of results.

            I                    .          j
     An output routine, OUTPUT, writes  the. results to  the file INDXOUT.DAT.

               -            • '   • '      • '    '   •             '                i
This file is a formatted sequential  file which  contains: the molecule's name


and  empirical  formula, the data run, enumeration  of  each subgraph  type and


order, and valence corrected, simple, and  framework  connectivity indices for


orders 0 to 10.  OUTPUT also  sends  this  information  to the  line printer in a


tabular form which requires one page per molecule.


                                     46      .

-------
     Glossary of-Programs and Files.  All  important  programs,  command files,




and data files used in the INDEX system are  listed below with  brief




explanations.  Alphabetical order  is used  with  the file  extensions  having




priority.  In general, file types  can be identified  by  their  file




extensions:




     .BIG - an IIF file  contining  "big" molecules only




     .BKP - master file  backup




     .CMD - command file




     .DAT - general data file




     .FTN - Fortran code (used here  for subroutines)




     .IIF - INDEX input  file




     .MED - an IIF file  containing "medium"  molecules only




     .SML - an IIF file  containing "small" molecules only




     .TOC - at table of  contents file




     .TSK - an executable task image of a  program




Note that many programs  are listed which will never  be-run directly and some




files are listed which are transparent to  the user.  Almost all user




executions will  use only .CMD and  .TSK listings below.   The program INDEX5




and its subroutines are  not described in the glossary  (see "Operation of the




INDEX5 Program", below).




xxxxxx.BIG     -  An IIF-type file written by the program SPLITTER which




                  contains only molecules  with  23 atoms  or more.




xxxxx.BKP      -  A backup file  for  the master  file  which was  created on the




                  date "xxxxx" (e.g. OCT17.BKP).  This  file is in the




                  INDXODT.DAT format.




DFILE.CMD      -  Command for duplicate search  of the master'file.




DSORT.CMD      -  Command for a PDS  sort used  i-n DFILE.CMD.




                                     47

-------
PSTOC.CMD       -   Command to print table of contents of the' master file




                   ordered by location- or by molecular formula.  The-master  ..




                   file  can also be output to the line printer in tabular  form




                   with  this command.




TOC.CMD         -   Command for a PDS sort used in PSTOC.CMD.




SHMIND.DAT      -   An INDXOUT-type file containing selected entries from the




                   master file written by the program FILEM.




DISCOM.FTN      -   This  subroutine takes a given compound with its




                   connectivity indices from the program IND1ST, then searches




                   the master file computing the Euclidean distance for each




                   compound in a connectivity 34-space.  The most and least




                   distant compounds are then output to the user's terminal.




MFORM.FTN    , :!-   This  routine .outputskmole.cularv formulas to the line    •.-  •;•'•'




                   printer.          .     "'•  •             .




PUTOUT.FTN      -,  This  routine takes  an entry from the INDEX5 master file and




                   outputs it to the line printer in table form.




RITE47.FTN      -   This  routine writes a sequential file with parameters in




                   the title field of  each record retrieved from a user




                   selected sequential file in INDXOUT.DAT format.




RITEM.FTN       -   This  routine retrieves an entry from the master file and




                   writes it to a sequential file on logical unit 3,.




RRITEM.FTN      -   This  routine writes a sequential file with parameters in




                   the title field of  each record retrieved from the INDEX5




                   master file.                                          '




SEQDIST.FTN     -   See write up for DISCOM.FTN.  The difference is that here




                   the initial compound is retrieved from a sequential file  in




                   INDXOUT.DAT format  rather than from the INDEX5-master file.




                                    '  48

-------
xxxxxx.IIF     -  A  file  in  Che  INDEX  input  format.   Contains  coded molecular




                  structures.  (e.g. PHENOLS.DAT)




MASTERxxx.IIF  -  An IIF  file which  is  saved.   Normally a number of .IIF




                  files are  appended together  to  form the type of file (eg.




                  MASTERS.IIF).




xxxxxx.MED     -  See xxxxxx.BIG;  .MED  contains only molecules with 12-22




                  atoms per  molecule.




xxxxxx.SML     -  See xxxxxx.BIG:  .SML  contains only molecules with 11 or




                  fewer atoms per  molecule.




TOC.TOC        -  A  transparent  temporary file for input to a  PDS sort




                  (TOC.CMD)  used in  the table  of  contents programs




                  (PTOC.CMD).




STOC.TOC       -  A  sorted transparent  temporary  file output from TOC.CMD in




                  the table  of contents programs.




BURP.TSK       -  This program initializes an  available space  list for the




                  master  filing  system  of INDEX5.




CHECKERS.TSK   -  This program checks  the validity of molecular structure




                  entries  in an  IIF  file.  The program is designed to be run




                  on a Tektronics  4010  terminal.   Each molecule is drawn;




                  hard copies are  optionally available.  "Wheat" and "chaff"




                  filing  optionally  available..




CINAMES.TSK    -  This program prints  the names of molecules on an




                  INDXOUT.DAT-type file.




DFILEM.FTN     -  This program is  part  of the  duplicate entry  search of the




                  INDEX5 master  file.   It writes  the.entire master into a




                  temporary  sequential  file  to be  used in. a PDS SORT command




                  (DSORT.CMD).                               :




                                     49

-------
DSFILE.FTN
FILEM.TSK
GETEM.TSK
HIBURP.TSK
IIFCOUNT.TSK
IIFNAMES.TSK
INDIST.TSK
INTMIND.TSK
LOADEM.TSK
This program uses the files created by DFILEM  and  DSORT  to




Locate duplicate entries in the INDEX5 master  file.




This program writes a sequential  file in INDXOUT.DAT  format




from sequentially or randomly retrieved compounds  from the




master file.




This program retrieves entries in the INDEX5 master  file by




location and outputs them to the  line printer  in table




form.




This program initializes the hash table for the INDEX5




master file.




This program counts the number of molecules in an  IIF .file




and prints a frequency histogram  of the number of  atoms  per




molecule.                       ..  . .     '           .




This program will print a list of the names of molecular




structures in an IIF file.




This program takes a user selected compound from the  INDEX5




master file or a sequential file  in INDXOUT.DAT format and




compares it to every non-garbage,  non-duplicate entry in




the master file and computes the  most and  least distant




compounds with respect to the connectivity indices in




34-space.




This program initializes! the master file MINDEX.DAT  for  the




INDEX5 filing system.




This program uses the output file (INDXOUT.DAT) created  by




INDEX5.  After the data in the file has been verified, this




program  is run to insert the entries  into  the  master  file.




                   50
                                                                          ฉ

-------
MOSS.TSK       -  This  is  an  interactive Molecular  Structure Specification




                  program  which outputs a verified  IIF-type file.   The user




                  can specify  entire  structures  or  modify any existing




                  structure already on an IIF  file.   Designed to  be run on a




                  Teletronics  4025 terminal.




PREPEM.TSK     -  This  program creates a user  named  output  file with a




                  parameter name  and  value  for each  logical record in the




                  title  field  for the compound.   Input  can  be taken from the




                  master file  or  any  file in INDXOUT.DAT format.




PSTOC.TSK      -  This  program prints out a table of contents for  the INDEX5




                  master file  sorted  by location or  molecular formula.  It




                  can also output the entire master  file to the  line printer




                  in tabular -form.




PTOC.TSK       -  This  program creates a temporary  sequential file to be used




                  for input for TOG.CUD and PSTOC.TSK.




SPLITTER.TSK   -  This  program splits an IIF  file into  three files (.SML,




                  .MED,  and .BIG) based on  the number of atoms per molecule.




WIPEM.TSK      -  This  program deletes entries from the master file given the




                  entry  location.




     Despite the  immediate usefulness of some  connectivity  indices and the




insignificant cost of calculating them for  large numbers of chemicals, there




are two problems  which  must be solved before  a complete set can  be prepared.




The first is that the computer program operates  from a  connection  table which




must be generated or purchased.   Even if 10 structures/hour could  be encoded




into the system,  over 500  working days would be  needed  to input  the TSCA in-




ventory in the connectivity program.  An alternative would  be to  purchase the
                                      51

-------
 connection cables from existing  files; however  funds  to  accomplish  this  have




 not been available to this project.  The second  limitation,  is  that  adequate




 valence correction factors have  not been determined  for  all  heteroatoms.




 Consequently, valence-corrected  indices cannot yet be  calculated  for




 organophosphorus chemicals.  This work can proceed if  funding  is  available.




      One of the potential uses of the connectivity indices  in  addition  to




 single preditions is the development of a "similarity  index" for  chemicals  in




 the _n dimensional structure  space.  The purpose  of this  project  is  to provide




 the Agency with the capability to "indict" a  chemical  because  of  its




 similarity to hazardous chemicals for which data are  available.   The




 "similarity index" may evolve into a ranking  system  for  testing  priorities.




 For example, each chemical structure can be represented  by  a single point  in




•_n dimensional (often'more than 30 dimensions) structure  space.   Chemicals




 with similar properties form clusters in structure space.   The distance  of




 any new chemical structure from  a given cluster  can  be calculated to




'determine statistically whether  the new structure  is  "inside"  or  "outside"




 the cluster and, if outside, how it is distributed between  clusters




 representing contrasting environmental behaviors.  This  part of  the QSAR




 project is evolving as a "back-burner" project  until  funds  can be redirected




 from testing to develop it more  rapidly.




      For a more detailed account on possible 'uses  of  the indices, the reader




 is referred to Kier and Hall (1976).  As a summary,  it has  been  shown that




 the boiling point of classes of  chemicals can be estimated  from  the following




 equations:
                                       52

-------
aLkanes           b.p. = 55.691X. + 4.7084Xpc - 96.13        r =  0.9969
alcohols          b.p. = ISl.SSix-  155. 23^ -  17.993X
                         + 21.743Xvp - 2.743xc  -  29.05      r =  0.99

alkyl halides     b.p. = 36.36^ +  16.363Xp +
                         2.8693XVC - 31.037                 R=  0.9982

The water solubility of alcohols can be estimated  from

          In S = 9.417 - 11.2661X +  8.643XXV       r =  0.9945

Log P can be estimated from a similar series of equations  such  as  for

aliphatic carboxylic acids

          Log P = -0.859 + 1.6151XV  - 0.550^      r =  0.9979

Similar equations have been developed for many  biological  activity endpoints

such as narcosis, conversion of cytochrome P-450  to P-420  in rat  livers,

thymidine phosphorylase inhibition, microbial inhibition,  vapor  toxicities,

and potency of mutagens in the Ames  test.  •     '
                                     53

-------
VII. QSAR INFORMATION SYSTEM                         •       .            ..'.-.




          An information storage and retrieval system was  first  begun  as  part  of




     the structure-activity studies.  It was clear  that  any computer modeling




     study would require the storage, retrieval and manipulation of  large  amounts




     of physical, chemical, and biological data.  Efforts  to determine  the




     availability of a commercial or public  information  system capable  of  meeting




     the needs of the project met with  little success and  work was begun  on  the




     design of a local interactive  information storage and retrieval system  and a




     corresponding data base.




          The first data base contained only 248 compounds with  the  following  data




     elements: a local identification number, Chemical Abstracts registry  number,




     molecular weight, molecular formula,  freezing  point,  boiling point, vapor




     pressure, solubility  in water', 'Wiswesser line  notation (WLN), logarithm of  •




     the partition coefficient (log P) , gas  chromatographic retention  data,




     bioconcentration factors, acute aquatic toxicity data (LC50), and  acid




     dissocation constant  data.




          Initially the information storage  and retrieval  system was quite




     primitive as searches could only be carried out for the following: one  local




     identification number, one registry number, a  specific or ranged  molecular




     weight, a ranged, partial or complete molecular* formula, or a specific  or




     ranged log P value.   The commands  available sit  that' time were:  •   •




                           SEARCH              '  :   COMMAND




                           DISPLAY                  END




                           CLEAR




          In the next phase of system development an additional  2,000  compounds




     were added to the data base along  with  new data elements such as  maximum




     allowable toxic concentration  and  8th and 9th  Chemical Abstracts  Collective




                                           54

-------
Index, systematic, common, and trade names.  Many  improvements were made  in




searching capabilities, for example, an exact name search was added and all




properties could be "inventoried".  Numerous new commands were added:




                      INTERSECT        .            SHOW




                      MERGE                        HELP




                      LIST                         STATUS




                      OUTPUT                       COMMENT




                      INFORM                       STOP




During this development the acronym ISHOW  Conformation jjystem for jlazardous




ta-ganics in V[ater) was coined, a User's Manual prepared, and the system was




made available to  the public.




     Additional data were added to the data base and  considerable effort  was




expended in continued software modification to provide for  rapid'searching




compatible with efficient storage.  A major effort was the  development of a




software manual along with complete documentation  of  the computer software.




     ISHOW is currently undergoing another major change to  test methods of




evaluating QSAR on large numbers of industrial chemicals.   Using information




obtained from the Office of Toxic Substances, the  ISHOW data base is  being




expanded to contain an additional 12,000-13,000 compounds as a training set




of data.  Through  the support of EPA Region V an expanded ISHOW will  contain




compounds manufactured or used in the Great Lakes  watershed or EPA Region V.




Corresponding information on manufacturers or users of these compounds  in the




Great Lakes is being added, along with new data elements which include




biochemical oxygen demand, chronic toxicity (aquatic), and  the Kovats index




for gas chromatography data.  Software to  store and search  these parameters"




is also being developed.
                                     55

-------
     The purposes of  this  part  of  the  project  are  twofold.   First,  this




project is attempting  to make data concerning  the  biological and  chemical .




activities as well  as  key  molecular descriptors  available  to many researchers




with the assumption that this field of study will  progress  more rapidly.




Second, it is attempting to  develop and test predictive  techniques  as  they




are developed on large numbers  of  chemicals such as  those  found in  the TSCA




inventory.  Such evaluations are  judged to be  essential  if  QSAR is  to  become




an acceptable screening technique  in evaluating  the  environmental efficacy of




organic chemicals on  a major scale.
                                      56

-------
ACTIVITY REPORTS




A.   Compilation of the Aquatic Toxicity Data Base




     A review of the aquatic toxicity data  in the  literature  indicated  early




in the study that the amount of data as well as  the quality of  comparable




data precluded serous QSAR modeling attempts.  This conclusion  was  a major




factor in the decision to generate a systematic  set of  toxicity data with




aquatic organisms.  In the event that data  in the  literature  can be used in




the QSAR models after the patterns begin to emerge  from the testing program,




the QSAR project is compiling an aquatic toxicity  data  base.




     The data are being compiled from manual and computerized literature




searches using the compiling form listed in Figure  8.   The form was designed




with the cooperation of ASTM and OTS and is intended  to meet  the needs  of




several computerized data bases.  The data  will  be  loaded  into  a computer




file at ERL-Duluth from which subsets of the data  will  be  sent  on magnetic




tapes to the ISHOW and HEEDA developers according  to  their respective needs.




     To date, information on the toxicity of approxiately  700 chemicals has




been compiled.  The data include tests on approximately 75 freshwater and




marine organisms.  Acute tests make up approximately  675 of the tests and




chronic tests constitute about 75 of the tests.  Fathead minnows, rainbow




trout, guppies, bluegills, daphnids, scuds, and  stoneflies were most




frequently tested.  In addition to the toxicity  data, approximately 50  tests




have been reported for bioconcentration.  At present  ISHOW contains test




information on 350 chemicals and information on  approximately 150 chemicals




is ready to be loaded into ISHOW.  Tests have been  performed  on about 200




chemicals and/or mixtures for which Chemical Abstract Service registry




numbers have not yet been located.
                                     57

-------
                                          Figure 8.   Aquatic Toxicity Data  Compiling  Form
                       CHEMICAL NAME
                                        UimilAI AI1SIHACI  HEGISIIU NIIMUEH (CAS ซ)
Ol
oo
IE SI UIIMICAI
Chemical Gratlo

III II ItlHCI

ItirtlllNCE

1(1 II HI NCI

IIIIIKINU

Source









INI) till
Effect
Cmlpoln





—



cr
:OMC.W
-------
B.   Estimation of the Bioconcentration Factor




     The measurement and estimation of the bioconcentration  factor  (BCF)  in




fish have been reviewed by Veith et al. (1979, 1980).   It has  been  shown  that




the BCF can be estimated using the equations:




                         log BCF = 0.85 log P - 0.70




                    or   log BCF = 0.76 log P - 0.23




depending on the data base selected.  The accuracy  of  predictions made  with




these QSAR models is limited by the inherent variances  in the  BCF test  and




the variation of species tested in the data base.   The  equations were




intended for an estimation of the bioconcentration  potential within an  order




of magnitude for screening purposes.  Chemicals for which metabolic routes




not reflected in the log P are possible may give BCF values  lower than  the




bioconcentration potential predicted by log P.  For the majority of chemicals




tested the present predictive capability can estimate  the BCF  within a  factor




of three.




     Improvements in these predictive capabilities  are  being  sought along




several fronts.  The first is the continuing effort to  compile BCF  data into




a QSAR computer file which is interfaced with automatic plotting and




regression programs.  This will permit frequent updating of  the QSAR for  BCF




at the request of the EPA program offices.




     The second project is a re-analysis of BCF test data to  improve the




calculated BCF values reported for individual chemicals using  methods




recommended by OTS or ASTM.  This effort began with a  study  of the  BCF  data




for trichlorobenzene, DDE, and hexachlorobenzene as part of  an EPA




round-robin test validation program for OTS.  A complete analysis of this




data will be reported elsewhere and summarized in subsequent  progress  reports




of this project.




                                     59

-------
     The third project is aimed at uncovering the variation of the BCF among




several species tested. . In an at tempt, to relate the BCF-of specific  ... .  .




chemicals between fathead minnows, bluegills, rainbow trout, etc., a




simplified bioenergetics model is being tested to determine the extent to




which the differences  in observed BCF between species is dependent on




different metabolic activities of the species.  Exposure equipment has been




constructed to measure the BCF and the oxygen consumption of fish




simultaneously.  Since oxygen consumption may reflect the amount of water




pumped across the gills, it is hypothesized  that this value may "normalize"




the BCF value for different species.  This study will be discussed in more




detail when appropriate.




C.   Estimation of Water Solubility




     The solubility of many organic chemicals has. been estimted from the  log




P by numerous investigators.  Hansch et al.  (1968) presented relationships




for 9 classes of organic chemicals showing that linear inverse relationships




exist over approximately six orders of magnitude of water solubility as shown




in Figure 9.  If the data are separated into chemical classes, a series of




equations estimating solubility from log P are available with a correlation




coefficient greater than 0.96 for most classes and greater than 0.93 for  all




classes.




     Because of the importance of water solubility in defining the aquatic




toxicity testing regime as discussed previously and in the distribution of




chemicals in the environment, a computer data base of solubility data has




been established at ERL-D.  Although there are insufficient funds at this




time to generate water solubility data in  this project,  an effort is being




made to update the data base from literature reviews.  Subsequent reports
                                      60

-------
                              Figure 9.   Relationship between Log P and  Water Solubility
(ฃ2)
O
         -6
                               0
6
                                                  LOG  P

-------
from this project will provide evaluations of the state-of-the-art  in




estimating water solubility by QSAR.
                                     62

-------
IX.   EVALUATION OF RAPID SCREENING TESTS FOR QSAR




          la Che event that the QSAR analysis of aquatic toxicity data  leads  to




     the need to test more chemical classes than is  feasible using  the  four day




     acute test, this project began evaluating several more rapid tests  for the




     reliability in estimating relative toxicity of  chemicals.  The  tests




     initially selected were the residual oxygen bioassay, the Microtoxฎ bacterial




     bioluminescence bioassay, and the fluorescence  algal  stress bioassay.




     A.   Residual Oxygen Bioassay




          The residual oxygen bioassay is a 6-8 hour test  performed  in  sealed




     bottles.  Briefly, about 3 gm of fish are placed into each of  a series of BOD




     bottles containing water saturated with oxygen  and varying concentrations of




     the test chemical.  The fish die quickly at lethal concentrations  of  the




     chemical or after longer periods at sub-lethal  concentrations  due  to  oxygen




     depletion.  The oxygen is measured in each bottle as  the last  fish  dies  and




     the residual oxygen concentration is plotted against  the concentration of the




     chemical.




          At high concentrations, the fish are killed quickly by the chemical




     before the oxygen can be reduced.  In control exposures, the residual oxygen




     remaining for fathead minnows is approximately  0.5 mg/L.  The  concentration




     of chemical at which the residual oxygen rises  above  that of the control




     residual is the lethal threshold for the chemical.  This study  is  an




     evaluation of the relationship between the lethal threshold and the 96 hour




     LC50 for fathead minnows.




          Tests were conducted in 300-ml BOD bottles containing the  desired




     concentrations of test chemicals.  Stock solutions were prepared by either of




     two ways.  For some solutions the calculated amount of chemical (solid or
                                          63

-------
liquid) was weighed  out,  transferred  to  a 2-liter  volumetric  flask,' and




diluted to volume; aliquots  of  this stock solution were  then  transferred to




the BOD test bottles and  dilution  water  was  added  to obtain the desired test




concentration.  For  other solutions the  stock  solution was  prepared by taking




2 liters of solution from a  saturator, analyzing this solution, and then




preparing the desired concentrations  from aliquots of the stock solution as




before.  The dilution water  used  in all  tests  was  Lake Superior water warmed.




to room temperature  (20-22ฐC).




     Fathead minnows (Pimephales promelas)  from the Environmental  Research




Laboratory-Duluth  stock were used  as  the test  fish; these fish ranged from




30-day-old fry  to  adults. Fish were  held in1the same water at the same




temperature as  that  used  for test  dilution water.   During holding, the fish




.were fed either frozen brine-.shrimp (San Francisco Bay Brand, Inc.,;Newwark,




CA) or a commercial  salmon ration  (Glencoe Mills,  Inc.,  Glencoe, MN)  twice




daily.  Fish were  not fed for at  least 16 hr prior to testing nor  during




testing.




     Tests were conducted by preparing the test solutions as  described above,




usually using 9-11 test concentrations plus  a  control (dilution water only)




for each test.  Fish were weighed  to  obtain a  3 to 3.5-g loading for each BOD




bottle, were rapidly placed  in the test  solution,,and the bottles  were




immediately capped.   When all the  fish at a given  concentration died, the




dissolved oxygen concentration and pH of the test  solution  were measured, and




the number of test fish was  counted.




     Dissolved  oxygen concentration was  determined using a Beckman 0260




oxygen analyzer calibrated by a Winkler  titration  (APHA  et  al., 1976) on




water  from a holding tank; pH was  determined with  a Corning Model  12 meter.
                                      64

-------
Averaging of pH was  accomplished  using  a  computer  program which converts pH




values to their corresponding hydrogen  ion  concentrations, .averages  these,




then reconverts the  average  to  pH units.




     Data Treatment.  The dissolved oxygen  concentration  and  toxicant




concentration test data  for  each  test were  analyzed  by computer to determine




the lethal threshold concentration.  The  lethal  threshold endpoint was




determined by means  of the computer program HOCK(ey-stick) which fits the




data to a model for  the  bioassay  and used "hockey  stick"  statistics  to




estimate the endpoint confidence  intervals.   This  program was developed to




offer maximum input/output flexibility.




     Version I enables the user to  input  data directly into  the HOCK program




and receive output from  the  line  printer.   This  is  for single time usage.




Version II uses a data file  to  store information for multiple use.  Version




III gives examples of data modification and terminal output  and can  be used




with either a data file  or terminal data  input.




     Figure 10(a) illustrates Version I where test data are  entered  directly




through the terminal for a test with 4-pentylphenol.  The threshold  is




computed to be 2.61 mg/1 with the 95% confidence interval being 2.37 to 2.88




mg/1.  Finally, the  data are plotted at the terminal to show  graphically how




well the model fits  the  data (Figure 10(b)).




     The chemicals tested in the  evaluation o'f this  screening method are




presented in Table 4 and the correlation  between the LC50 and the  lethal




threshold is presented in Figure  11.  The correlation for this set of




chemicals produced an R^ = 0.92.  Although  the data  indicate  that  a




relationship exists between  these two endpoints  for  toxicity, it is  the




current opinion that the use of the residual  oxygen  bioassay  as a  screening




test is o.f marginal value.  The test requires all  of the  fish, culturing




                                     65

-------
         Figure  10.   Input of Residual  Oxygen Test Data .to Program HOCK
        IAS  PROGRAM DEVELOPMENT SYSTEM    :
                11:50:01          3-OCT-80

            ENVIRONMENTAL RESEARCH LAB SYSTEM
PDS> LOGIN EPARCR RCR
User EPARCR  UIC [350,40] TT01:   Job-id 333  11:50:13   3-OCT-80
NOTE:  NEU DEC-MOS MEMORY HAS BEEN INSTALLED (UED.  24-SEP-80).

LAST SYSTEM BACKUP:  TUES. 30-SEP-80    (4 1/2  TAPES UERE USED.)  GETTING BETTER!
PDS> RUN HOCK
11:50:26
  PROGRAM HOCK(EY-STICK)
RESIDUAL DO BIOASSAY  ANALYSIS
    WHAT IS THE INPUT FILE? IF TERMINAL TYPE
                       TI:
TI:
  ENTER PAIRUISE DATA ( CTOX3 & R-DO ). CTRL-Z TO QUIT
0.09,0.39                                   ..,-•-
0.96,0.24            .             " .           •
2.9,0.96
4.8,4.91
6.8,6.99                      '.'..'
8.7,8.48
10.6,7.84
14.5,8.22
24.,8.59 "
48.,8.48
97.,8.10
"I
ft
1
2
3
4
5
6
7
8
9
10
11
CTOX3
9.0000004E-02
0.9600000
2.900000
4.800000
6.800000
8.700000
10.60000
14.50000
24.00000
48.00000
97.00000
LOGtTOX]
-1.045758
-1.7728778E-02
0.4623981
0.6812413
0.8325090
0.9395193
1.025306
1.161368
1.380211
1.681241
1,986772
R-DO
0.3900000
0.2400000
0.9600000
4. j9 10000
6 ,)99'0000
3.480000
7.840000
3.220000
3.590000
8.480000
8.100000
   IS  THAT  OK?  (YES  OR  NO)
 YES
   NAME THE PLOT (120 CHARS.  MAX.)
 TEST  97   4-PENTYLPHENOL   6/9/80
   TYPE 5 TO GET THE OUTPUT HERE, TYPE  6  TO  SEND PRINT  IT
 6
                                         66

-------
                               Figure 11.   Residual Oxygen Test  Results  for 4-pentylphenol
                                                            <ป-PENTYl.PHENOli'   6/9/HO




















II
t:
s
i
0
u
A
1,

m. v------------- 	 	 	 -________•_________ — ____-.___-_______-_-__--__ — --_...--_._.-....--__._..__.---_------.--
I . :
1 in<1CTt'Y ThMHSHfiLD = i.fallOy . PPM. 95% LIMITS ARt 2.38945 TO ^.S)7V3fa
I . ' . . .
1 i)HSKiHVATinris=iu)
y.U-1 MllDKLs********* '.''''.'
1 IH/tHljAP ztfUiaBaMlo
i HKI-SCM.N: DATA=!
I 0 0 tJ
in TuPf SimLP Rt.<", ION .' ซ*ป*ปซซซ
-------
Table 4
Compound
Phenol
2 ,4-Dichlorophenol
2 ,4 ,6-Trichlorophenol
Pent ach lor ophenol
2 ,4 ,6-Tribromophenol
4-Nitrophenol
2 ,4-Dinitrophenol
2,6-Dimethylphenbl ;.':!•.''••
2 ,4-Ditnethylphenol
2 ,4-Dinitro-o-crosol
3-Methoxyphenol
Tetrachloroethylene
1 ,1 ,2,2-Tetrachloroethane
2-Chlorophenol
1 ,1 ,2-Trichloroethylene
1 ,2-Dichloropropane
1 ,2-Dichloroethene
Lethal
Threshold
(mg/L)
20
4.2
1.3
0.40
12 .
11
-3.2
;--;-;>-.36--:-;^; .;
.-/. ;/-32 • •.:•-
0.32
120
14
19
58
66
40 ;
. . 138 i .' •
96-Hour
LC50
(mg/L)
28.8
8.23
9.17
0.23
, ' 6.64
60.5
'-'••'.'. 16.7
:,-^.--:v;;..;,:22.6:V;
16.8
2.04
76.3
13.5
20.3
12.4
44.1
. . 139
118
       68

-------
Table 4.  (Continued)
Compound
Methanol
Ethanol
2-Propanol
3,4-Xylenol
1-Butanol
sec-Butyl alcohol
1-Octanol
1-Hexanol - • : • • .
2-Methyl-l-propanol
l-Amino-2-propanol
o-Phenylphenol
1-Naphthol
4-Butylphenol
2-(2-Ethoxyethoxy) ethanol
4-Pentylphenol
Cyclohexanol
4-Phenylazophenol .
4-Chloro-3-methylphenol
Lethal
Threshold
(mg/L)
17,600
12,700
5,490
30.8
2,090
2,650
13
128 .
2,020
147
5.8
8.9
3.1
17,700
2.6
755 '
3.4
8.2
96-Hour
LC50
(mg/L)
28,200
15,000
9,640
14
1,740
1,910
14.0
. 97.2'
1,460
327
6.0
4.24
5.13
26,363
1.09
705
1.09
7.58
                               69

-------
 LI
 
-------
and handling facilities of  Che 4 day  acuce  test  and,  since the test requires




an entire working day as a minimum, the  time  saved  to produce an indirect




estimate of the acute toxicity is  judged to be  insufficient to recommend its




use as a screening test.
                                      71

-------
 B.    Microtoxฎ Bacterial Bioluminesence Test

      This rapid screening technique is the bacterial assay developed by


 Beckman Instruments Inc., which measures the decrease in natural light


 emission from the luminescent bacteria Photobacterium phosphoreum in response


 to  a toxicant.  The decrease in light is expressed as a 5-min EC50 (the


 concentration effecting a 50% reduction in light output).
                                         •

      Bacterial ECSOs for 76 compounds were measured to evaluate the


 relationship between the EC50 and the 96-hr LC50 values for one species of


 fish, the fathead minnow (Pimephales promelas).


      To eliminate variability in the fathead minnow LC50 values due to


 changes in water quality, chemicals were selected for bacterial testing only


 if.-the fathead minnow LC50s.,were derived from tests conducted at the


-. Environmental Research Laboratpry-Duluth.  Chemical, selection was further  .


 restricted to.tests conducted in flow-through systems using Lake Superior


 water, and when exposure concentrations were measured.


      Preparation of Test Solutions.  All solutions for bacterial bioassay


 were prepared in distilled, deionized water.  If the compound was readily


.spl.uble in water, concentrations were not measured and bacterial EC50 values


 were calculated from nominal concentrations.  When testing less soluble

 compounds, a saturator consisting of a volumetric flask and a magnetic


 stir-rer was used.  Chemical concentrations fo'r less soluble compounds were


 measured by gas chromatography.


      Bacterial EC50 Determination.  All bacterial bioassays were conducted


 using the Microtox Toxicity Analyser" (Beckman Instruments, Inc., 6200 El


 Camino Real, Carlsbad, CA).  Testing was done following the method described
                                      72

-------
by Beckman (1979) who supplied the reagents and  lyophilized bacteria.

     Inhibition of luminescence (M) was calculated  for each concentration:

X= (initial light output x blank ratio/final  light  output) -  1

where the Blank ratio = final light output of  the blank/initial  light

                        output of the blank.

A graph of log^g  ^vs. logio concentrations was  plotted and the

concentration causing 50% inhibition of luminescence was determined  from  that

graph.

      For most compounds it was possible to prepare  a  saturated  stock

solution that was at least double the EC50 concentration.  When  this was  not

possible, the saturated solution was tested and  the  reduction  in light  due  to

the toxicant was  calculated as follows:

                         mean chart reading for  the  Cl Blank  '    ~1
                     mean chart reading for 100% Toxicant  solution


     Statistical  Methods.  The regression correlation  for  logjo  96-hour

fathead minnow LC50 and logjQ 5-minute bacterial EC50 was  calculated for

selected ethanes, alcohols, and ketones to determine the effects of  chemical

substitution on toxicity to fish and bacteria.

     In addition, log^Q EC50 vs. number of carbons  and log^Q  fish

LC50 vs.  number  of carbons was plotted for a  homologous series  of

unsubstituted ketones and alcohols, and simil'ar  plots  of EC50  and LC50  vs.

number of chlorines were drawn for a homologous  series of  ethanes.

Prediction intervals were calculated at the 95%  level.

     Evaluation of the Bacterial Bioassay.  The  correlations  between EC50 and

LC50 are summarized in Table 5.  The correlation comparing 96-hour  fathead

minnow LCSOs to 5-minute bacterial EC50s for unsubstituted alcohols

(R^=.96), unsubstituted ketones (R^=.81) and chlorinated ethanes

                                     73

-------
         were evident.  Toxicity of  chemicals  to  both  fish  and  bacteria




increased with addition of chlorine  groups  to  ethanes.   Addition  of  carbon




groups also increased  the toxicity of  alcohols and  ketones  to bacteria and




fish.  A comparison of the reproducibility  of  acute fish and bacteria assay




results (Table 6) shows no statistically  significant differences  in  the




standard deviation in  the percent difference between replicates  in the two




tests.




     The bacterial bioassay  is  a rapid and  relatively  inexpensive test.  The




entire test, including temperature equilibration  for all reagents takes about




30 minutes to complete.  The bacterial bioassay exhibits precision comparable




to the precision of the acute bioassay 'with fish.   Bulich (1979)  found a




coefficient of variance of 18.6  for  81 tests on sodium lauryl  sulfate using a




Microtoxฎ Toxicity Analyzer.   ;'•.•.'.••"....''.'.   • ;,. /;   .   i        :    :'''    -•'••/ .




     From data obtained in this  evaluation  (Table 7),  bacteria  EC50  data




could be used to screen certain  types  of  chemicals  for potential  fish




toxicity.  This could  be done using  the prediction  limits indicated  in the




graphs (Figure 12) and may be demonstrated  using  a  hypothetical  example in




which a tier testing scheme  included an acute  toxicity trigger  at an LC50 of




10 mg/L.  If it were decided that extensive fish  tests would not  be  required




for chemicals with a LC50 greater than 10 mg/1, the data obtained in the




evaluation could be partitioned  accordingly. 'The point where  the lower




prediction limit  intersects  LC50 = 10  defines  the EC50 (501 mg/1)




corresponding to that  fish toxicity  value.   The lower  prediction  limit is




chosen to ensure that  95 percent of  the compounds having an EC50  greater than




the value will have an LC50  greater  than  10 mg/1.   Based on these data for




organic chemicals, chemicals with an EC50 greater than 501  mg/1 would have an




LC50 greater than 10 mg/1 and additional  fish  tests would not be  necessary.




                                      74

-------
   Table 5.  Correlations for Classes of Chemicals  of




  LC50 (96-Hr Fathead Minnow) vs.  Log   5-Min EC50  (Bacteria)
Chemical Class
All Organic Chemicals                 68               0.71




Unsubstituted Alcohols                 9               0.96




Chlorinated Ethanes                    4               0.98




Unsubstituted Ketones                 10               0.81
                                75

-------
Table 6.  Summary of 5-minute EC50 and 96-Hr LC50 Results of Replicate




   Testing of Selected Chemicals in Bacteria and in Fathead Minnows
Chemical Compound
Acetone
2-Butanone
2-(2-Ethoxyethoxy)
Et Hanoi
Hexanol
2-Methyl-2 ,4-Pentanediol
2-Octanone
Phenol
5-Min Bacteria
EC50 (mg/1)
22,000
21,000
5,750
4,350
1,290
1,000
42.7
40:4
3,300
3,200
; 2,710 -...
20.5
15.0
39.8
40.7
Percent Difference in
Replicated ECSOs
5.0
28.0
25.3
3.12
7.8
30.98
2.23

Chemical Compound
1-Amino-Propanol
Butanol
Cyclohexanone
96^Hr Fathead
Minnow LC50
(mg/1)
327
287
16.2
13.4 '
732 ' ;
527
Percent Difference in
Replicated LCSOs
15.5
19.0
32.56
                                  76

-------
Table 7.  96-Hr Fathead Minnow LC50s  and  5-Min  Bacterial EC50s.
Chemical Name
1-Decanol
*l-0ctanol
l-Amino-2-Propanol
l-Amino-2-Propanol
2-Pehnoxyethanol
1-Hexanol
1-Hexanol
Cyclohexanol
2 , 3-Dibromopropanol
2-Methyl-l-Propanol
2,2,2-Trichloroethanol •
1-Butanpl
2-Methyl-2,4-Pentanediol
2-Methyl-2 ,4-Pentanediol
2-Methyl-2,4-Pentanediol
2-Chloroethanol
2-Propanol
Ethanol
Methanol
96-Hr
Fathead Minnow
LC50 (mg/1)
2.3
13.4
280
327
346
97.2
97.2
705
71
1,460
. . ;298 -
1,740
10,700
10,700
10,700
37
9,640
14,200
28,200
5-Min
Bacterial EC50
(mg/1)
1.16
6.3
27.2
27.2
32.7
40.2
40.4
115
320
1,670
1,800
2,300. .
. - -' • •
2,710
3,200
3,300
13,400
35,000
44,000
125,000
                              77

-------
Table 7.-  (Continued)
Chemical Name
*2-Decanone
*2-Decanone
*2-Octanone
6-Methyl-5-Hepten-2-one
Cyclohexanone
Cyclohexanone
*2-Octanone
4,4-Dimethyl-Amino-3-Methyl-2-Butanone
4-Methyl-2-Pentanone
2-Butanone Oxime
5-Methyl-2-Hexanone
2,4-Pentanedione
5-Methyl-2-Hexanone
2-Butanone
2-Butanone
Acetone
Acetone
*p-tert-Butylphenol
Pent ach lor ophenol
4-Ch loro-3-Methy Iphenol
Phenylazophenol
96-Hr
Fathead Minnow
LC50 (mg/1)
5.7
5.7
37
85.8
527
732
37
8.4
509
' - 844 ;•••';'
158
96
158
3,200
3,200
8,140
8,140
1 1
5.15
0.220
7.59
1.09
5-Min
Bacterial EC50
(mg/1)
6.1
9.7
15
17.5
18.5
18.5
20
42.1
80
950
980
1,050
1,448
4,350
5,750
21,000
22,000
0.21
0.08
0.58
0.96
                                      78

-------
Table 7.  (Continued)
Chemical Name
4-Chloro-3-Methylphenol
*0-Phenylphenol
2 ,4 ,6-Tribromophenol
2 ,4-Dichlorophenol
2,4-Dimethylphenol
*l-Napthol
4,6-Dinitro-o-Cresol
2 ,4,6-Trichlorophenol
*2-Allylphenol
*p-Nitrophenol
2,4-Dinitrophenol
o-Chlorophenol
*4-Amino-2-Nitrophenol
Phenol
Phenol
*HexachloroeChane
*Penc ach lor oe chane
*Tetrachloroe thane
*Trich lor oe thane
*Kel thane
*Permethrin
96-Hr
Fathead Minnow
LC50 (mg/1)
7.59
6.0
6.64
8.23
16.81
4.24
2.04
9.17
•••'•' 15.9
60.5
16.7
12.4
34.3
28.8
28.8
1.53
7.3
20.3
81.7
0.51
0.017
5-Min
Bacterial EC50
(mg/1)
1.86
2.05
2.7
3.63
4.4
5.66
6.6
7.20
10
13.0
15.8
22.1
35.9
39.9
40.4
0.14
0.75
8.6
105
0.45
0.56
                                     79

-------
 -Table 7.  (Continued)
          Chemical Name
    96-Hr
Fathead Minnow
  LC50 (mg/1)
  ..;.:'. 5-M in
Bacterial EC50
    (mg/1)
*Diazinon

 Dimethyl Formamide

*N-Butyl Ether

 .2-(2-Ethoxyethoxy)-Ethanol

 2-(2-Ethoxyethoxy)-Ethanol

 •Triethylene Glycol

*Butanal

*Ethanal
        6.65

   10,600

       32.5

   26,400

   26,400

   61,000

       16.2

       31.2
        9.8

   20,000

       63.0

    1,000

    1,290

   33,000

       16.4

      342
                                      80

-------
                    Figure 13.  Relationship Between 96-hr LC50 and Microtoxฎ EC50
I    7
6-fl
2C

a
8-
<
k,

K
X
 i
(O
CO

Q
2.5
      0.0
     -2.5
     -6.0
           -2
                      0
6
                        LOQ10 6-MINUTE BACTERIA EC50

-------
If this screening technique had been  applied  Co  the  compounds  in  this




evaluation, fish tests would not have been  performed on  35  of  the 76




compounds.  Because a bacterial bioassay costs considerably less  than  the




estimated $2,000 for a fish test,  the savings would  be substantial.




     Compounds which fall between  the upper and  lower prediction  limits  for




LC50 = 10 would require  additional  testing  with  fish.  Another  decision




scheme might use the screening test to circumvent  intermediate  testing on




compounds likely to be highly toxic and "flag" them  for  long term testing.




Those compounds could be determined from the  regression,  plot using the same




process described above  and substituting the  upper prediction  limit.   As




testing continues on several species  at the ERL-Duluth,  it  will be possible




to make similar comparisons on other  homologous  series of chemicals.   At




present it appears that  bacteria can  be used  to  determine the  relative




toxicity to fish of selected groups of organic chemicals.
                                     82

-------
C.   Algal Fluorescence Test




     The "Algal Fluorescence Test" being developed  is  a.  sensitive  and  rapid




test procedure based on the fluorescence of chlorophyll  &_ from algae.




Normally, chlorophyll _a in algae will re-emit  a  small  percentage of  absorbed




light as red fluorescence, the remainder being used  in photosynthesis.   If




the overall health of the algal cell deteriorates the  percentage of




fluorescence increases.  The increase in fluorescence  has  been determined  to




be inversely proportional to the photosynthetic  capacity of  the cell.   By




poisoning the cell with DCMU, a very potent herbicide, it  is  possible  to




totally block photosynthesis and achieve maximum fluorescence.  DCMU causes  a




larger percent fluorescence increase in healthy  cells, whereas in  cells




stressed by a toxicant this increase will be reduced.  Using  these principles




we are refining our techniques for assaying toxicity using algae.




     Cultivation.  The green algae Selenastrum capricornutum  and the blue




green algae Anabaena flos-aquae' were* obtained  from EPA-Corvallis,  Oregon and




were cultivated in AAP medium (U.S. EPA, 1978).  Algal cultures were main-




tained on magnetic stir plates at 25ฐC.  All cultures  and  test systems




received continous irradiance of 350 ft. c. provided by  40 W  "cool white"




fluorescence bulbs.  Exponentially growing cultures  were used in all tests.




     Test Procedures.  The test system consisted of  24-5 ml  test tubes  in  a




plexiglass rack.  Algae were either innoculat-ed  into lake  water containing




various concentrations of toxicant or the toxicant was added  directly  to an




algae culture in AAP medium.  Final test volume  was  4  ml.  Volatile  compounds




was tested in stoppered tubes.  The test system  was  kept under continuous




light until actual fluorometer readings were taken,  usually  2 hours.
                                     83

-------
     Fluorescence Measurements.  All  fluorescence measurements  were  done, on a
Baird-Atomic, Model SFR'1'00 Spectrofluorimeter at an excitation of 430 nm and
an emission of 680 nm.  The test system was kept under  continous illumination
until actual measurements were taken  so as not to alter  the  photosynthetic
state of the culture.  After the incubation period  the  culture  tube  was read
directly in the fluorometer and an  initial reading  was  recorded after  the
signal stabilized (F^) .  Similarly  a  second reading was  recorded after the
addition of enough DCMU (3-(3,4-dichlorophenyl)-l,1-dimethyl-urea),  a
herbicide which blocks the electron transport pathway  in photosynthesis,  to
obtain a 5 MM solution (Fp).  These readings were used  to calculate, a
percent fluorescence  increase:
               (Fp -  FM) x 100 = %  fluorescence  increase.
                   %     .'.:....    ...   ;  :.-,...      ..: .  ....  ......

These values were- compared statistically  to the  control to detnnine. effect
concentrations of a variety of organic compounds.
     Evaluation of the Algal Fluorescence Test.  Typical results with  several
organic toxicants are presented in  Table  8.  Although  data are  insufficient
to be statistically analyzed with data from other algal tests,  the results
indicate the relative sensitivity of  algae to selected  compounds.  A dose-
response curve is presented  in Figure 14. The method  appears  to be
sensitive, reliable,  and fast (2 hours) to a broad  range of  toxicants.   With
the chemicals tested  thus  far, effect concentrations  are generally in  the
range of the fathead  acute values being generated.  The initial results with
natural surface waters being  tested for the "Ambient Metals  Study -  EPA"
indicate that the method can  also detect  metal toxicity.
     The method will  continue to be refined with respect to  test techniques
for analyzing the data.  A systematic testing of those  chemicals being used
                                    .  84

-------
      Table 8.  Effect concentrations using-the algal  fluorescence  test
Compound
Ethanol
Methanol
Butanol
Hexanol
Hexanal
Hexanal
Acetone
DMF
Tetrachlorophenol
1 ,2 ,4-Trichlorobenzene
1 , 3-Dichlorobenzene
1 ,1 ,2,2-Tetrachloroethane
Algae
Selenastrum
Selenastrum
Selenastrum
Selenastrum
Selenastrum
Anabaena
Selenastrum
Selenastrum
Selenastrum
Selenastrum
Selenastrum
Selenastrum
Effect3
Concentration
50 ppm*b
>32,000 ppm
300 ppm**
>320 ppm
.' .80 , ppm*
160 ppm*
4,000 ppm*
6,250 ppm**b
15.4 ppm**
• ' . 8 ppm*
60 ppm**
51.1 ppm**
Test
Duration
2 hr
2 hr
2 hr
2 hr
2 hr
2 hr
2 hr
2 hr .
2 hr
24 hr
3 hr
1.5 hr
a Lowest concentration tested showed significant difference  from control




b Lowest concentration tested




* P = 95



**p = 99
                                     85

-------
                                  Figure 14.  Dose-Response of Algal Fluorescen
                                                                     ce  to Trichlorobenzene
       J26  t-
I

UJ
CO
I
o
co

I
cf
       100
76  |-
        60  H
26   1-
                                            1,2,4 TRICHLOROBENZENE
CL

-------
in che QSAR program will be performed.  Plans  have  been made Co use a green




algae and a blue green algae Co determine  comparacive  sensicivicies Co




Cesc compounds.




     The literature and data generaCed  chus  far  suggest that Che reduce ion or




enhancement of chlorophyll a_ fluorescence  before  Che addition of DCMU may not




only indicate toxicity, buC also  Che  sice  of action of Che  parCicular




inhibicor.  This variable fluorescence  response  has Che potential Co




delineaCe between various modes of  action  of structurally different




compounds.  For example, DCMU Cype  compounds block  non-cyclic phoCosynChecic




eleccron transport and increase FJJ, whereas benzoquinone, a well known Hill




oxidanc, lowers FJJ.  Along wich our continued  effort Co improve Che method




we will also be considering Chese "behavioral" responses of chlorophyll _a




fluorescence'in an attempt to describe  mode specific'chemical accivicy'. '  A




literature review on Che poCency  and  modes of  action of chemicals in




inhibiting Che Hill reacCion in photosynthesis will be Che  subjecC of




subsequent reporCs.
                                     87

-------
X.   LITERATURE REVIEW ON NARCOSIS                                         :




          In an effort to summarize  the major modes of  action  of  common  industrial




     chemicals, a series of  literature reviews are planned.  These  reviews are  an




     integral part of QSAR because they help to determine  truely  anomalous




     behavior within a model  from changes  in mode of action which  suggest an




     additional model is needed.  The  first review is concerned with  narcosis  for




     the reasons stated previously in  this report.




          Narcosis is a reversible state of arrested activity  of  various




     protoplasmic structures  under the influence of certain chemicals.  Kaufman




     (1977) defines "general  anesthesia as a pharmacologically induced reversible




     disruption resulting- in  a  coordinated sequence of  changes in  neural




     activity".  The terms narcosis  and anesthesia are  used interchangeably  in




     medical literature.  Narcosis of'the  central, nervous  system  is. referred  to as




     general anesthesia.  The chemicals or drugs producing arrested activity  of




     any cellular organization  are referred to as narcotics, biological




     depressants, hypnotics,  and general anesthetics (Albert 1965).   A variety  of




     chemical- agents of different  chemical structures  and properties produce




     narcosis in animals.  These agents may'be gases OHesser et al.,  1978)




     aliphatic or aromatic hydrocarbons, chlorinated hydrocarbons,  alcohols,




     ethers, ketones, aldehydes, weak  acids, weak bases,  and some  aliphatic




     nitrocompounds (Albert,  1965; Gero, 1965; Roth, 1980).




          Most of the biological cells are susceptible  to  the  actions of  general




     anesthetics (Roth, 1980).  However, it appears that  all anesthetics  or




     narcotics produce a general depression of the central nervous system (CNS) of




     the organisms.  Vandam  (1965) pointed out that whatever may  be the mechanism




     of narcosis, the clinical  signs of general anesthesia are related to the
                                           88

-------
amount of anesthetic made available to the CNS.  Anesthetics may affect  not




only the CNS but a variety of cell types and several synthetic model




preparations (Halsey et al., 1974; Miller and Millet, 1975; Roth, 1979;




Seeman, 1972).  In a recent study, Schwab and  Schwab (1979) have shown that




protozoan (Allogromia laticollaris Arnold) cells after  treatment with 1  mM of




the volatile anesthetic, halothane, showed disappearance of cytoplasmic




microtubles.  These workers also demonstrated  the appearance of tubulin




polymers which were helices of paracrystalline structures.




     During the last few years new information has accumulated concerning the




pharmacological effects, of narcotic agents.  With the development of novel




monitoring devices and in vitro techniques, it has been demonstrated that




anesthetics can exert profound and direct effects on cardiac muscles and




alter regional blood flows in a number of organs (Price and Ohnishi, 1980;




Longnecker and Harris, 1980).  Many narcotic agents are capable of altering




tissue and plasma concentrations of cyclic nucleotides  such as cAMP or GMP




and of neurohumoral substances such as catecholamines,  dopamine, serotonin,




and renin (Altura, 1980; Rahwan, 1975; Zaleska, 1979).  In addition these




anesthetics have the ability to perturb lipid-bilayer membranes and alter ion




transport (Roth, 1980; Lee, 1976; Schettini and Furniss, 1979).




     A mechanism of action of some reactive anesthetics has been sought  in an




inhibition of enzymic processes.  For example, a number of anesthetics




inhibit mitochondria! oxygen consumption.  The mode of  action-in mitochondria




appears to be chiefly inhibition of electron transport  with slight uncoupling




of.oxidative phosphorylation (Altura, 1980).




     In spite of intensive pharmacodynamics, physicochemical, biochemical and




clinical research,- the .exact .changes in. the CNS -remain  unknown.  It is quite




puzzling how some chemically inert agents and  others which are rather ..




                                     89

-------
reactive chemicals may  induce  reversible  changes  in the  functions  of the  CNS




in such a way  that loss  of consciousness  and  insensitivity  to  pain occurs




(Kuschinsky  and Lullmann, 1973; Adriani and Naraghi,  1977).




     There are several  hypotheses which have  been proposed  to  explain the




mode of action of narcosis.  It  is  generally  accepted that  the majority of




general anesthetics  affect most biological cells  in a non-specific fashion.




However, due to their diverse  chemical structures,  a similar  or unitary




mechanism of action  appears  to be inconceivable.   Recent  studies suggest  that




all anesthetics may  not  act  via  the same  mechanism (Stockard  and Bickford,




1975; Trudell, 1977).




     One of  the oldest  and most widely quoted theories of anesthetic action




is the Meyer-Overton "lipoid hypothesis of narcosis". Meyer  (1899) and




Overton (1901) independently proposed that anesthetic.action  correlates with




a.high oil/water partition coefficient.   K. H. Meyer (1937) stated that




"narcosis (anesthesia)  commences when any chemical substance  has attained a




certain molar  concentration  in the  lipids of  the  cells."   It  is true that




most anesthetics, especially volatile anesthetics,  have  high  oil/water




.partition coefficients,  but  this does little  to  provide  an  explanation of how




they function  as depressants (Roth,  1980; Gero,  1965; Koblin  and Eger, 1979).




This theory  may only explain why such anesthetics accumulate  in the lipid




rich CNS (Roth and Seaman, 1971).   In addition,  there are some outstanding




•exceptions to  the Meyer-Overton hypothesis' of anesthesia,  such as  alcohols




which are nonspecifically acting anesthetics  despite their  low oil/water




distribution coefficient (Albert 1965).   Furthermore, oils  and certain other




substances with very high partition coefficients, do not  produce narcosis.




     Traube  (1904) proposed  a  different physicochemical  hypothesis of




narcosis.  This theory  describes a  lowering of surface tension between water




                                     90

-------
and a second phase such as oil or  air.  The  alcohols  in  homologous  series  fit




this generalization, but even familiar hydrocarbons  like  chloroform and




ethylchloride do not lower interfacial tension  at  the  surface  boundary of




water/air or water/oil.  Furthermore, many detergents  which  are  known  to




lower surface tension are not depressants.




     Ferguson (1939, 1951) provided a physicochemical  explanation based  on




the thertnodynamic or chemical potential of the  depressants  in  the  internal




phase and the biophase (site of biological action).  He  pointed  out that a




stable level of narcosis is achieved when an equilibrium of  drugs  is reached




between the two phases.  Ferguson's principle provides a satisfactory




correlation between chemical potential and the  activity  of  nonspecifically




acting drugs, but does not explain the true  mode of  action  of  depressants.




Depression of physiological function.generally  appears to be-the typical




action of compounds which do not enter into  specific reactions with the




organisms (Albert, 1965).




     Pauling (1961) postulated that hydrophobia depressants  cause  the




formation of hydrate microcrystals in the brain.   Such hydrates  are known  to




be formed at low temperatures around nonpolar molecules  and  nonpolar groups




within molecules, and are referred to as "clathrates".  These  clathrates are




ordinarily unstable at body temperature, but it is also  known  that  clathrates




formed around several kinds of molecules or  groups (mixed clathrate hydrates)




are more stable (Gero, 1965).  Pauling further  assumed that  clathrate




hydrates are formed in brain cells in the presence of  volatile anesthetics




which then combine with the alkyl  groups of  the proteins, and phospholipids.




These water clathrates are stable  under physiological  conditions and1they




cause water molecules in their vicinity to change  to the •denser  structure  of




ice.  The ice microcrystals formed in this way  tie up  water, increasing  the




                                     91

-------
impedance of Che neural network of conductors  and  hindering- Che  movement  of.




ions.in the brain.  Unlike other theories, Pauling's  hypothesis  centers




around Che aqueous phase and not on  the  lipid  material  of  the  CNS,  and  also




seeks  to connect depressant action with  a process  of  reversible




disorganizacion.




    More recenC Cheories of narcosis  seek the mechanism of depressanc  action




in interactions of anesthetics with  cell membranes (McElroy,  1947;  Miller-and




Pang,  1976; Rosenberg et al., 1975;  Ashcroft et  al.,  1977ab;  Haydon et  al.,




1977;  Jain et al., 1978).  It is well  accepted that conduccion of  impulses  in




Che neurons and muscle  fibers depends  on their cable  structure and  occurence




of specific changes in  Che ion permeability of the surface  membranes (Keynes,




1972).  Biological membranes are composed of a phospholipid bilayer and




cholesterol, arranged wich polar (hydrophill'ic)  headgroups  facing  outward and




Che lipophillic or hydrophobic moeity  inward-.   The protein  of Che  membrane  on




eicher side of Che phospholipid bilayer  forms  a  trilaminar  sandwich (Keynes,




1972) .  Membrane structures have pores or channels through  which sodium and




potassium can move across the membrane during  the  transmission of  nerve




impulses (Ashcroft et al., 1977).  Most  investigators have  envisioned




membrane lipids as the  targec of aneschecic actions.   However, some believe




chac general anesthetics inaccivate  proteins,  which are essential  for CNS




function (Eyring et al., 1973; Kaufman,  1977)'.




Recent Theories of Anesthesia             Si




     (a)   Phase Transition - Three  recent theories propose that anesthetics




disrupt the lipid bilayer of the membrane.  Lee  (1976)  postulated  an "annular




transition model", also called a "phase  transition hypothesis".   This




hypothesis proposed that the sodium  channel of an  excitable membrane is




surrounded by  an annulus of  lipid which  is  in  the  "crystalline" or gel state




                                     92

-------
(solid phase), which helps Co maintain an  open  sodium channel.   Addition of




an anesthetic triggers a change in the lipid  to  a  fluid  or  liquid  crystalline




state and this alteration allows  the  closure  of  the  sodium  channel,  resulting




in anesthesia.  An increase in temperature can  also  alter the gel  phase  to  a




liquid phase, and the temperature at  which this  transition  occurs  is called




the phase-transition temperature  (Roth,  1979).   The  "phase  transition




hypothesis" is supported by the studies  of Hill  (1978) on a phospholipid




membrane model, and of Karaaya et  al.  (1979) who  showed that high pressure




antagonizes both anesthesia and the anesthetic-induced decrease  in phase




transition temperature of dipalmitoyl phosphatidylcholine bilayers.   However,




a recent study by Pringle and Miller  (1978) which  utilized  structural isomers




of tetradecanol, indicated no direct  relationship  between lipid  phase




transition and the mechanism of anesthesia. .  Gases,-barbiturates,  steroids




and alcohol anesthetics have all  been shown to  fluidize




phosphatidylcholine-cholestrol lipid  bilayers and  some biological  membranes.




Miller and Pang (1976) using different lipid  composition and anesthetics




demonstrated that the fluidizing  efficacy  of  an  anesthetic  varied  with




different anesthetics and also with the  lipid composition of the membrane.




     (b)   Lateral Phase Separation - This hypothesis by Trudell (1977)




proposes that lipids under normal conditions  exist as gel and liquid phases.




The region where a gel phase coexists with a.liquid  phase is defined as  a




lateral phase separation (Trudell, 1977).  Conversion of one phase to another




phase allows the membranes to expand  or  contract with less  energy  than would




be necessary if the membrane was  only in gel  or  fluid form. Also  the regions




(fluid or gel) may be associated  with functional proteins.   Trudell-proposed




that anesthetics may reduce the gel phase  in'membranes and"thus  decrease the




lateral phase separation.  This decrease in the  lateral  phase separation




                                      93

-------
would prevent the functional protein from 'altering  its conformation  necessary




for trans location'of ions across the post synaptic  membrane or  for the




release of a neurotransmiter, thus resulting  in depression or narcosis




(Trudell, 1977; Trudell et al., 1975).




     (c)   Fluidized Lipid Hypothesis - This  hypothesis  postulates that




anesthetics can disorder the motion of the  lipid membrane components (Gage




and Hamill, 1976), and that due to increases  in the lateral and  rotational




motion of the lipid component, the stability  of the channel proteins  is




reduced.  This facilitates the closure of the  ionic channels  in  the  membrane




(Gage and Hamill, 1976; Miller and Pang, 1976; Miller, 1977; Roth, 1980).




     Theories discussed above as "phase transition" (Lee 1976),  "lateral




phase separations" (Trudell 1977) and the "fluidized lipid" hypothesis all




propose that. the.  phenomenon of narcosis is  dependent on  fluidization of  the




lipid membrane.   The spin label study of Pringle and Miller (1978) showed




that unsaturated  long chain alcohols and cis  or trans-tetradecanol could




fluidize egg lecithin cholesterol membranes to similar levels.   However,




contrary to that, cis-tetradecanol lowered  the transition temperature of




dipalmitoylphosphatidylcholine while trans-tetradecanol  elevated this




transition temperature.  These results are  consistent with a  lipid fluidity  ...




model of anesthetic action, but provide little support for a  "lateral phase




separation" theory of anesthesia.  The author's suggest direct interaction  of




alcohols with the membrane protein (Pringle; and Miller,  1978).   Studies




conducted by Jain et al. (1975) on the transition  profile of  dipalmetoyl




lecithin liposomes exposed to a variety of  drugs support the  theory  of




fluidity disorder of the lipid bilayer.




     (d)   Critical Volume Hypothesis - Mullins (.1954) proposed  a critical




volume hypothesis.  According to"this hypothesis,  narcosis is produced by




                                     .94

-------
expansion of the critical hydrophobic site due to the adsorption of




anesthetic molecules.  Mull ins (1954) suggested that anesthetic molecules




should have a complementary shape or size to "fit" into the  free spaces of




the membrane's chemical lattice.  This critical volume of occupation will




thus result in decreased sodium permeability and depression  of membrane




excitability.                                 ...••-•




     Several studies lend support to Mullin's critical volume expansion




hypothesis.  For example, application of hydrostatic pressure (pressure




reversal of anesthesia) removed, at least partially, the effects of general




anesthetics in certain amphibians and mice (Lever et al., 1971; Miller et




al., 1973).  The phenomenon of pressure reversal of  anesthesia was  first




observed in luminous bacteria (Johnson et al., 1942), which  led Lever et al.




(1971) to formulate a unifying concept of critical volume expansion as a




general mechanism of narcosis.  They postulated that anesthetics expand the




size of a critical hydrophobic site, while the hydrostatic pressure contracts




this site and causes reversal of anesthesia.  Lever  et al. (1971) calculated




that general anesthetics at critical concentrations  in the lipids would




produce an expansion of lipids of the order of 0.4%.  Further studies on




quantitative measurements of erythrocyte membranes (Roth and Seeman, 1971,




1972ab), and of lipoprotein films (Clements and Wilson, 1962) support the




estimate of 0.5% expansion of biomembranes in the presence of general




anesthetics.  High pressure of approximately 100 atmospheres indeed




antagonizes the effect of general anesthesia, but the degree of the high




pressure reversal varies" with different anesthetics, and the antagonism of




high pressure may not be proportional or linear with the' pressure (Halsey et




al., 1978).                              ....     	               . .
                                     95

-------
     A multisite expansion hypothesis was  proposed  by Ha-Lsey  et  al.  (1978.).




They stated that a general' anesthetic'inay  influence the  expansion  of more




than one molecular site with different  physical  properties, and  the  physical




properties of molecular sites may be  influenced  by  anesthetics and pressure.




In addition, the molecular sites have a finite size and  limited  occupancy,




and the pressure does not necessarily influence  the same site as anesthetics.




Finally the molecular sites  for narcosis may  not be disturbed by decreases  in




temperature in a manner analogous to  increases in pressure.   The multisite




expansion hypothesis based .on intact  animal studies is more relevant,




particularly with complete membranes  in which  the molecular sites  are




energetically linked together (Halsey et al.,  1978).




     (e)   Degenerate Protein Perturbation Hypothesis -  Previous studies




discussed in this review  all support  a  common structural.'perturbation, in the'




excitable membrane which  is caused by anesthetics of diverse  chemical




structures.  This unitary hypothesis  of narcosis was derived  from  experi-




mental evidence that anesthetic potency correlates  approximately with




oil/water partition coefficients of narcotic  agents (Seeiaan,  1972; Miller  et




al., 1977).  However, direct experimental  evidence  for any unitary mechanism




has not been presented  to describe the  interactions of anesthetics with




target protein structures (Richards et  al., 1978).   Studies conducted by"




Richards et al. (1978)  utilizing n-alkanols do not  provide support to a




unitary mechanism of narcosis.  These worker's  have  suggested  that  anesthesia




is produced by direct interactions between the anesthetic molecules  and




target membrane protein.  They assume that smaller  anesthetic molecules




(halothane or cyclopropene) distribute  in  one set of hydrophobic sites of




appropriate dimensions  within the protein, while larger  molecules  like




barbituates, steroids,  and aromatic amines may bind to distinct  sets of sites




                                      96

-------
on the target protein (Richards et al., 1978).  Eyring  et  al.  (1973)  had  also




proposed that anesthetics may interact with  the hydrophobic  region  of a




protein and change its conformation to make  it  less  active.  The  protein




pertubation theory of anesthesia does not provide a  distinct mechanism of




narcosis (Koblin and Eger, 1979), rather  indirect evidence  is  derived from




thermodynamic principles (Eyring et al. ,  1973)  and the  lack  of correlation




between lipid membrane fluidization with  n-alkanols.




     (f)   Cyclic AMP Metabolism - Cohn and  Cohn (1980)  proposed  that anes-




thetics interact with the activity of a specific enzyme  protein hydrophobic




site which binds with the adenine moeity  of  cAMP.  According to these authors




cAMP in vivo regulates the duration of narcosis without  altering  the  cAMP




concentrations in the brain.  These workers  have shown  that  pentobarbital




causes dose dependent inhibition of brain adenylate  deaminase  which converts




AMP to inosine (Cohn and Cohn, 1980).  It has been shown that  dibutyryl




cyclic AMP shortens the duration of narcosis induced by  a  wide variety of




narcotics and hypnotics, and protects against the lethal doses of amobarbital




(Cohn et al., 1980).  The effects of anesthetics on  transmembrane regulatory




systems may require further studies.  The role  of hormone  and




neurotransmitter-dependent adenylate cyclase is known in transmembrane




regulatory systems and production of cAMP (Rimon et  al., 1978).




     Our knowledge of the mode of action  of  anesthetic  chemicals  awaits




further understanding of the structure of phospholipid bilayers,  axonal




transport, conformation of nerve proteins, synaptic  transmission  in the




central nervous system..and the interactions  of  anesthetics with endogenous




components of the neurons.  Current theories of narcosis emphasize  lipid




membrane disorders in. the CNS without- specif ic.ity.-(Roth,.. 1980).
                                     97

-------
      In mammalian systems,  four, definite stages  of-anesthesia can  be       •




 distinguished after the administration of anesthetics  (Vandam,  1965).   Stage




 I anesthesia is characterized by motor excitation  and  inability to coordinate




 motor activity (ataxia).  Stage II anesthesia  is characterized  by  delirium




 when the animals show hallucinatory  and cataleptic  behaviour.   Stage  III




 anesthesia, also called surgical anesthesia, is  further  progression of  stage




 II with increasing immobility and loss of reflexes.  During  this stage  the




 subject becomes unresponsive to painful stimuli.   Further  increases in




 anesthetic dosage may cause a depression of  the  total  central nervous  system




 activity (Stage IV), and  irreversible damage to  the  brian  which results in




 medullary paralysis and death.  These different  stages of  anesthesia




 classified by Guedel (1951) as unidirectional  schema were  revised  on  the




 basis of neufophysiological studies  (Winters,  1976).




      Fishes treated with  anesthetic  agents have  been shown to undergo  similar




 stages of anesthesia (McFarland and  Klontz,  1969;  Klontz and Smith,, 1968;




 Bonath, 1977).  The phenomenon of narcosis in  fishes has not been  extensively




 studied.  Hunn and Allen, (1974) reviewed the  recent studies on MS222  and




 quinaldine, used widely as  fish anesthetics.   Such  compounds are readily




 absorbed through the gills  of fish,  as demonstrated  with rainbow trout  (Salmo




 gairdneri), channel catfish (ictalurus punctatus),  marine  dog fish (Squalus




..acanthias), brook trout (Salvelinus  fontinalis), bluegills (Lepomis




 macrochirus) and largemouth bass (Micropterus  salmoides).  High lipid




 solubility and rapid rates  of diffusion across  the  gill  membrane appear to be




 important properties for  fish anesthetic agents  (Hunn  and  Allen, 1974). For




 example, MS222 has been shown to be  readily  taken  up through the gills  of




 freshwater fishes according to its concentration gradient.  From the  gills it




 is distributed to many body tissues, especially  the  cerebrospinal  fluid (CSF)




                                      •98

-------
in che brain (Hunn, 1970; Hargens et al., 1974;  Stenger  and Maren,  1974;




Daxboeck and Holeton, 1980).




     Fish have been shown to possess superficial  and  internal  (or  central)




oxygen receptors which initiate active cardiac and ventilatory responses




under environmental hypoxia.  Anesthetics may  interact with these  receptors




(Daxboeck and Holeton, 1978; Smith and Jones,  1978; Bamford 1974a,b; Marvin




and Burton, 1973) .  Mechanisms by which  such chemical agents disrupt various




physiological functions  in  fishes largely remains unknown  and  deserves




further studies.




     Studies conducted in our laboratory with  the fathead  minnow (Pimephales




promelas) to determine 24,  48, and 96 hour median lethal concentrations




(LC50) of a group of alcohols, aldehydes, ketones and ethers  indicate  that




these chemicals^cause fish  immobility, hyperexcitation .and loss  of  equili-.




brium before the onset of toxic symptoms.  These  symptoms  are  apparently




dependent on the chemical dosage and the exposure time.  It is probable that




these chemicals produce  lethal effects due to  medullary  paralysis,  as  shown




in mammals after dosing  with higher concentrations of the  narcotics.  The




typical signs of the stage  I anesthesia  (indicated by motor excitation and




ataxia) in fish may be determined _in vivo without killing  the  test  fish and




it may provide useful insight.




     Previous studies on long chain alcohols have shown  that  they  induce




narcosis and anesthesia  in  different animal species.  Jain and Wray (1978)




have shown that the lipid bilayer/water  partition coefficient  for  n-alkanols




are at least 3 fold smaller than their partition  coefficient  in  the bulk




solvents.  They imply that  alcohols may  induce changes in  the  lipid




biomembranes at much lower  concentrations than can.be predicted  from their




octanol/H20 partition coefficients (Jain and Wray,' 1978).   Similar  studies




                                -     99.

-------
on the partition coefficients  of  a variety of  chemicals with actual     :...




biological tissues (eig'. with-.-fish- g-ill/^O or-'blood/^O)  may-provide'-




useful informtion on  the movement  and  uptake of  these  chemicals,  and may have




some degree of correlation with their  toxicity.   McCreery  and Hunt, (1978)




studied 62 compounds,  including a  broad  range  of alcohols  for their abilities




to intoxicate rats in  vivo.  These workers suggested that  the shape of a




molecule had little effect on,its  ability  to intoxicate, but the  compound's




amphiphillicity was important  for  its  ability  to partition into neuronal




membranes (McCreery and Hunt,  1978).   Several  studies  have indicated that




membrane/water partition coefficients  are. more meaningful  than partition




coefficients with artificial solvents  (Roth and-Seeman, 1972a,b;  Jain and




Wray, 1978; Mullins,  1954; Seaman, 1972).   Biologically active compounds must




reach a definite target site in'order-.to-elicit-a particular response, and




such compounds cross multiple  hydrophyllic and 'hy.dr.ophobic barriers before




reaching the site of  action (Kubinyi, .1978).
                                      100

-------
                                 REFERENCES




Adrian!, J. and M. Naraghi.  1977.  The pharmacoLogic principles of regional




     pain relief.  Ann. Rev. Pharmacol. 17: 223-242.




Albano, C., W. Dunn III, U. Edlund, E. Johansson,  B. Norden, M. Sojostrom and




     S. Worl.  1978.  Four levels of pattern recognition.  Anal. Chim. Acta




     103: 429-443.




Albert, A.  1965.  Selective Toxicity. 3rd edition, John Wiley & Sons, Inc.,




     New York, pp. 3-380.




Altura, B. M.  1980.  Cardiovascular actions of general anesthetics.  Fed.




     Proceed.  39: 1574.




Ashcroft, R. G., H. G. L. Coster, and J. R. Smith.  1977a.  Local anesthetic




     benzyl alcohol • increases membrane thicknes's-.   Nature 269: 819-820;




Ashcroft, R. G., H. G. L. Coster, and J. R. Smith.  1977b.  .The.molecular




     organization of bimolecular lipid membranes.   Biochem. Biophys. Acta




     469: 13-22.                                                        .




Bamford, 0. S.  1974a.  Oxygen reception in the rainbow trout (Salmo




     gairdneri.  Comp. Biochem. Physiol. 48A: 69-76.




Bamford, 0. S.  1974b.  Respiratory neurons in rainbow trout (Salmo




     gairdneri).  Comp. Biochem. Physiol. 48A: 77-83.




Biagi, G. L., A. M. Barbaro, M. C. Guerra, M. Babbini, M. Gaiardi, and M.




     Bartoletti.  1980.  ^ values and structure-activity of




     benzodiazepines.  J. Med. Chem. 23: 193-201.




Bonath, K.  1977.  Information on' experimental animals _+4 Narcosis of




     repitiles, amphibians and fish, Part 3.  pp.  99-149. Published by




     Verlag-Paul. Parey, Berlin, Germany.      _. 	   '    ....-•  ••••••  .
                                     101

-------
Bulich, A. A..M. W. Greene, and D. L. Isenberg.  1979.  The reliablity of one




     bacterial luminescence- assay-for the determination of toxic ity of pure




     compounds and complex effluents.  ASTM, Proceedings of Chicago Meeting,




     1979, (unpublished).




Cammarata, A.  1972.  Interrelationship of the regression models used for




     structure-activity analyses.  J. Med. Chem. 15: 573-577.




Cammarata, A. and G. K. Menn.  1976.  Pattern recognition classification of




     therapeutic agents according to pharmacophores.  J. Med. Chem. 19:




     739-747.




Chu, K. C., R. J. Feldmann, M. B. Shapiro, G. F. Hazard, Jr., and R. I.




     Geran.  1975.  Pattern recognition and structure-activity relationship




     studies.  Computer-assisted prediction of antitumor activity in




     strkucturally diverse drugs in an experimental-mouse brain tumor system.•:




     J. Med. Chem. 18(6): 539-545.




Clements, J. A. and K. M. Wilson.  1962.  The affinity of narcotic agents for




     interfacial films.  Proceed. Natl. Acad. Sci. USA 48: 1008-1014.




Cohn, M., M. L. Cohn, S. K. Wolfson, and F. M. Steichen.  1980.  Dibutyryl




     cyclic AMP - an effective antidote against lethal doses of amobarbital




     in rhesus monkey.  Fed. Proceed. 39: 406.




Cohn, M. L. and M. Cohn.  1980.  A new molecular theory of anesthesia.  Fed.




     Proceed. 39: 406.




Craig, P. N.  1971.  Interdependence between physical parameters and




     selection of substituent groups for correlation studies.  J. Med.




     Chem.14: 680-684.




Daxboeck, C. and G. F. Holeton.  .1978.  Oxygen receptors in the rainbow trout




     (Salmo gairdneri).  Can  .J. Zool. 56: 1254-1259.
                                     102

-------
Daxboeck, C. and G. F. Holeton.  1980.  The effect of MS222 on the hypoxic




     response of rainbow trout (Salmo gairdneri).




Draper, N. R. and H. Smith.  1966.  Applied Regression Analysis.  Wiley and




     Sons, New York, NY.




Dunn, N. J., Ill, and S. Wold.  1978.  A structure-carcinogenicity study of




     4-nitroquinoline 1-oxide using the SIMCA method of pattern recognition.




     J. Med. Chem. 21: 1001-1007.




Eyring, H., J. W. Woodbury, and J. S. D'Arrigo.  1973.  A molecular mechanism




     of general anesthesia.  Anesthesiology 38: 415-424.




Ferguson, J.  1939.  The use of chemial potentials as indices of toxicity.




     Proceed. Roy Soc. London 127B: 387-404.




Ferguson, J.  1951.  Mechanisme de la narcose.  IN: Colloques Internat'ionaux




;     du Centre National de la Recherche Scieritifique, Paris, p. 25.




Frank, R. E., W. E. Massy and D. G. Morrison.  .1965.  Bias  in multiple




     discriminant analysis.  J. Market. Res. 2: 250-257.




Fujita, T. and T. Ban.  1971.  Structure-activity of phenethylamines as




     substrates of biosynthetic enzymes of sympathetic transmitters.  J. Med.




     Chem. 14: 148-152.




Gage, P. W. and P. 0. Hamill.  1976.  Effects of several inhalation




     anesthetics on the kinetics of postsynaptic conductance changes in mouse




     diaphram.  Br. J. Pharmacol. 57: 263-272.




Gero, A.  1965.  Intimate study of drug-action III: Possible mechanism of




     drug action IN: Drill's Pharmacology in Medicine (ed. J. R. DiPalma),




     Blackiston Div., McGraw Hill Book Co.., New York, 3rd edition, pp.




     47-69.                                                       .




Glave, W. R. and C. Hansch.  1972.  Relationship between lipophilic character




     and anesthetic activity.  J. Pharm. Sci. 61(4): 589-591.




                               '103

-------
Gray, N. A. B.   1976.   Constraints  on "learning machine" classification




     methods.  Anal.  Chem.  48:  2265-2268.




Guedel, A. E.  1951.   Inhalation  anesthesia,  2nd ed.  pp. 10-52. McMillan Co.,




     New York.




Halsey, M. J., R. A.  Millar,  and  J.  A.  Button.   1974.  Molecular mechanisms




     in general  anesthesia.   Published  by  Churchill-Livingstone, Edinburgh,




     England.




Halsey, M. J., B. W.  Smith,  and:C.  J Green.   1978.   Pressure reversal of




     general  anesthesia - a multisite expansion hypothesis.  Br. J. Anaesth.




     50: 1091-1097.




Hansch, C. and T. Fujita.   1964.   p-^-^analysis A method for the correlation




     of biological activity and chemical  structure.  J. Amer. Chem. Soc. 68:




  '•  16=16-.-'   v  •'.•'•.•.'- >.';•:->; :':;':;:;:.'-;/-•'"-^';'''^; ;•-;--••'--••;'  =V-".V--. v   ''.''•-'::



Hansch, C., J. E. Quinlan,  and  G. L. Lawrence.   1968.  The linear free-energy




     relationship between partition coefficients and the aqueous.solubility




     of organic  liquids.  J.  Org. Chem. 33(1):  347-350.




Hansch, C., A. Leo,  S.  H. Unger,  K.  H.  Kim,  D.  Nikaitani, and E.  J.  Lien.




     1973.  "Aromatic" sub.stitue.nt  constants for structure-activity




     correlations.   J.  Med.  Chem. 16(11):  1207-




Hansch, C, S. H. Unger,  and A.  B. Forsythe.   1973.   Strategy in drug design.




     .cluster  analysis as an aid in the  selection of- substituents.  J. Med.




     Chem. 16: 1217-1222.                 '  >•




Hargens, A. R.,  R. W. Millard,  and K. Johanse.n.  1974.  High capillary




     permiability  in fishes.   Comp.  Biochem. Physiol. 48A: 675-680.




Haydon, D. A., B. M.  Hendry,  S. R.  Levinson, and.J. Requena.  1977.




     Anesthesia  by n-alkanes.-  A  comparative study, of nerve impulse blockage
                                      ,104 .

-------
      and the properties of black lipid bilayer membranes.   Biochem.  Biophys.




      Acta 470:  17-34.




 Henry,  D.  R. and J.  H.  Block.   1979.   Classification of drugs by discriminant




      analysis using  fragment molecular connectivity values.  J.  Med.  Chrm.




      22: 465-472.




 Hesser, C. M.,  L.  Fagraeus, and J.  Adolfson.   1978.  Roles of nitrogen,




      oxygen and carbon  dioxide in compressed  air narcosis.  Undersea Biomed.




      Res.   5: 391-400.




.Heymans, F., L. Le Therizien,  J.-J. God.froid.   .1.979.. .Quantitative




      structure-activity relationships for N-[(N1,N"-distributed-amino)acetyl]




      jarylamines for local anesthetic activity and acute toxicity.  J. Med.




      Chera. 23:  184-193.




 Hill,.M. W.  1978.  Interaction of lipid vesicles with-anesthetics.   Ann..




      N.Y.   Acad.,  Sci.  308: 101-110.




 Hopfinger, A. J.,  R. Potenzone, Jr.,  R. Pearlstein, 0. Kikuchi,  M. Shapiro,




      G.-W. A. Milne, and B. R. Heller.  1979.   Structure activity analyses




      int he classification of toxic chemicals.  Proc. Sym. on Safe Handling




      of Chemial carcinogens.  Ann Arbor Press, Ann Arbor,  MI  Vol 2,  p




      385-407, 1979.




 Hudson, D. R.,  G.  E. Bass and W. P. Purcell.   1970.  Quantitative




      structure-activity modesl, some  conditions for application  and




      statistical interpretation.  J.  Med. Chrm. 13: 1184-1189.




 Hunn, J. B.  1970.  Dynamics of MS222 in the  blood and brain of  freshwater




      fishes during anesthesia.  U.S.  Bur. Sport Fish Wildlife 42:  1-8.




 Hunn, J. B. and J. L.  Allen. 1974.   Movement  of drugs across the gills of




    •  "fishes.  Ann. Rve. Pharmacol.  14: 47-55.
                                      105

-------
Iwamura, H.  1980.  Structure-taste relationship of perillartine and nitro-




     and cyanoanline derivatives.  J. Med. Chem. 23: 308-312.




Jain, M. K., J. Gleeson, A. Opreti, and G. C. Upreti.  1978.  Intrinsic




     perturbing ability of alkanols in lipid bilayers.  Biochem. Biophys.




     Acta 507: 1-8.




Jain, M. K. and L. V. Wray, Jr.  1978.  Partition coefficients of alkanols in




     lipid bilayer/water.  Biochem. Pharmacol. 27: 1294-1296.




Jain, M. K., N. Y. Wu. and L. V. Wray.  1975.  Drug-induced phase change in




     bilayer as possible mode of action of membrane expanding drugs.  Nature




     255: 494-495.            .




Johnson, F. H., D. E. S. Brown, and D. A. Marsland.  1942.  Pressure reversal




     of the actions of certain narcotics.  J. Cell COmp. Physiol. 20:




  ;   269-276.        •    ''"•.""'":   -:' ;  ' '   ' •'" • :'<"^  •  ' ''  ••"]"•'"'•'' '  ".   '




Jurs, P. C., B. R. Kowalski,  and T. L. Isenbour.  1969.  Computerized




     learning machines applied to chemical problems.  Anal. CHem. 41: 21-27.




Kamaya, H., I. Ueda, and P. S. Moore.  1979.  Antagonism between high




     pressure and anesthetics in the thermal phase-transition of dipalmitoyl




     phosphatidylcholine bilayer.  Biochim. Biophys. Acta 550: 131-137.




Karickhoff, S. W., D. S. Brown and T. A. Scott.  1979.  Sorption of




     hydrophobic pollutants on natural sediments.  Water Res. 13: 241-248.




Kaufman, R. D.  1977.  Biophysical mechanisms' of anesthetic action:




     Historical perspective and review of; c'urrent theories.  Anesthesioiogy




     46: 49-62.




Keynes, R. D.  1972.  Excitable membranes.  Nature 239: 29-32.




Kier, L. B., and L. H. Hall.  1976.  Molecular Connectivity in Chemistry and




     Drug Research.  Academic Press, Inc., New York, N.Y.  257 pp.
                                     106

-------
Klontz, G. W. and L. S. Smith.  1968.  Methods of using  fish  as biological




     research subjects in method of animal experimentation  (ed. Cay, W.  I.)




     Vol. 3, pp. 323-379. Academic Press, New York.




Koblin, D. D. and E. I. Eger.  1979.  Current concepts theories of narcosis.




     N. Engl. J. Med. 301(22): 1222-1224.




Kowalski, B. R. and C. F. Bender.  1974.  The application of  pattern




     recognition to screening prospective anticancer drugs.   Adenocarcinoma




     755 biological activity test.  J. Amer. CHem. Soc.  93: 916-918.




Kubinyi, H.  1979.  Lipophilicity and biological activity.  Drug Res. 29(11),




     Nr. 8: 1067-1080.




Kubinyi, H.  1978.  Non-linear dependence of biological  activity on




     hydrophobic characteri: the bilinear model.  II Farmaco, Ed.  Sc.  34:




     248-276.   "    '       ' '   '    •     :  '  '/••''-  •'..'•'•'.  . ':  -•: ;"';'-/ •••".:'-":--'



Kubinyi, H.  1977.  Quantitative structure-activity relationships. 7. Tje




     bilinear model, a new model for nonlinear dependence of  biological




     activity on hydrophobic character.  J. Med. Chem. 20: 625-629.




Kubinyi, H. and O.-H. Kehrhahn.  1976a.  Quantitative  structure-activity




     relationships. 1. The modified Free-Wilson approach.  J. Med. Chem. 19:




     578-586.




Kubinyi, H. and O.-H. Kehrhahn.  1976b.  Quantitative  structure-activity




     relationships. 3. A comparison of different Free-Wilson  models.  J. Med.




     Chem. 19(8): 1042-1049.




Kuschinsky, G. and Lullmann, H.  1973.  Textbook of Pharmacology, Chapter 4.




     Acad. Press Inc., New York, N. Y. 1-407.




Lee, A. G.  1976.  Model for action of local anaesthetics...  .Nature 262:




  .   545-548.  '         '             ..           •
                                      107

-------
Lever, M. J., K. W. Miller, W. D..M. Paton, and E. B. Smith.  1971.  Pressure




   ''""reversal of anesthesia.  Nature 231: 368-371.




Llinas, R. R. and E. J. Henser.  1977.  Depolarization-release coupling




     systems in neurons.  Neurosci. Res. Program Bull. 15: 556-687.




Longnecker, D. E. and P. D. Harris.  1980.  Microcirculatory actions of




     general anesthetics.  Fed. Proceed. 39: 1580-1583.




Malinowski, E. R.  1977a.  Determination of the number of factors and the




     experimental error in a data matrix.  Anal. Chem. 49(4): 612-617.




Malinowski, E. R.  1977b.  Theory of error in factor analysis.  Anal. chem.




     49(4): 606-612.




Martin, Y. C. and H. N. Farias.  1979.  Mathematical considerations in series




     design.  J. Med. Chem. 22: 784-791.




Marvin, D; E. and D.< T. 'Burton.  ,1973.  Cardiac and'respiratory .responses-of .




     rainbow trout, blue gills and brown bullhead catfish during rapid'




     hypoxia and recovery under normaxic conditions.  Comp. Biochem. Physio.




     46A: 755-766.




McCreery, M. J. and W. A. Hunt.  1978.  Physico-chemical correlates of




     alcohol intoxication.  Neuropharmacol.  17: 451-461.




McElroy, W. D.  1947.  Mechanism of inhibition of cellular activity by




     narcotics.  Quart. Rev.. Biol. 22: 25-58.




McFarland, J. W.  1970.  On the parabolic relationship between drug potency




     and hydrop'hobicity.  J. Med. CHem. 13:' 1192-1196.




McFarland, W. N. and G. W. Klontz.  1969.  Anesthesia in fishes.  Fed.




     Proceed. 28: 1535-1540.
                                     108

-------
Meyer, K. H.   1899.  Zur Theorie der Alkoholnarkose.  Arch. Exper. Path. U.




     Pharmakol. 42: 109-118.




Meyer, K. H.  1937.  Contributions to the theory of narcosis.  Trans. Faraday




     Soc. 33: 1062-1068.




Miller, K. W., W. D. M. Paton, and R. A. Smith.  1973.  The pressure reversal




     of general anesthesia and the critical volume hypothesis.  Mol.




     Pharmacol. 9: 131-143.




Miller, J. C. and K. W. Miller.  1975.  Approaches to the mechanisms of




     action of general anesthetics.  Blaschko, H. F. K. ed., Physiological




     and pharmacological biochemistry.  Biochem, Series One, Vol. 12, .




     Publisher Butterworth, london, pp. 33-76.




Miller, K. W.  1977.  Towards the molecular bases of anesthetic action.




   .  Anesthesiology 46: 2-4. : ':  ' .   ,    ;          .  .     :       '.-••'




Miller, K. W. and K. Y. Y. Pang.  1976.  General anesthetics can selectively




     perturb lipid bilayer membranes.  Nature 263: 253-255.




Morrison, D. G.  1969.  On the interpretation of discriminant analysis.  J.




     Market. Res. 6: 156-163.




Mullins, L. J.  1954.  Some physical mechanisms in narcosis.  Chem. Rev. 54:




     289-323.




Ogino, A., S. Matsumura, and T.  Fujita.  1980.  Structure-activity study of




     antiulcerous and antiinflammatory drugs "by discriminant analysis.  J.




     Med. Chrm. 23: 437-444.




Operating Instructions Microtox™ Toxicity Analyzer Model 2055, Interim Manual




     Number 110679, Beckman Instruments, Inc., Microbics.Operations,   .




     Carlsbad, CA, 1979.  .       .        •   '-."..        .. .'--      '    '
                                     109

-------
Otto, P., M. Steel, J; Ladiki, R. Muller.  1979.  From retrospective to




     predictive structure-activity correlations.  J. Theor. Biol. 78:..




     197-210.




Overton, E.  1901.  Studied uber de narcose Zugleich ein Beitrag zur




     allgemeinen pharmakologie Jena: Verlag von Gustav Fischer: 1901.




Pauling, L.  1961.  Molecular theory of general anesthesia.  Science 134:




     15-21.




Price, H. L. and S. T. Ohnishi.  1980.  Effects of anesthetics on heart.




     Fed. Proceed. 39: 1575-1579.




Pringle,,M. J. and K. W. Miller.  1978.  Structural isoraers of tetradecenol




     discriminate between lipid fluidity and phase transition theories of




     anesthesia.  Biochem. Biophys. Res. Commun.  85: 1192-1198.




Rahwan, R. G.  :1975.  Toxic effects of e;thanpl: possible role' of




     acetaldehyde, tetrahydroisoquinolines and tetrahydro-B-carbolines.




     Toxicol. Appl. Pharmacol. 34: 3-27.




Rasmussen, G. T., T. L. Isenhour, S. R. Lowry, and G. L. Ritter.  1978.




     Principal component anlaysis of the infrared spectra of mixtures.  Anal.




     Chim. Acta 103: 213-221.




Richards, C. D., K. Martin, S. Gregory, C. A. Keightley, .T. R. Hesketh, G. A.




     Smith, G. B. Warren and J. C. Metcalfe.  1978.  Degenerate perturbations




     of protein structure as the mechanism of anesthetic actions.  'Nature




     276: 775-779.                       ' ,:




Rimon, G., E. Hanski, S. Braun, and A. Levitzki.  1978.  Mode of coupling




     between hormone receptors and adenyLate eyelase elucidated by modulation




     of membrane fluidity.  Nature 276: 394-396.
                                     110.

-------
Rosenberg, Ph. H., E. Hansjorg, and S. Anton.  1975.  Biphasic effects of




     halothane on phospholipid and synaptic plasma membranes: a spin label




     study.  Mol. Pharmacol. 11: 879-888.




Roth, S. H.  1979.  Physical mechanism of anesthesia.  Ann. Rev. Pharmacol.




     Toxcol. 19: 159-178.




Roth, S. H.  1980.  Membrane and cellular actions of anesthetic agents.  Fed.




     Proceed. 39: 1595-1599.




Roth, S. H. and P. Seeraan.  1971.  All lipid soluble anesthetics protect red




     cells.  Nature New Biol. 231: 284-285.




Roth, S. H. and P. Seeman.  1972a.  Anesthetics expand erythrocyte membranes




     without causing loss of potassium.  Biochem. Biophys. Acta 255:




     190-198.




Roth, S. and P. Seeman.  1972b.  The membrane concentrations of' neutral and  ;




     positive anesthetics (alcohols, chlorpromzaine, morphine) fit the




     Meyer-Overton rule of anesthesia: negative narcosis do not.  Biochem.




     Biophys. Acta 255: 207-219.




Santara, N. J. and K. Augary.  1975.  Non-computer approach to




     structure-activity study.  An expanded fibonacci search applied to




     structurally diverse types of compounds.  J. Med. Chem. 18: 959-963.




Saxberg, B. E.H., D. E. Duewer, J. L. Booker and B. R. Kowalski.  1978.




     Pattern recognition and blind assay techniques applied to forensic




     separation of whiskies.  Anal. Chim. Acta 103: 201-212.




Schaad, L. J. and B. A. Hess, Jr.  1977.  Theory of linear equations as




     applied to quantitative structure-activity correlations.  J.-Med. Chem.




     20: 619-625.           ..               '.         '    -
                                     111

-------
Schettini, A. and W. W.  Furniss.   1979.   Brain water and electrolyte




     distribution during  the  inhal-ation: of  halothane.   Br. J. Anaesth. 51:




     1117-1124.




Schwab, S. H. and D. Schwab.   1979.   The transformation of cytoplasmic




     microtubules into helices  and paracrystals by halothane in the




     foraminifer Allogromi  l-'at icollar'is'. . Arnold Z. Mikrbsk Anat. Forsch




     93(4): 751-762.




Seeman, P.  1972.  Membrane actions  of anesthetics and tranquillizers.




     Pharmacol. Rev. 24:  583-655.




Smith, F. M. and D. R. Jones'.   1978.   Localization of receptors cau'sing




     hypoxic bradycardia  in trout  (-Sai'mo gairdneri).  Can. J. Zool. 56:




     1260-1265.




Steriger, V. G. 'arid' T. Hv Mar Vri.-f  1974.   The; pharmacology;'of .;MS222" (ethyl:-  ;




     m-aminobenzoate)  in Squalus acanthias.  Com'p. Gen. Pharmacol. 5: 23-35.




Stockard J. and R. Bickford'.   1975.   The neurophysiology of anesthesia.




     Gorden E., ed.  A basis  and practice of neuroanesthesia.  Amsterdam:




     Excepta-Med. pp.  3-4,6.




Topliss, J. G.  1977.  A manual method for  applying the Hansch approach to




     drug design.  J.  Medy  Chem!. 20('4>:. 463-469'.




Toplis.s, J.. G. and R.  P.   Edward's.   1-979'v   Chance factors in studies of




     quantitative structure-activity  relationships.  J. . Med...Chem. 22:   ,,




     1238-1244.                  •




Traube, J.  1904.  Theorie-der Osmose- und Narkbse.  Arch. Geo. Phys:iol.   105:




     541-558.




Trudell, J. R.  1977.-  A unitary  theory of  anestriesda bas'ed on lateral, phase




     separations in the.  nerve  membranes-. Ariesthesiblogy 46: 5-10'.
                                      112'

-------
Trudell, J. R., D. G. Payan, and J. H. Chin.  1975.  The antagonistic effect

     of an inhalation anesthetic and high pressure on the phase diagram of

     mixed dipalmitoyl-dimyristoylphosphatidylcholine bilayers,.  Proceed.

     Natl. Acad. Sci. USA 72: 201-210.

Unger, S. H. and C. Hansch.  1973.  On model building in structure-activity

     relationships.  A reexamination of adrenergic blocking activitr of

     B-Halo-B-arylalkylamineS;  J. Med. Chem. 16(7): 745-749.

U.S. EPA.  1978.  Experimental design, application and data interpretation

     protocol.  U.S. EPA, ORD-Corvallis, OR.  EPA-600/9/78-018.  126 pp.

Vandam, L. 0.  1965.  Uptake and transport of anesthetics and  stages of

 1    anesthesia IN:  Drill's Pharmacology in Medicine (ed. J.  R. DePalma),

     Blakiston Div. McGraw-Hill Book Co., New York, 3rd edition, pp. 85-99.

Veith, G. D., N. M. Austin, and R. T.- Morris.  .1979.  A rapid  method for.  ;

     estimating log P for organic chemicals.  Water Res.' 13: 43-47.
          \   • ,  - • -           ' .  •  •       • • •     • '• •    ,     ~     - '   '  '' -      •.
Veith, G. D., D. L. DeFoe, and B.'V: Bergstedt.  1979.  Measuring and

    • estimating the bioconcentration factor of chemicals in fish.  J. Fish.

     Res. Board Can. 36: 1040-1048.

Warren, G. B., M. D. Housley, J. C. Metcalfe, and N. J. M. Birdsall.  1975.

     Cholesterol is excluded from the phospholipid annulus surrounding an
          -*' ^
     active calcium transport protein.  Nature 255: 684-687.

Weisel, C. P. and J. L. Fasching.  1977.  Dec-eptive "correct"  separation by

     the  linear learning machine.  Anal. Chem. 49: 2114-2116.

Winters, W. D.  1976.  Effects of drugs on the electrical activity of the

     brain: anesthetics.  Ann. Rev. Pharmacol. 413-426.
                                     113

-------
Wu, R. S., M. K. Wolpert-DeFilippes,  and F.  R. Quinn.   1980.   Quantitative


     structure-activity  correlations  of rifamycins  as  inhibitors  of  viral


     RNA-directed DNA'polymerase  and  mammalian a  and  8 DNA  polymerases.  J.


     Med. Chem. 23:  256-261.


Yalkowsky, S. H. and G.  L.  Flynn.   1973.. Transport  of alkyl  homologs  across


     synthetic and .biological membranes:  A  new model  for chain


     length-activity relationships.   J. Pharm. Sci.  62(2):  210-217.


Zaleska, M.  1979.   Effect.of pentobarbital  anaesthesia on  serotonin

                       •      .".-•••     '       -'^ >  ..'. '-.
     metabolism in hypoxic  rat  brain.  Pol.  J. Pharraaco.l. Pharm.  31(3):


     179-186.
                                     . 114

-------

-------