{"title": "Presynaptic Neural Information Processing", "book": "Neural Information Processing Systems", "page_first": 154, "page_last": 163, "abstract": null, "full_text": "154 \n\nPRESYNApnC NEURAL INFORMAnON PROCESSING \n\nL.  R.  Carley \n\nDepartment of Electrical and  Computer  Engineering \n\nCarnegie  Mellon  University,  Pittsburgh  PA  15213 \n\nABSTRACT \n\nThe  potential  for  presynaptic  information  processing  within  the  arbor \nof a  single  axon  will  be  discussed  in  this  paper.  Current  knowledge  about \nthe  activity  dependence  of  the  firing  threshold,  the  conditions  required  for \nconduction  failure,  and  the  similarity  of  nodes  along  a  single  axon  will  be \nreviewed.  An  electronic  circuit  model  for  a  site  of low  conduction  safety  in \nan  axon  will  be  presented.  In  response to  single  frequency  stimulation  the \nelectronic circuit acts  as  a  lowpass filter. \n\nI.  INTRODUCTION \n\nThe  axon  is  often  modeled  as  a  wire  which  imposes  a  fixed  delay on  a \npropagating  signal.  Using  this  model,  neural  information  processing  is \nperformed  by  synaptically  sum m ing  weighted  contributions  of  the  outputs \nfrom  other  neurons.  However,  substantial  information  processing  may  be \nperformed  in  by  the  axon  itself.  Numerous  researchers  have  observed \nperiodic  conruction  failures  at  norma! physiological  impulse  activity  rates \n(e.g.,  in  cat,  in  frog 2,  and  in  man  ).  The  oscillatory  nature  of  these \nconduction  failures  is  a  result  of  the  dependence  of the  firing  threshold  on \npast impulse  conduction  activity. \n\nThe  simplest  view  of  axonal  (presynaptic)  information  processing  is \nas  a  switch:  the  axon  will  either  conduct  an  im pulse  or  not.  The  state  of \nthe  switch  depends  on  how  past  impulse  activity  modulates  the  firing \nthreshold,  which  will  result  in  conduction  failure  if firing  threshold  is  bigger \nthan  the  incoming  impulse  strength. \nIn  this  way,  the  connectivity  of  a \nsynaptic  neural  network  could  be  modulated  by  past  impulse  activity  at \nsites  of  conduction \nthe  network.  More  sophisticated \npresynaptic  neural  information  processing  is  possible  when  the  axon  has \nmore than  one  terminus,  implying  the  existence  of  branch  points  within  the \naxon.  Section \nII  will  present  a  general  description  of  potential  for \npresynaptic  information  processing. \n\nfailure  within \n\nThe  after-effects  of previous  activity  are  able  to  vary  the  connectivity \nof  the  axonal  arbor  at  sites  of  low  conduction  safety  according  to  the \ntemporal  pattern  of  the  impulse  train  at  each  site  (Raymond  and  LeUvin, \n1978; Raymond,  1979).  In  order  to  understand  the  inform ation  processing \npotential  of presynaptic  networks  it  is  necessary  to  study  the  after- effects \nof  activity  on  the  firing  threshold.  Each  impulse  is  normally  followed  by  a \nbrief  refractory  period  (about  10m s  in  frog  sciatic  nerve)  of  increased \n\n\u00a9 American Institute of Phvl'if:<'  1 qR~ \n\n\f155 \n\nthreshold  and  a  longer  superexcitable  period  (about  1  s  in  frog  sciatic \nnerve)  during  which  the  threshold  is  actually  below \nlevel. \nDuring  prolonged  periods  of  activity,  there  is  a  gradual  increase  in  firing \nthreshold  which  can  persist  long  (>  1  hour  in  frog  nerve)  after  cessation \nof  im pulse  activity  (Raymond  and  Lettvin,  1978). \nIII,  the \nmethods  used  to  measure  the  firing  threshold  and  the  after-effects  of \nactivity will  be  presented. \n\nIn  section \n\nits  resting \n\nIn  addition  to  understanding  how  impulse  activity  modulates  sites  of \nlow  conduction  safety,  it  is  important  to  explore  possible  constraints  on \nthe  distribution  of  sites  of  low  conduction  safety  within  the  axon's  arbor. \nSection  IV  presents  results  from  a  study  of  the  distribution  of  the  after(cid:173)\neffects  of activity along  an  axon. \n\nSection  V  presents  an  electronic  circuit  model  for  a  region  of  low \nconduction  safety  within  an  axonal  arbor.  It  has  been  designed  to  have  a \nfiring  threshold  that depends  on  the  past activity  in  a  manner  similar  to  the \nactivity  dependence  measured  for frog  sciatic  nerve. \n\nII.  PRESYNAPTIC  SIGNAL  PROCESSING \n\nConduction  failure  has  been  observed  in  many  diffe~e~t  organisms, \n\nincluding  man,  at  normal  physiological  activity  rates. 1,  ,  The  after(cid:173)\neffects  of  activity  can  \"modulate\"  conduction  failures  at  a  site  of  low \nconduction  safety.  One  common  place  where  the  conduction  safety  is  low \nis  at branch  points  where  an  impedance  mismatch  occurs  in  the  axon. \n\nIn  order  to  clarify  the  meaning  of  presynaptic  information  processing, \na  simple  example  is  in  order.  Parnas  reported  that  in  crayfish  a  single \naxon  separately  activates  the  medial  (DEA~~ and  lateral  (DEAL)  branches \nof  the  deep  abdominal  extensor  muscles.'  At  low  stimulus  frequencies \n(below  40-50  Hz)  impulses  travel  down  both  branches;  however,  each \nimpulse evokes  much  smaller contractions  in  DEAL than  in  DEAM  resulting \nin  contraction  of  DEAM  without  significant  contraction  of  DEAL.  At  higher \nstim ulus  frequencies  conduction  in  the  branch  leading  to  D EAM  fails  and \nDEAL  contracts  without  DEAM  contracting.  Both  DEAL  and  DEAM  can  be \nstim ulated  separately  by  stim ulus  patterns  more  com plicated  than  a  single \nfrequency. \n\nThe  theory  of  \"fallible  trees\",  which  has  been  discussed  by  Lettvin, \nMcCulloch  and  Pitts,  Raymond,  and  Waxman  and  Grossman  among \nothers,  suggests  that  one  axon  which  branches  many  times  forms  an \ninformation  processing  element  with  one  input  and  many  outputs.  Thus, \nthe  after-effects  of  previous  activity  are  able  to  vary  the  connectivity  of \nthe  axonal  arbor  at  regions  of  low  conduction  safety  according  to  the \ntemporal  pattern  of the  impulse train  in  each  branch.  The transfer function \nof  the  fallible  tree  is  determined  by  the  distribution  of  sites  of  low \nconduction  safety and the  distribution  of superexcitability  and  depressibility \nat  those  sites.  Thus,  a  single  axon  with  1000  terminals  can  potentially  be \nin  21000  different  states  as  a  function  of the  locations  of sites  of conduction \nfailure  within  the  axonal  arbor.  And,  each  site  of  low  conduction  safety  is \n\n\f156 \n\nmodulated  by  the  past impulse activity at that  site. \n\nto  cause  different \nfallible \n\nFallible  trees  have  a  number  of  interesting  properties.  They  can  be \nused \nto  excite  different  axonal \nterminals.  Also, \ntiming \ninformation  in  the  input  signal;  Le.,  starting  from  rest,  all  branches  will \nrespond  to  the first  impulse. \n\ninput \ntrees,  starting  at  rest,  will  preserve \n\nfrequencies \n\nIII. AFTER- EFFECTS  OF ACTIVITY \n\nIn  this  section,  the  firing  threshold  will  be  defined  and  an  experimental \nmethod  for  its  measurem ent  will  be  described. \neffects  of  activity  will  be  characterized  and \ncharacterization  process will be given. \n\nIn  addition,  the  after(cid:173)\ntypical \n\nresults  of \n\nthe \n\nThe  following  method  was  used  to  measure  the  firing  threshold. \nWhole  nerves  were  placed  in  the  experimental  setup  (shown  in  figure  1). \nThe  whole  nerve \nfiber  was  stim ulated  with  a  gross  electrode.  The \nresponse  from  a  single  axon  was  recorded  using  a  suction  microelectrode. \nFiring  threshold  was  measured  by  applying  test  stimuli  through  the  gross \nstimulating  electrode  and \nthe  suction \nm icroelectrode. \n\nresponse \n\nlooking \n\nfor  a \n\nin \n\nF ixed-duration \n\nvariable-amplitude \ncurrent stimulator \n\nAg-AgCI \nelectrode \n\n0\u00b74 mm diameter \nf--MOVES~ \n\nMotor(cid:173)\ndriven \nvernier \n\nmicrometer \n\n,  . \n\nSuction electrOdep' . t' . Refer~nce \n\n~;  lh \n\nSingle axon \n\nt .  \n\n~  suctIon \nelectrode \n\n\u2022\u2022 \n\nA \n\nWhole nerve \n\n\\ \n\nFigure  1. Drawing  of the  experimental  recording  chamber. \n\nThreshold  Hunting,  a  process  forschoosin g  the  test  stimulus  strength, \nwas  used  to  characterize  the  axons. \nIt  uses  the  following  paradigm.  A \ntest  stimulus  which  fails  to  elicit  a  conducting  impulse  causes  a  small \nincrease  the  strength  of  subsequent  test  stimuli.  A  test  stim ulus  which \n\n\f157 \n\nelicits  an  im pulse  causes  a  small  decrease  in  the  strength  of  subsequent \ntest  stimuli.  Conditioning  Stimuli,  ones  large  enough  to  guarantee  firing  an \nimpulse,  can  be  interspersed  between  test  stimuli  in  order  to  achieve  a \ncontrolled  overall  activity  rate.  Rapid  variations  in  threshold  following  one \nor  more  conditioning  impulses  can  be  measured  by  slowly  increasing  the \ntime  delay  between  the  conditioning  stimuli  and  the  test  stimulus.  Several \nphases  follow  each  impulse.  First,  there  is  a  refractory  period  of  short \nduration  (about  10ms  in  frog  nerve)  during  which  another  impulse  cannot \nbe  initiated.  Following  the  refractory  period  the  axon  actually  becomes \nmore  excitable  than  at  rest  for  a  period  (ranging  from  200ms  to  1 s  in  frog \nnerve,  see  figure  2).  The  superexcitable  period  is  measured  by  applying  a \nconditioning  stimulus  and  then  delaying  by  a  gradually  increasing  time \ndelay  and  applying  a  test  stimulus  (see  figure  3).  There  is  only  a  slight \nincrease \nfollowing  multiple \nim pulses?  The  superexcitability  of  an  axon  was  characterized  by  the  % \ndecrease  of  the \nlevel  at  the  peak  of  the \nsuperexcitable  period. \n\nthe  peak  of  the  superexcitable  period \n\nits  resting \n\nthreshold \n\nfrom \n\nin \n\n5'(1)  fo,  P, 0.50 \n\n\u2022 5 \n\n+ \n\no~! ____  ~~~I ----~~~I ~--17~~'---IOc~~Td \n\nINTERVAL 'm .. c) \n\nCONO I TlONING \n\n:_TO'Ald-~ \n\nT[ST  5t IMULU5 1 \n\n:......-TO~ \n1 \n:_FRAMC  1 - ; - rRAMC  ?-:-\n\nco NOI T 10NING \n\nT [5T  5T IMULUS \n\n. \n\nFigure  2.  Typical  superexcitable \nperiod  in  axon  from  frog  sciatic \nnerve. \n\nFigure  3.  Stim ulus  pattern  used \nfor  measuring  superexcitability. \n\nDuring  a  period  of  repetitive  impulse  conduction,  the  firing  threshold  may \ngradually  increase.  After  the  period  of increased  im pulse  activity  ends,  the \nthreshold  gradually  recovers  from  its  maximum  over  the  course  of  several \nminutes  or  more  with  complete  return  of  the  threshold  to  its  resting  level \ntaking  as  long  as  an  hour or two  (in  frog  nerve)  depending  on  the  extent of \nthe  preceding  im pulse  activity.  The  depressibility  of  an  axon  can  be \ncharacterized  by  the  initial  upward  slope  of  the  depression  and  the  time \n\n\f158 \n\nconstant  of  the  recovery  phase  (see  figure  4).  The  pattern  of  conditioning \nand  test  stimuli  used  to  generate the  curve in  figure  4  is  shown  in  figure  5. \nDepression  may  be  correlated  with  microanatomical  changes  which \noccur  ira  the  glial  cells  in  the  nodal  region  during  periods  of  increased \nactivity.  During  periods  of  repetitive  stim ulation  the  size  and  num ber  of \nextracellular  paranodal  intramyelinic  vacuoles  increases  causing  changes \nin  the  paranodal geom etry. \n\nThreshold \\percenl.gt of rHling level) \n\nCond.t.on.,,!! \n\nburst \n\nTest \n\n200 \n\n120 \n\n40 \n\n\u00b0o+-' --5~-tO--15--2-0--2+-5--:'30 \n\nTime (min) \n\nFigure  4.  Typical  depression  in  an \naxon  from  frog  sciatic  nerve.  The \naverage  activity \nrate  was  4 \nimpulses/sec  between  the  5  min \nmark  and  the  10  min  mark. \n\nl' \nr-- ~ \n\n5 min  >\"  T \nOff \n\nOn \n\nTime \n\nFigure  5.  Stim ulus  pattern  used \nfor measuring  depression. \n\nIV. CONSTRAINTS  ON  FALLIBLE  TREES \n\nThe  basic  fallible  tree  theo ry  places  no  constraints  on  the  distribution \nof sites  of conduction  failure  among  the  branches  of a  single  axon.  In  this \nsection  one  possible  constraint  on  the  distribution  of  sites  of  conduction \nfailure  will  be  presented.  Experiments  have  been  performed  in  an  attempt \nto  determine  if  the  extremely  wide  variations  in  superexcitability  anS \ndepressibility  found  between  nodes  from  different  axons  in  a  single  nerve \n(particularly  for  depressibility)  also  occur  between  nodes  from  the  same \naxon. \n\nA  study  of  the  distribution  of  the  after-effects  of  activity  along  an \nunbranching  length  of  frog  sciatic  nerve isund  only  sm all  variations  in  the \nafter- effects  along  a \nsuperexcitability  and \ndepressibility  were  extremely  consistent  for  nodes  from  along  a  single \nunbranching  length  of axon  (see  figures  6  and  7).  This  suggests  that there \nmay  be a  cell-wide regulatory  system  that maintains  the  depressibility and \n\nsingle  axon. \n\nBoth \n\n\fsuperexcitability  at  com parable  levels  throug hout  the  extent  of  the  axon. \nThus,  portions  of  a  fallible  tree  which  have  the  same  axon  diameter  would \nbe  expected  to  have the  same  superexcitability and  depressibility. \n\n159 \n\n3.() \n\n95 \n\nSuperexcitability (%1 \n\n30 \n\n0 -8 \n\n2-5 \n\n8 -0 \n\n25 \n\n80 \n\nUpward slope  (\"'/minl \n\nFigure  6.  PDF  of  Superexcitabili(cid:173)\nty.  The  upper  trace  represents \nthe  PDF  of  the  entire  population \ntwo \nof  nodes  studied  and \nlower \nthe \nseparate  populations  of  nodes \nfrom  two  different axons. \n\nrepresent \n\ntraces \n\nthe \n\nFigure  7.  PDF  of  Depressibility. \nThe  upper  trace  represents  the \nPDF  of  the  entire  population  of \nnodes  studied  and  the  two  lower \nthe  separate \ntraces \npopulations  of  nodes \ntwo \ndifferent axons. \n\nrepresent \n\nfrom \n\nThis  study  did  not  examine  axons  which  branched,  therefore  it  cannot  be \nconcluded  that  superexcitability  and  depressibility  must  remain  constant \nthroughout  a  fallible  tree.  For  example,  it  is  quite  likely  that  the  cell \nactually  regulates  quantities  like  pump- site  density,  not  depressibility. \nIn \nthat  case,  daughter  branches  of  smaller  diameter  might  be  expected  to \nshow  consistently  higher  depressibility.  Further  research  is  needed  to \ndetermine  how  the  activity  dependence  of  the  threshold  scales  with  axon \ndiameter  along  a  single  axon  before  the  consistency  of  the  after-effects \nalong  an  unbranching  axon  can  be  used  as  a  constraint  on  presynaptic \ninformation  processing  networks. \n\nV.  ELECTRICAL  AXON  CIRCUIT \n\nThis  section  presents  a  simple  electronic  circuit  which  has  been \ndesigned  to  have  a  firing  threshold  that  depends  on  the  past  states  of  the \noutput  in  a  manner  similar  to  the  activity  dependence  measured  for  frog \nsciatic  nerve.  In  response  to  constant frequency  stimuli,  the  circuit  acts  as \n\n\f160 \n\na  low pass  filter  whose  corner  frequency  depends  on  the  coefficients  which \ndetermine the  after-effects of activity. \n\nFigure  B  shows  the  circuit  diagram  for  a  switched  capacitor  circuit \nwhich  approximates  the  after- effects  of  activity  found  in  the  frog  sciatic \nnerve.  The  circuit  employs  a  two  phase  nonoverlapping  clock,  e  for  the \neven  clock  and  0  for  the  odd  clock,  typical  of  switched  capacitor  circuits. \nIt  incorporates  a  basic  model  for  superexcitability  and  depressibility.  VTH \nrepresents  the  resting  threshold  of  the  axon.  On  each  clock  cycle  the  V'N \nis  com pared  with  VTH+ Vo- Vs. \n\nThe two  capacitors  and  three  switches  at the  bottom  of figure  B model \nthe  change  in \nthreshold  caused  by  superexcitability.  Note  that  each \nimpulse  resets  the  comparator's  minus  input  to  (1-cx.)VTH,  which  decays \nback  to  VTH  on  subsequent  clock  cycles  with  a  time  constant  inversely \nproportional  to  Ps.  This  is  a  slight  deviation  from  the  actual  physiological \nsituation  in  which  multiple  conditioning  im pulses  will generate  slightly  more \nsuperexcitability than  a  single  impulse? \n\nThe  two  capacitors  and  two  switches  at  the  upper  left  of  figure  B \nmodel  the  depressibility  of  the  axon.  The  current  source  represents  a \nfixed \nthreshold  with  every  past  impulse.  The \ndepression  voltage  decays  back  to  0  on  subsequent  clock  cycles  with  a \ntime  constant inversely proportional to  PO. \n\nincrement  in \n\nthe  firing \n\nFigure  B.  Circuit diagram  for  electrical circuit analog  of nerve threshold. \n\nThe  electrical  circuit  exhibits  response  patterns  similar  to  those  of \nneurons  that  are  conducting  intermittently  (see  figure  9).  During  bursts  of \nconduction,  the  depression  voltage  increases  linearly  until  the  comparator \n\n\f161 \n\nfails  to  fire.  The  electrical  axon  then  fails  to  fire  until  the  depression \nvoltage  decays  back  to  (1 +aOV)VTH'  The  connectivity  between  the  input \nand  output  of  the  axon  is  defined  to  be  the  average  fraction  of  impulses \nwhich  are  conducted. \nIn  terms  of  connectivity,  the  electrical  axon  model \nacts  as  a  lowpass filter  (see  figure  10). \n\nriftiNG  VD ' \n\ntll4  Vs \n\nYES \n\nNO \n\n.. ~ \n\nrlUINC  \"'I1ACTI(lN \n\n, . \\ \u2022\u2022 \n\n\u2022\u2022 \n\no  : \n\nv S \n\n,00 \n\n10 \n\n300 \n\nT,,.a:  \u00abSl:CONUS  I \n\no. :.I--~~----;-t-----;2r-------.c. \nINruT  rR(: QVE~' \n\ntrace \n\nFigure  9.  Typical  waveform s  for \nconduction. \nintermittent \nThe \nupper \nindicates  whether \nimpulses  are  conducted  or  not. \nVD  and  Vs  are \nthe  depression \nvoltage  and \nthe  superexcitable \nvoltage  respectively. \n\naxon  model. \n\nFigure  10.  Frequency  response  of \nThe \nelectrical \nconnectivity  is  reflected  by \nthe \nfraction  of \nimpulses  which  are \nconducted  out  of  a  seq uence  of \n100.000 \nthe \nfrequency  is  in  stim uli/second. \n\nstimuli \n\nwhere \n\nFor  a  fixed  stim ulus  frequency.  the  average \n\nfraction  of  im pulses \nwhich  are  conducted  by  the  electrical  model  can  be  predicted  analytically. \nThe  expressions  can  be  greatly  simplified  by  making  the  assumption  that \nVD  increases  and  decreases  in  a  linear fashion.  Under  that  assumption.  in \nterms  of the  variables  indicated  on  the  schematic  diagram, \n\nwhere  M  is  the  number  of  clock  cycles  between  input  stimuli.  which  is \ninversely  proportional  to  the  input frequency.  The  frequency  at which  only \nhalf  of  the  impulses  are  conducted  is  defined  as  the  corner  frequency  of \nthe  low pass  filter.  The  corner frequency  is \n\n\f162 \n\nf(P  ==  0.5) _...!.  ==  log(1-~D) \naD \nlog(1--) \n\nM \n\naOV \n\nUsing \nfrequency  can  be  designed. \n\nthe  above  equations, \n\nlowpass \n\nfilters  with  any  desired  cutoff \n\nThe  analysis  indicates  that  the  corner  frequency  of  the  lowpass  filter \ncan  be  varied  by  changing  the  degree  of  conduction  safety  (aov)  without \nchanging  either  depressibility  or  superexcitability.  This  suggests  that  the \nexistence  of  a  cell- wide  regulatory  system  maintaining  the  depressibility \nand  superexcitability  at  comparable  levels  throughout  the  extent  of  the \naxon  would  not prevent the  construction  of  a  bank  of  low pass  filters  since \ntheir  corner  frequencies  could  still  be  varied  by  varying  the  degree  of \nconduction  safety  (aov). \n\nVI.  CONCLUSIONS \n\nimportant \n\nrole \n\nin \n\ninterfering  with \n\nimpulse  conduction.11  This  suggests \n\nRecent  studies  report  that  the  primary  effect  of  several  common \nanesthetics  is  to  abolish  the  activity  dependence  of  the  firing  threshold \nwithout \nthat \npresynaptic  processing  may  play  an \nhuman \nconsciousness.  This  paper  has  explored  some  of  the  basic  ideas  of \npresynaptic  information  processing,  especially  the  after- effects  of  activity \nand  their  modulation  of  impulse  conduction  at  sites  of  low  conduction \nsafety.  A  switched  capacitor  circuit  which  sim ulates  the  activity  dependent \nconduction  block  that  occurs  in  axons  has  been  designed  and  simulated. \nSimulation  results  are  very  similar  to  the  intermittent  conduction  patterns \nmeasured  experimentally \ninformation \nprocessing  possibility  for  the  arbor  of  a  single  axon,  suggested  by  the \nanalysis  of  the  electronic  circuit,  is  to  act  as  a  filterbank;  every  terminal \ncould  act as a  lowpass filter  with  a  different corner frequency. \n\nfrog  axons.  One  potential \n\nin \n\nBIBLIOGRAPHY \n\n[1]  Barron  D.  H.  and  B.  H.  C.  Matthews,  Intermittent  conduction  in  the \n\nspinal  chord.  J.  Physiol.  85,  p.  73-103 (1935). \n\n\f163 \n\n[2]  Fuortes  M.  G.  F.,  Action  of  strychnine  on \n\n\"intermittent \nconduction\" of impulses  along  dorsal  columns  of the  spinal  chord  of \nfrogs.  J.  Physiol.  112,  p.42  (1950). \n\nthe \n\n[3]  Culp  W.  and  J.  Ochoa,  Nerves  and  Muscles  as  Abnormal  Impulse \n\nGenerators.  (Oxford  University  Press,  London,  1980). \n\n[4]  Grossman  V.,  I.  Parnas,  and  M.  E.  Spira,  Ionic  mechanisms  involved \nin  differential  conduction  of  action  potentials  at  high  frequency  in  a \nbranching  axon. J.  Physiol.  295,  p.307 - 322  (1978). \n\n[5]  Parnas  I.,  Differential  block  at  high  frequency  of  branches  of  a \nsingle  axon  innervating  two  muscles.  J.  Physiol.  35,  p.  903-914, \n1972. \n\n[6]  Carley,  L.R.  and  S.A.  Raymond,  Threshold  Measurement: \nApplications  to  Excitable  Membranes  of  Nerve  and  Muscle.  J. \nNeurosci.  Meth.  9,  p.  309 - 333  (1983). \n\n[7]  Raymond  S.  A.  and  J.  V.  Lettvin,  After-effects  of  activity \n\nin \nperipheral  axons  as  a  clue  to  nervous  coding.  In  Physiology  and \nPathobiology of Axons,  S.  G. Waxman  (ed.),  (Raven  Press,  New  York, \n1978), p.  203 - 225. \n\n[8]  Wurtz  C.  C.  and  M.  H.  Ellisman,  Alternations  in  the  ultrastructure  of \nrepetitive  action \n\nperipheral  nodes  of  Ranvier  associated  with \npotential propagation.  J.  Neurosci.  6(11),  3133- 3143  (1986). \n\n[9]  Raym ond  S.  A.,  Effects  of  nerve  im pulses  on  threshold  of  frog \n\nsciatic  nerve fibers. J.  Physiol.  290,273- 303  (1979). \n\n[10]  Carley,  L.R.  and  S.A.  Raymond,  Com parison  of the  after- effects  of \nimpulse  conduction  on  threshold  at  nodes  of  Ranvier  along  single \nfrog  Sciatic axons.  J.  Physiol.  386,  p.  503 - 527  (1987). \n\n[11]  Raymond  S.  A.  and  J.  G.  Thalhammer,  Endogenous  activity(cid:173)\n\ndependent  mechanisms  for  reducing  hyperexcitability  ofaxons: \nEffects  of  anesthetics  and  CO 2 , \nIn  Inactivation  of  Hypersensistive \nNeurons,  N.  Chalazonitis  and  M.  Gola,  (eds.),  (Alan  R.  Liss  Inc.,  New \nVork,  1987), p.  331-343. \n\n\f", "award": [], "sourceid": 84, "authors": [{"given_name": "L.", "family_name": "Carley", "institution": null}]}