{"title": "Dynamic Modulation of Neurons and Networks", "book": "Advances in Neural Information Processing Systems", "page_first": 511, "page_last": 518, "abstract": null, "full_text": "Dynamic Modulation of Neurons  and Networks \n\nEve Marder \n\nCenter for Complex Systems \n\nBrandeis University \n\nWaltham,  MA 02254 USA \n\nAbstract \n\nBiological neurons have a variety of intrinsic properties because of the \nlarge number of voltage dependent currents  that control their activity. \nNeuromodulatory substances modify both the balance of conductances \nthat determine intrinsic properties and  the strength of synapses.  These \nmechanisms alter circuit dynamics,  and suggest that functional circuits \nexist only in the modulatory environment in which  they operate. \n\n1  INTRODUCTION \n\nMany  studies of artificial neural  networks employ model neurons and  synapses  that are \nconsiderably simpler than their biological counterparts.  A variety of motivations underly \nthe use of simple  models  for neurons  and  synapses in artificial neural  networks.  Here, \nI discuss some of the properties of biological neurons and networks that are lost in overly \nsimplified models of neurons and synapses. A fundamental principle in biological nervous \nsystems is that neurons and networks operate over a wide range of time scales,  and that \nthese are modified by neuromodulatory substances.  The flexible,  multiple time scales  in \nthe nervous system allow smooth transitions between different modes of circuit operation. \n\n2 NEURONS HA VE DIF'FERENT INTRINSIC PROPERTIES \n\nEach  neuron has  complex dynamical  properties that depend on  the number and kind of \nion  channels  in  its  membrane. \nIon  channels  have  characteristic  kinetics  and  voltage \n\n511 \n\n\f512 \n\nMarder \n\ndependencies  that depend on the  sequence  of amino acids of the protein.  Ion channels \nmay  open  and  close  in  several  milliseconds;  others  may  stay  open  for  hundreds  of \nmilliseconds or several  seconds. \n\nSome  neurons  are  silent  unless  they  receive  synaptic  inputs.  Silent  neurons  can  be \nactivated  by  depolarizing  synaptic  inputs,  and  many  will  fire  on  rebound  from  a \nhyperpolarizing input (postinhibitory rebound).  Some neurons are tonically active in the \nabsence of synaptic inputs, and synaptic inputs will increase or decrease their firing rate. \n\nSome neurons  display rhythmic bursts of action potentials.  These bursting  neurons  can \ndisplay  stable  patterns  of oscillatory  activity,  that  respond  to  perturbing  stimuli  with \nbehavior  characteristic  of  oscillators,  in  that  their  period  can  be  stably  reset  and \nentrained.  Bursting  neurons  display  a  number of different  voltage and  time  dependent \nconductances  that  interact  to  produce  slow  membrane  potential oscillations with  rapid \naction potentials  riding  on  the  depolarized  phase.  In a  neuron  such  as  R15  of Aplysia \n(Adams  and  Levitan  1985)  or  the  AB  neuron  of the  stomatogastric  ganglion  (STG) \n(Harris-Warrick and Flamm 1987), the time scale of the burst is in the second range,  but \nthe individual action potentials are produced in the 5-10msec time scale. \n\nNeurons  can  generate  bursts  by  combining  a  variety  of different  conductances.  The \nparticular balance  of these conductances can  have  significant impact on the  oscillator's \nbehavior (Epstein and Marder 1990; Kepler et aI1990; Skinner et aI1993), and therefore \nthe choice of oscillator model to use must be made  with care (Somers and Kopell  1993). \n\nSome  neurons  have  a  balance  of  conductances  that  give  them  bistable  membrane \npotentials,  allowing  to  produce  plateau potentials.  Typically,  such  neurons  have  two \nrelatively stable states,  a hyperpolarized silent state,  and a sustained depolarized state in \nwhich they fire action potentials. The transition between these two modes of activity can \nbe made with a short depolarizing or hyperpolarizing pulse (Fig.  1).  Plateau potentials, \nlike \"flip-flops\" in electronics, are a \"short-term memory\"  mechanism for neural circuits. \n\nThe intrinsic properties of neurons can  be  modified by sustained changes  in membrane \npotential.  Because  the  intrinsic  properties  of  neurons  depend  on  the  balance  of \nconductances that activate and inactivate in different membrane potential ranges and over \na variety of time scales, hyperpolarization or depolarization can switch a neuron between \nmodes of intrinsic activity (Llinas  1988;  McCormick  1991;  Leresche  et aI1991). \n\nAn  interesting  \"memory-like\"  effect  is  produced by  the  slow inactivation properties of \nsome  K+  currents  (McCormick  1991;  Storm  1987).  In  cells  with  such  currents  a \nsustained  depolarization  can  \"amplify\"  a  synaptic \ninput  from  subthreshold  to \nsuprathreshold,  as  the  sustained  depolarization  causes  the  K+  current  to  inactivate \n(Marom and Abbott 1994; Turrigiano, Marder and Abbott in preparation). This is another \n\"short-term memory\"  mechanism  that does not depend  on changes in synaptic efficacy. \n\n\fDynamic Modulation of Neurons and Networks \n\n513 \n\nA.  CONTROL \n\n\\20mV \n\nI04nA \n\n1..., \n\nFigure 1:  Intracellular recording from the DG neuron ofthe crab STG.  A:  control saline, \na  depolarizing  current  pulse  elicits  action  potentials  for \nIn \nSDRNFLRFamide, a short depolarization elicits a plateau potential that lasts until a short \nhyperpolarizing current pulse terminates it.  Modified from Weimann  et al  1993. \n\nits  duration.  B: \n\n2  INTRINSIC MEMBRANE PROPERTIES  ARE MODULATED \n\nsystems  use  many \n\nBiological  nervous \nsubstances  as  neurotransmitters  and \nneuromodulators. The effects of these substances include opening of rapid, relatively non(cid:173)\nvoltage  dependent  ion  channels,  such  as  those  mediating  conventional  rapid  synaptic \npotentials.  Alternatively,  modulatory  substances  can  change  the  number  or  type  of \nvoltage-dependent conductances  displayed  by  a  neuron,  and  in  so  doing  dramatically \nmodify  the  intrinsic  properties  of a  neuron.  In  Fig.  1,  a  peptide,  SDRNFLRFamide \ntransforms  the  DG  neuron of the crab  STG  from  a state in which it fires  only during a \ndepolarizing pulse to one in which it displays  long-lasting plateau properties (Weimann \net  al  1993).  The  salient  feature  here  is  that  modulatory  substances  can  elicit  slow \nmembrane  properties not otherwise expressed. \n\n3  SYNAPTIC  STRENGTH IS MODULATED \n\nIn most neural network models  synaptic weights are  modified by learning rules,  but are \nnot  dependent  on  the  temporal  pattern  of presynaptic  activity.  In  contrast,  in  many \nbiological synapses the amount of transmitter released depends on the frequency of firing \nof the presynaptic neuron.  Facilitation, the increase in the amplitude of the postsynaptic \ncurrent when' the presynaptic neuron is activated several times in quick succession is quite \ncommon.  Other synapses  show depression.  The  same  neuron  may  show facilitation at \nsome  of  its  terminals  while  showing  depression  at  others  (Katz  et  al  1993).  The \nfacilitation  and  depression properties of any  given synapse can  not be  deduced on  first \nprinciples,  but must be  determined empirically. \n\nSynaptic efficacy is often modified by modulatory substances. A dramatic example is seen \nin the Aplysia gill withdrawal reflex, where serotonin significantly enhances the amplitude \nof the  monosynaptic connection from  the  sensory  to  motor neurons (Clark and  Kandel \n1993;  Emptage and  Carew  1993).  The effects  of modulatory  substances  can occur on \ndifferent  branches  on  a neuron  independently  (Clark  and  Kandel  1993),  and  the  same \nmodulatory  substance may  have different actions at different sites of the same neuron. \n\n\f514 \n\nMarder \n\nElectrical synapses are also subject to neuromodulation (Dowling,  1989).  For example, \nin the retina dopamine reversibly  uncouples horizonal cells. \n\nModulation of synaptic strength can  be  quite extreme;  in some  cases  synaptic contacts \nmay be virtually invisible in some modulatory environments, while strong in others. The \nimplications of this for  circuit ooeration will be discussed below. \n\nHormones \n.DA \nLJ5-HT \n\u2022  Oct \nII CCAP \nII cCCK \nIIlomTK \n\u2022  APCH \n\nNeuromodulators \n\u2022  ACh \n.OA \nrnm  GABA \n\n.HA \n\n\u2022  Oct \n\n\u2022  AST \n\u2022  Buc \nID  cCCK \n\n.LK \n\nI'IlomTK \n\n\u2022  Myomod \n1'1  Proc \n\u2022  RPCH \n\u2022  SOAN \n1'1  TNRN \n\nSensory Transmitters \n\n~~~ . \n\n\u2022  ACh  } \n\nFigure 2:  Modulatory substances  found in inputs to  the STG.  See Harris-Warrick et aI., \n1992  for details.  Figure courtesy of P.  Skiebe. \n\n4  TRANS:MITTERS  ARE COLOCALIZED IN NEURONS \n\nThe time course of a  synaptic potential evoked by  a neurotransmitter or modulator is a \ncharacteristic  property  of the  ion  channels  gated  by  the  transmitter  and/or  the  second \nmessenger system activated by the signalling molecule. Synaptic currents can be relatively \nfast,  such  as  the rapid  action of ACh  at  the  vertebrate  skeletal  neuromuscular junction \nwhere  the  synaptic  currents  decay  in  several  milliseconds.  Alternatively,  second \nmessenger activated synaptic events may have durations lasting hundreds of milliseconds, \nseconds,  or even  minutes.  Many  neurons  contain  several  differen.t neurotransmitters. \nIt is common to find  a small  molecule  such  as  glutamate or GABA  colocalized with an \namine  such  as  serotonin or histamine and  one or more  neuropeptides.  To describe  the \nsynaptic actions of such neurons, it is necessary to determine for each- signalling molecule \nhow  its  release  depends  on  the  frequency  and  pattern  of activity  in  the  presynaptic \n\n\fDynamic Modulation of Neurons and Networks \n\nSIS \n\nterminal,  and characterize  its  postsynaptic actions.  This is  important,  because  different \nmixtures  of cotransmitters,  and  consequently  of postsynaptic  action  may  occur  with \ndifferent presynaptic patterns of activity. \n\n5  NEURAL NETWORKS  ARE MULTIPLY MODULATED \n\nNeural  networks  are  controlled by  many  modulatory  inputs and  substances.  Figure 2 \nillustrates  the  patterns  of modulatory  control  to  the  crustacean  stomatogastric  nervous \nsystem,  where the motor patterns produced by  the only 30 neurons of the stomatogastric \nganglion are controlled by about 60 input fibers (Coleman et a11992) that contain at least \n15  different substances,  including a  variety  of amines,  amino  acids,  and  neuropeptides \n(Marder and Weimann  1992;  Harris-Warrick et aI1992).  Each  of these \nmodulatory substances produces characteristic and different effects on the motor patterns \nof the  STG  (Figs.  3,4).  This  can  be  understood  if one  remembers  that  the  intrinsic \nmembrane  properties  as  well  as  the  strengths  of the  synaptic  connections  within  this \ngroup of neurons are all subject to modulation. Because each cell has many conductances, \nmany  of which are  subject to  modulation,  and  because of the large number of synaptic \nconnections,  the  modes  of circuit operation are theoretically large. \n\n6 CIRCUIT RECONFIGURA TION BY MODULATORY CONTROL \n\nFigure  3  illustrates  that  modulatory  substances  can  tune  the  operation  of  a  single \nfunctional  circuit.  However,  neuromodulatory  substances  can  also  produce  far  more \nextensive changes  in the  functional organization of neuronal  networks.  Recent work on \nthe  STG  demonstrates  that  sensory  and  modulatory  neurons  and  substances  can  cause \nneurons  to  switch between different functional  circuits,  so  that the same neuron  is  part \nof several  different  pattern  generating  circuits  at  different  times  (Hooper  and  Moulins \n1989;  Dickinson et  al  1990;  Weimann  et  al  1991;  Meyrand  et  al  1991;  Heinzel  et  al \n1993). \n\nIn the example shown in Fig.  4, in control saline the LG neuron is firing in time with the \nfast pyloric rhythm (the LP neuron is also firing in pyloric time), but there is no ongoing \ngastric  rhythm.  When  the  gastric  rhythm was  activated  by  application of the  peptide \nSDRNFLRFNHz,  the  LG  neuron  fired  in  time  with  the  gastric  rhythm  (Weimann  et  al \n1993).  These and other data lead  us  to conclude  that it is  the  modulatory environment \nthat  constructs  the functional  circuit  that  produces  a  given  behavior  (Meyrand  et  al \n1991).  Thus,  by \ntuning  intrinsic  membrane  properties  and  synaptic  strengths, \nneuromodulatory  agents  can  recombine  the  same  neurons  into  a  variety  of  circuits, \ncapable of generating  remarkably  distinct outputs. \n\nAcknowledgements \n\nI thank Dr.  Petra Skiebe for  Fig 3 art work.  Research  was  supported by  NSI7813. \n\n\f516 \n\nMarder \n\nCONTROL \n\nPILOCARPINE \n\nSEROTONIN \n\nFigure 3:  Different forms of the pyloric rhythm different modulators.  Each panel,  the top \ntwo traces:  simulataneous intracellular recordings from LP and PD neurons of crab STG; \nbottom  trace:  extracellular  recording,  Ivn  nerve.  Control,  rhythmic  pyloric  activity \nabsent.  Substances  were  bath  applied,  the  pyloric  patterns  produced  were  different. \nModified from Marder and Weimann  1992. \n\ndgn \n\nFigure 4:  Neurons  switch between different pattem-genreating circuits.  Left panel,  the \ngastric rhythm not active (monitored by DG neuron), LG neuron in time with the pyloric \nrhythm  (seen  as  activity  in  LP  neuron).  Right  panel,  gastric  rhythm  activated  by \nSDRNFLRFamide, monitored  by  the DG  neuron  bursts  recorded on  the dgn.  LG  now \nfired  in  alternation  with  DG  neuron.  Pyloric  time  is  seen  as  the  interruptions  in  the \nactivity of the VD  neuron.  Modified from Marder and  Weimann  1992. \n\n\fDynamic Modulation of Neurons and Networks \n\n517 \n\nReferences \n\nAdams  WB  and  Levitan  IB  1985  Voltage  and  ion  dependencies  of the  slow  currents \n\nwhich mediate bursting in Aplysia neurone R,s'  J  Physiol 360 69-93 \n\nClark GA, Kandel ER 1993 Induction oflong-tenn facilitation in Aplysia sensory neurons \nby  local  application of serotonin to  remote synapses.  Proc Natl Acad Sci  USA \n90:  1141-11415 \n\nColeman MJ,  Nusbaum MP,  Cournil I,  Claiborne BJ  1992 Distribution of mopdulatory \ninputs to the stomatogastric ganglion of the crab,  Cancer borealis. J Comp Neur \n325:  581-594 \n\nDickinson  PS,  Mecsas  C,  Marder  E  1990  Neuropeptide  fusion  of two  motor-pattern \n\ngenerator circuits.  Nature 344:  155-158 \n\nDowling JE 1989 Neuromodulation in the retina:  the role of dopamine.  Sem  Neur 1:35-\n\n43 \n\nEmptage NJ and Carew TJ 1993 Long-term synaptic facilitation in the absence of short(cid:173)\n\nterm facilitation in Aplysia neurons.  Science 262 253-256 \n\nEpstein  IR,  Marder  E  1990  Multiple  modes  of a  conditional  neural  oscillator.  Bioi \n\nCybern  63:  25-34 \n\nHarris-Warrick RM, Flamm RE  1987 Multiple mechanisms of bursting in a conditional \n\nbursting neuron.  J  Neurosci 7:  2113-2128 \n\nHarris-Warrick RM, Marder E, Selverston AI, Moulins M eds 1992 Dynamic Biological \n\nNetworks:  The Stomatogastric Nervous System.  MIT Press  Cambridge \n\nHeinzel H-G, Weimann JM, Marder E  1993 The behavioral repertoire of the gastric mill \nin  the  crab,  Cancer pagurus:  An  in  vivo  endoscopic  and  electrophysiological \nexamination. J  Neurosci  13:  1793-1803 \n\nHooper  SL,  Moulins  M  1989  Switching of a neuron  from  one  network  to  another by \n\nsensory-induced changes  in membrane  properties.  Science 244:  1587-1589 \n\nKatz  PS,  Kirk  MD,  and  Govind  CK  1993  Facilitation  and  depression  at  different \nbranches  of  the  same  motor  axon:  evidence  for  presynaptic  differences  in \nrelease.  J  Neurosci  13:  3075-3089 \n\nKepler TB,  Marder E, Abbott LF 1990 The effect of electrical coupling on the frequency \n\nof model  neuronal oscillators.  Science 248:  83-85 \n\nLeresche N,  Lightowler S,  Soltesz I, Jassik-Gerschenfeld D, and Crunelli V  1991  Low(cid:173)\nfrequency  oscillatory activities  intrinsic  to  rat  and  car  thalamocortical  cells.  J \nPhysiol 441  155-174 \n\nLlinas  RR  1988  The  intrinsic  electrophysiological  properties  of mammalian  neurons: \n\ninsights into central nervous  system function.  Science  242  1654-1664 \n\nMcCormick DA  1991  Functional properties of slowly inactivating potassium current in \n\nguinea pig dorsal  lateral  geniculate relay  neurons.  J  Physiol 66  1176-1189 \n\nMarom  S  and  Abbott  LF  1994  Modeling  state-dependent  inactivation  of  membrane \n\ncurrents.  Biophysical J  in press \n\nMarder  E,  Weimann  JM  1992  Modulatory  control  of mUltiple  task  processing  in  the \nstomatogastric  nervous  system.  IN:  Neurobiology  of  Motor  Programme \nSelection:  new  approaches  to  mechanisms  of  behavioral  choice,  Kien  J, \n\n\f518 \n\nMarder \n\nMcCrohan C.  Winlow W  eds Pergamon Press Oxford \n\nMeyrand  p.  Simmers I. Moulins,  M  1991  Construction of a  pattern-generating circuit \n\nwith neurons of different networks.  NaJure  351:  60-63 \n\nSkinner FK. Turrigiano 00. Marder E  1993 Frequency and burst duration in oscillating \n\nneurons and  two cell networks.  Bioi Cybern  69:  375-383 \n\nSomers D. Kopell N  1993 Rapid synchronization through fast threshold modulation. Bioi \n\nCybern  68:  393-407 \n\nStorm IF  1987 Temporal integration by a slowly inactivating K+  current in hippocampal \n\nneurons.  NaJure 336:  379-381 \n\nWeimann  1M.  Meyrand  p.  Marder  E  1991  Neurons  that  form  multiple  pattern \ngenerators: Identification and multiple activity patterns of gastric/pyloric neurons \nin the crab stomatogastric system. J  Neurophysiol 65:  111-122 \n\nWeimann 1M,  Marder  E,  Evans B.  Calabrese  RL  1993  The effects  of SDRNFLRFNHl \nand TNRNFLRFNHl on the motor patterns of the stomatogastric ganglion of the \ncrab.  Cancer borealis. J Exp  Bioi 181:  1-26 \n\n\f", "award": [], "sourceid": 780, "authors": [{"given_name": "Eve", "family_name": "Marder", "institution": null}]}