{"title": "Cholinergic Modulation Preserves Spike Timing Under Physiologically Realistic Fluctuating Input", "book": "Advances in Neural Information Processing Systems", "page_first": 111, "page_last": 117, "abstract": null, "full_text": "Cholinergic Modulation Preserves Spike \nTiming Under Physiologically Realistic \n\nFluctuating Input \n\nAkaysha C.  Tang \nThe Salk Institute \n\nHoward Hughes  Medical Institute \n\nComputational Neurobiology Laboratory \n\nLa Jolla, CA  92037 \n\nAndreas M.  Bartels \n\nZoological Institute \nUniversity of Zurich \n\nZiirich \n\nSwitzerland \n\nTerrence J. Sejnowski \n\nThe Salk Institute \n\nHoward Hughes Medical Institute \n\nComputational Neurobiology Laboratory \n\nLa Jolla,  CA  92037 \n\nAbstract \n\nNeuromodulation  can  change  not  only  the  mean  firing  rate  of a \nneuron,  but  also  its  pattern  of firing .  Therefore,  a  reliable  neu(cid:173)\nral  coding  scheme,  whether  a  rate  coding  or  a  spike  time  based \ncoding,  must  be  robust  in  a  dynamic  neuromodulatory  environ(cid:173)\nment.  The common observation that cholinergic modulation leads \nto  a  reduction  in  spike  frequency  adaptation  implies  a  modifica(cid:173)\ntion  of spike  timing,  which  would  make  a  neural  code  based  on \nprecise spike timing difficult to maintain.  In this paper, the effects \nof cholinergic modulation were  studied to test  the hypothesis that \nprecise  spike timing can serve  as a  reliable neural code.  Using  the \nwhole  cell  patch-clamp technique  in rat neocortical  slice  prepara(cid:173)\ntion and compartmental modeling techniques,  we show that cholin(cid:173)\nergic  modulation, surprisingly,  preserved  spike  timing in response \nto a fluctuating  inputs that resembles  in  vivo conditions.  This re(cid:173)\nsult suggests that in vivo spike timing may be much more resistant \nto changes in neuromodulator concentrations than previous physi(cid:173)\nological studies have  implied. \n\n\f112 \n\nA. C.  Tang, A.  M.  Bartels and T.  J. Sejnowski \n\n1 \n\nIntroduction \n\nRecently,  there has  been  a  vigorous debate concerning  the nature of neural coding \n(Rieke  et al.  1996;  Stevens and Zador 1995; Shadlen and Newsome 1994).  The pre(cid:173)\nvailing view  has  been  that the mean firing  rate conveys  all information about  the \nsensory  stimulus in  a  spike train  and the  precise  timing of the individual spikes  is \nnoise.  This belief is,  in part, based on a lack of correlation between the precise tim(cid:173)\ning of the spikes and the sensory qualities of the stimulus under study,  particularly, \non a lack of spike timing repeatability when identical stimulation is delivered.  This \nview  has been challenged by a  number of recent studies,  in which highly repeatable \ntemporal  patterns  of spikes  can  be  observed  both  in  vivo  (Bair  and  Koch  1996; \nAbeles  et al.  1993)  and  in  vitro (Mainen  and  Sejnowski  1994).  Furthermore, appli(cid:173)\ncation of information theory to the coding problem in the frog  and house fly  (Bialek \net  al. 1991; Bialek and Rieke  1992)  suggested  that additional information could be \nextracted from spike  timing.  In the absence of direct  evidence for  a  timing code in \nthe cerebral  cortex,  the role of spike  timing in neural coding remains controversial. \n\n1.1  A  necessary condition for a  spike timing code \n\nIf spike  timing  is  important  in  defining  a  stimulus,  precisely  timed  spikes  must \nbe  maintained  under  a  range  of physiological  conditions.  One  important  aspect \nof  a  neuron's  environment  is  the  presence  of  various  neuromodulators.  Due  to \ntheir  widespread  projections  in  the  nervous  system,  major neuromodulators,  such \nas acetylcholine  (ACh)  and norepinephrine  (NA), can have a  profound influence on \nthe firing  properties of most neurons.  If a  change in concentration of a neuromodu(cid:173)\nlator completely alters the temporal structure of the spike train , it would be unlikely \nthat  spike  timing could  serve  as  a  reliable  neural  code.  A  major effect  of cholin(cid:173)\nergic  modulation on cortical  neurons  is  a  reduction  in spike frequency  adaptation, \nwhich  is  characterized  by  a  shortening  of inter-spike-intervals  and  an  increase  in \nneuronal excitability  (McCormick  1993;  Nicoll  1988) .  One  obvious  consequence  of \nthis cholinergic effect  is  a  modification of spike timing (Fig.  1 A).  This modification \nof spike  timing  due  to  a  change  in  neuromodulator concentration  would  seem  to \npreclude  the  possibility of a  neural  code  based on precise spike timing. \n\n1.2  Re-examination of the cholinergic modulation of spike timing \n\nDespite  its  popularity,  the  square  pulse  stimulus used  in  most  eletrophysiological \nstudies  is  rarely  encountered  by  a  cortical  neuron  under  physiological  conditions. \nThe corresponding behavior of the neuron at the input/output level may have lim(cid:173)\nited  relevance  to  the  behavior of the  neuron  under  its  natural  condition,  which  is \ncharacterized  in  vivo  by  highly fluctuating  synaptic  inputs.  In  this  paper,  we  re(cid:173)\nexamine the effect  of cholinergic modulation on spike timing under two contrasting \nstimulus  conditions:  the  physiologically  unrealistic  square  pulse  input  versus  the \nmore plausible fluctuating  input.  We  report  that under physiologically more realis(cid:173)\ntic fluctuating inputs, effects of cholinergic modulation preserved the timing of each \nindividual spike  (Fig.  IB).  This result  is  consistent  with  the  hypothesis  that spike \ntiming may be  relevant  to information encoding. \n\n\fCholinergic Modulation Preserves Spike Timing \n\n113 \n\n2  Methods \n\n2.1  Experimental \n\nUsing the whole cell patch-clamp technique, we  made somatic recordings from layer \n2/3  neocortical  neurons  in  the  rat  visual  cortex.  Coronal  slices  of 400  p.m  were \nprepared from  14  to  18  days old Long  Evans  rats  (for details see  (Mainen  and  Se(cid:173)\njnowski 1994).  Spike trains elicited by current injection of 900 ms were  recorded for \nthe square pulse inputs and fluctuating  inputs with equal mean synaptic inputs,  in \nthe absence and presence of a cholinergic agonist carbachol.  The fluctuating  inputs \nwere constructed from Gaussian noise and convolved with an alpha function with a \ntime constant of 3 ms, reflecting the time course of the synaptic events.  The ampli(cid:173)\ntude of fluctuation  was such  that the subthreshold  membrane potential fluctuation \nobserved  in  our  experiments  were  comparable  to  that  in  whole-cell  patch  clamp \nstudy  in  vivo  (Ferster  and  Jagadeesh  1992).  The cholinergic  agonist  carbachol  at \nconcentrations of 5,7.5, 15,30 p.M  was delivered through bath perfusion (perfusion \ntime:  between  1 and 6 min).  For each cell, three sets of blocks were recorded  before, \nduring and after carbachol perfusion at a given concentration.  Each block contained \n20  trials of stimulation under  identical experimental conditions. \n\n2.2  Simulation \n\nWe used a compartmental model of a neocortical neuron to explore the contribution \nof three potassium conductances affected  by cholinergic modulation (Madison et ai. \n1987).  Simulations were  performed  in  a  reduced  9  compartment  model,  based  on \na  layer  2  pyramidal cell  reconstruction  using  the  NEURON  program.  The  model \nhad  five  conductances:  gNa,  gK v ,  9KM'  gCa,  gK(Ca)'  Membrane  resistivity  was \n40KOcm2 ,  capacitance  was  Ip.F/ p.m 2 ,  and  axial resistance  was  2000cm.  Intrinsic \nnoise was simulated by injecting a randomly fluctuating current to fit  the spike jitter \nobserved  experimentally.  Different  potassium  conductances  were  manipulated  as \nindependent variables and the spike timing displacement was measured for multiple \nlevels of conductance change corresponding to multiple concentrations of carbachol. \n\n2.3  Data analysis \n\nFor  both  experimental  and  simulation data,  first  derivatives  were  used  to  detect \nspikes  and  to  determine  the  timing for  spike  initiation.  Raster  plots  of the  spike \ntrains  were  derived  from  the  series  of membrane  potentials  for  each  trial,  and  a \nsmoothed  histogram  was  then  constructed  to  reflect  the  instantaneous  firing  rate \nfor  each  block of trials under identical stimulation and pharmacological conditions. \nAn event  was then defined  as a  period of increase  in  instantaneous firing  rate that \nis  greater  than  a  threshold  level  (set  at  3 times of the  mean firing  rate  within  the \nblock  of trials)  (Mainen and Sejnowski  1994). \n\nThe effect  of carbachol on spike  timing under fluctuating  inputs  was  quantified  by \ndefining  the displacement in spike timing for  each  event,  dj,  as  the  time difference \nbetween the nearest peaks of the events under carbachol and control condition.  The \nweight  for  each  event,  Wi,  is  determined  by  the  peak of the  event.  The higher the \npeak,  the less  the spike jitter.  The mean displacement is \n\n(1) \n\n\f114 \n\nA.  C.  Tang, A. M.  Bartels and T.  J.  Sejnowski \n\nwhere  i=  1,  2,  ... nth event  in  the control condition. \n\n3  Results \n\n3.1  Experimental \n\nThe  effects  of carbachol  on  spike  timing  under  the  square  pulse  and  fluctuating \ninputs  are  shown  in  Fig.  lA  and  B  respectively.  In  the  absence  of carbachol,  a \nsquare  pulse  input  produced  a  spike  train  with  clear  spike  frequency  adaptation \n(Fig.  lAl) .  Similar to  previous  reports  from  the  literature,  addition of carbachol \nto  the  perfusion  medium  reduced  spike  frequency  adaptation  (Fig.  lA2).  This \nreduction in spike frequency  adaptation is reflected  in the shortening of inter-spike(cid:173)\nintervals  and  an  increase  in  the  firing  frequency.  Most  importantly,  spike  timing \nwas  altered  by  carbachol  perfusion.  When  a  fluctuating  current  was  injected,  the \nstrong spike frequency adaptation observed under a square pulse input was no longer \napparent  (Fig.  IBl).  Unlike the results  under  the square pulse  condition,  addition \nof carbachol to the bath medium preserved  the timing of the spikes  (Fig.  IB2).  An \nincreased  excitability was  achieved  with the  insertion of additional spikes  between \nthe existing spikes. \n\nAt \n\n8t \n\n82 \n\nFigure 1:  Response of a  cortical neuron  to square pulse current injection (A)  and a \nfluctuating input (B).  The membrane potential during the 1024 ms sampling period \nis  plotted  as  a  function  of time for  the two  types of inputs  (onset:  5  ms;  duration: \n900  ms).  The grey lines show where  the spikes occurred  in the  upper traces. \n\nPreservation of spike  timing under carbachol  was  examined at concentrations of 5, \n7.5,  15,  and 30J.tM,  here shown in one cell  (Fig. 2,  5J.tM).  The smoothed histograms \n(as  described  in  section  2.3)  were  plotted  for  blocks  of 20  identical  trials  under \nthe same fluctuating  input.  The alignment of the events  between  the  control  and \ncarbachol indicates that spike timing was well  preserved.  The table gives the mean \nspike  displacement,  D,  for  a  range  of carbachol  concentrations.  The  spike jitter \nwithin the control  and carbachol conditions  was  approximately 1 ms,  and was  not \nchanged significantly by carbachol  (control:  0.96 \u00b1 0.3; carbachol:  0.94 \u00b1 0.42 ms.) \n\n3.2  Simulation \n\nThe model captured  the  basic  characteristics of experimental data.  In response  to \nfluctuating  inputs,  the  model neurons  showed  reduced  spike frequency  adaptation \nand  preservation  of spike  timing.  The  in  vitro  experiment  were  limited  to  only \ntwo  levels  of stimulus  fluctuation.  To  show  that  reduced  adaptation  in  response \n\n\fCholinergic Modulation Preserves Spike Timing \n\n115 \n\nControl \n\nCbolinergic Modification of Spike Timing \n\n1 \nr \n\n1 \n\nl \n\nCarbachol \n\nD(ms) \n\nN \n\nCarbachol \n(microM) \n\n2.76\u00b1 0.38 \n\n15 \n\n5-7.5 \n\n3.33 \n\n9.3 \n\n1 \n\n2 \n\n15 \n\n30 \n\nFigure  2:  Preservation  of spike  timing  for  a  range  of carbachol  concentrations. \nLeft:  the top  portion is  the  histogram for  the control condition;  the  bottom is  the \nhistogram for  the carbachol condition shown inverted.  The alignment of the events \nbetween  the control  and carbachol  indicates  preserved  timing.  Right:  statistics  of \nspike displacement. \n\n~ \nc:: \nQ \n'::I \nS \nQ, \n\n.a < \n\n55 \n\n48 \n\n41 \n\n34 \n\n27 \n\n20 \n\n0 \n\n50 \n\n100 \n\n150 \n\nFluctuation (pA) \n\nFigure  3:  Reduced  adaptation  as  a  function  of increasing  stimulus  fluctuation. \nAdaptation measured  as a  normalized spike count difference  between  the  first  and \nsecond  halves of the 900 ms stimulation:  (C2-Cl)/Cl. \n\nto fluctuating  inputs  is  a  general  phenomenon,  in  the  model neuron  we  measured \nadaptation for  multiple levels of stimulus fluctuation.  As shown in Fig. 3, spike fre(cid:173)\nquency adaptation decreased  as a function of increasing stimulus fluctuation over  a \nrange of fluctuation  amplitude.  The effects  cholinergic modulation on spike timing \nwere  studied  under simulated cholinergic  modulation.  Similar to the  experimental \nfinding,  increased  neuronal  excitability to  fluctuating  inputs  was  accompanied  by \ninsertion of additional spikes  (Fig.  4  left)  and spike  timing was  preserved  simulta(cid:173)\nneously  (Fig.  4 right). \n\nIn real neurons,  the total effects of cholinergic modulation depends on its effects  on \nat least three potassium conductances.  Using the model, we  examined the effects  of \nmanipulating each of the  three  potassium conductances on spike displacement and \nspike jitter.  We  found  that  (1)  spike  displacement  due  to  reduction  in  potassium \nconductances were all very small, on the order of a few  milliseconds (Fig. 5 top row); \n(2)  Compared to the conductances underlying 1M  and I, eak , spike displacement was \nmost  sensitive  to  changes  in  the  conductance  underlying  IAHP  (Fig.  5  top  row), \nwhose  reduction  alone  led  to  the  best  reproduction  of the  experimental  data;  (3) \nspike  jitters  of  approximately  1  ms  were  independent  of the  values  of the  three \n\n\f116 \n\nA.  C.  Tang, A.  M.  Bartels and T.  1.  Sejnowski \n\nlOOms \n1 30mV \n\nI \n\nFigure 4:  Preservation of spike  timing in the model neocortical  neuron.  Left:  Re(cid:173)\nsponses  of the  model neuron  to fluctuating  input.  Top:  replicating  data from  the \ncontrol condition.  Bottom:  reproducing  the carbachol effect  by blocking the  adap(cid:173)\ntation current,  IAHP.  Right:  histogram display of preservation of spike timing in  a \nblock of 20  trials. \n\npotassium  conductances  (Fig.  5  bottom row) .  These  results  make  predictions  for \nnew  experiments where  each  individual current  is  blocked selectively. \n\n4  Conclusions \n\nThe  results  showed  that  under  the  physiologically realistic  fluctuating  input,  the \neffects  of  cholinergic  modulation  on  spike  timing  are  rather  different  from  that \nobserved  when  unphysiological  square  pulse  inputs  were  used. \ning  the  spikes  forward  in  time  by  shortening  the  inter-spike-intervals,  cholinergic \nmodulation preserved  spike timing.  This preservation of spike timing was  achieved \nsimultaneously with an increase  in neuronal excitability. \n\nInstead  of  mov(cid:173)\n\nAccording  to the classical  view  of neuromodulation, one  would  have expected  that \na  spike  timing  based  neural code  would  be  difficult  to  maintain across  a  range  of \nneuromodulator concentrations.  The fact  that spike timing was  rather resistant  to \nchanges in the neuromodulatory environment raises the possibility that spike timing \nmay serve some function  in the cortex. \n\nThe differential effect  of cholinergic modulation on spike timing observed under the \nsquare  pulse  and fluctuating  inputs also  calls for  caution in generalizing  an  obser(cid:173)\nvation  from  one  set  of parameter  values  to  another,  especially  when  generalizing \nfrom  in  vitro to in  vivo.  This concern for  external validity is particularly important \nfor  computational neuroscientists  whose  work involves integrating phenomena from \nthe cellular, systems and finally,  to behavioral levels. \n\nAcknowledgments \n\nSupported  by  the  Howard  Hughes  Medical  Institute.  We  are  grateful  to  Zachary \nMainen,  Barak  Pearlmutter,  Raphael  Ritz,  Anthony  Zador,  David  Horn,  Chuck \nStevens,  William Bialek, and Christof Koch for  helpful discussions. \n\nReferences \n\nAbeles,  M.,  Bergman,  H.,  Margalit,  E.,  and  Vaadia,  E.  (1993).  Spatiotemporal \n\n\fCholinergic Modulation Preserves Spike Timing \n\n117 \n\nc u EL __ \n\",8 - '-' \n8'\" \n0.. \n'\" is \n\n5 \n\ngKleak \n\n: ~ \n~~ \n\n00  5  10  15  20  25 \n\n1:~ 1:~1:~ \n\n25 \n\n50 \n\n75 \n\n0 \n\n15 \n\n30 \n\n45 \n\n0  5  10  15  20  25 \n\n0.5 \no \n\no \n\n0.5 \n0 \n\n0.5 \n0 \n\nReduction of conductance (%) \n\nFigure  5:  Effects  of individual  conductance  changes  on spike  timing.  Top:  spike \ndisplacement  as  a  function  of changing  conductances.  Bottom:  spike  jitter  as  a \nfunction of changing conductances.  Each conductance was reduced from its control \nvalue  which  was  determined  by  fitting  experimentally observed  spike  trains.  The \nrange  of change for  the  leak  conductance  was  constrained  by  the  experimentally \nobserved  resting membrane potential changes  (avg.  5 m V.) \n\nfiring  patterns  in  the frontal  cortex  of behaving  monkeys.  J.  Neurophysiol., \n70,  1629-1638 . \n\nBair,  W .  and  Koch,  C.  (1996).  Temporal precision  of spike  trains  in extrastriate \ncortex of the  behaving  Macaque  monkey.  Neural  Computation,  8(6),  1184-\n1202. \n\nBialek, W. and Rieke, F. (1992).  Reliability and information transmission in spiking \n\nneurons.  Trends  Neurosci., 15,428-434. \n\nBialek,  W .,  Rieke,  F .,  de  Ruyter  van  Stevenick,  R.  R.,  and  Warland,  D.  (1991) . \n\nReading a  neural code.  Science,  252,  1854-7. \n\nFerster, D.  and Jagadeesh,  B. (1992) .  EPSP-IPSP interactions in cat visual cortex \nstudied  with  in  vivo  whole-cell  patch  recording.  J.  Neurosci.,  12(4),  1262-\n1274. \n\nMadison, D. V. , Lancaster,  B.,  and Nicoll,  R.  A. (1987) .  Voltage clamp analysis of \n\ncholinergic action in the hippocampus.  J. Neurosci.,  7(3), 733-741. \n\nMainen, Z.  F. and Sejnowski, T. J.  (1994).  Reliability of spike timing in neocortical \n\nneurons.  Science,  268,  1503-6. \n\nMcCormick,  D.  A.  (1993).  Actions  of acetylecholine  in  the  cerebral  cortex  and \n\nthalamus and implications for function ..  Prog.  Brain  Res.,  98,  303- 308. \n\nNicoll, R.  (1988).  The coupling of neurotransmitter receptors to ion channels in the \n\nbrain.  Science,  241, 545-550. \n\nRieke,  F. ,  Warland,  D.,  de  Ruyter  van  Steveninck,  R.,  and  Bialek,  W .  (1996). \n\nSpikes:  Exploring  the  Neural  Code.  MIT Press. \n\nShadlen,  M.  N.  and  Newsome,  W .  T.  (1994) .  Noise,  neural  codes  and  cortical \n\norganization.  Current  Opinion  in  Neurobiology,  4, 569-579. \n\nStevens,  C.  and  Zador,  A.  (1995) .  The enigma of the  brain.  Current Biology,  5, \n\n1-2. \n\n\f", "award": [], "sourceid": 1319, "authors": [{"given_name": "Akaysha", "family_name": "Tang", "institution": null}, {"given_name": "Andreas", "family_name": "Bartels", "institution": null}, {"given_name": "Terrence", "family_name": "Sejnowski", "institution": null}]}