{"title": "Characterizing Neurons in the Primary Auditory Cortex of the Awake Primate Using Reverse Correlation", "book": "Advances in Neural Information Processing Systems", "page_first": 124, "page_last": 130, "abstract": "", "full_text": "Characterizing Neurons in the Primary \nAuditory Cortex of the Awake Primate \n\nU sing Reverse Correlation \n\nR.  Christopher deC harms \n\ndecharms@phy.ucsf.edu \n\nMichael M .  Merzenich \n\nmerz@phy.ucsf.edu \n\nw. M.  Keck  Center for  Integrative Neuroscience \nUniversity of California, San Francisco CA 94143 \n\nAbstract \n\nWhile  the  understanding of the  functional  role  of different  classes \nof neurons in the awake primary visual cortex has been extensively \nstudied since the time of Hubel and Wiesel (Hubel and Wiesel, 1962), \nour  understanding  of the  feature  selectivity  and  functional  role  of \nneurons in  the primary  auditory  cortex  is  much farther  from  com(cid:173)\nplete.  Moving bars have long been recognized as an optimal stimulus \nfor  many visual cortical neurons,  and this finding  has recently been \nconfirmed  and extended in detail using reverse correlation methods \n(Jones and Palmer,  1987;  Reid  and Alonso,  1995; Reid  et al.,  1991; \nllingach et al.,  1997).  In this study, we recorded from neurons in the \nprimary auditory cortex of the awake  primate,  and used  a  novel  re(cid:173)\nverse correlation technique to compute receptive fields  (or preferred \nstimuli), encompassing both multiple frequency components and on(cid:173)\ngoing time.  These spectrotemporal receptive fields  make clear that \nneurons in the primary auditory cortex, as in the primary visual cor(cid:173)\ntex, typically show considerable structure in their feature processing \nproperties, often including multiple excitatory and inhibitory regions \nin their receptive fields.  These neurons can be sensitive to stimulus \nedges in frequency  composition or in time,  and sensitive to stimulus \ntransitions  such as  changes in  frequency.  These  neurons  also  show \nstrong responses and selectivity to continuous frequency  modulated \nstimuli analogous to visual drifting gratings. \n\n1 \n\nIntroduction \n\nIt  is  known  that  auditory  neurons  are  tuned  for  a  number  of independent  feature \nparameters of simple  stimuli including frequency  (Merzenich et  al.,  1973), intensity \n(Sutter and Schreiner, 1995), amplitude modulation (Schreiner and Urbas, 1988), and \n\n\fCharacterizing Auditory Cortical Neurons Using Reverse Correlation \n\n125 \n\nothers. In addition, auditory cortical responses to multiple stimuli can enhance or sup(cid:173)\npress one another in  a  time dependent fashion  (Brosch and Schreiner,  1997;  Phillips \nand Cynader, 1985; Shamma and Symmes,  1985), and auditory cortical neurons can \nbe highly selective for species-specific vocalizations (Wang et al.,  1995; Wollberg and \nNewman,  1972), suggesting complex acoustic processing by these cells.  It is  not  yet \nknown  if these  many  independent selectivities  of auditory  cortical neurons reflect  a \ndiscernible underlying pattern of feature decomposition, as has often been suggested \n(Merzenich et al., 1985; Schreiner and Mendelson, 1990;  Wang et al., 1995).  Further, \nsince sustained firing  rate responses in the auditory cortex to tonal stimuli are typ(cid:173)\nically  much  lower  than visual  responses to drifting bars  (deCharms and  Merzenich, \n1996b), it has been suggested that the preferred type of auditory stimulus may still \nnot  be known  (Nelken et al.,  1994).  We  sought  to develop  an unbiased  method  for \ndetermining the full feature selectivity of auditory cortical neurons, whatever it might \nbe, in frequency  and time based upon reverse correlation. \n\n2  Methods \n\nRecordings  were  made  from  a  chronic  array of up  to 49  individually placed  ultra(cid:173)\nfine  extracellular Iridium  microelectrodes,  placed  in  the primary auditory cortex of \nthe adult  owl  monkey.  The electrodes had tip lengths  of 10-25microns, which yield \nimpedance values of .5-SMOhm and good isolation of signals from individual neurons \nor  clusters of nearby  neurons.  We  electrochemically  activated  these  tips  to add  an \nultramicroscopic coating of Iridium Oxide, which leaves the tip geometry unchanged, \nbut  decreases  the tip impedance  by more than  an  order of magnitude,  resulting  in \nsubstantially  improved  recording  signals.  These  signals  are  filtered  from  .3-8kHz, \nsampled at 20kHz,  digitized, and sorted.  The stimuli used were a variant of random \n\nVlsuII Cortn: Reveree Correlltlon \nU.lng 2\u00b7D VI.nl Pltternl In  Time \n\nAuditory Cortex: Rever.e Correlltlon \nU.lng 1\u00b7D Auditory Pltternl (Chordl) In  Tim. \n\nt -Om.ec \n\nt- 20m.ec \n\nt-40msec \n\nt - 40m.ec \n\nII \n\nSplkeT .. ln. \n\nx \n\nII \n\nx \n! \n\nSpltlotemporal Receptive Field \n\nSpectrotempoul Receptive Field \n\nFigure 1:  Schematic of stimuli used for reverse correlation. \n\nwhite  noise  which was  designed to allow  us to characterize the responses of neurons \nin time and in frequency.  As  shown in figure  1, these stimuli are directly  analogous \nto stimuli that have  been used  previously to characterize the response properties of \nneurons  in  the  primary  visual  cortex  (Jones  and  Palmer,  1987;  Reid  and  Alonso, \n1995;  Reid  et al.,  1991).  In the visual case,  stimuli consist  of spatial checkerboards \nthat span some portion of the two-dimensional visual field  and  change pattern with \na  short  sampling  interval.  In  the  auditory  case,  which  we  have  studied  here,  the \nstimuli  chosen  were  randomly  selected  chords,  which  approximately  evenly  span  a \n\n\f126 \n\nR  C. deChanns and M  M.  Merzenich \n\nportion of the one-dimensional receptor surface of the cochlea.  These stimuli consist \nof combinations  of pure  tones,  all  with  identical  phase  and  all  with  5  msec  cosine(cid:173)\nshaped  ramps  in  amplitude when  they individually  turn on or off.  Each  chord  was \ncreated by  randomly  selecting frequency  values  from  84  possible  values  which  span \n7  octaves from  110Hz  to  14080Hz in  even  semitone steps.  The  density  of tones  in \neach stimulus was 1 tone per octave on average, or 7 tones per chord, but the stimuli \nwere selected stochastically so a given chord could be composed of a variable number \nof tones  of randomly  selected  frequencies.  We  have  used  sampling  rates  of 10-100 \nchords/second,  and the data here  are from  stimuli  with  50  chords/second.  Stimuli \nwith  random,  asynchronous onset times of each tone produce similar results.  These \nstimuli  were  presented  in  the  open sound field  within  an acoustical isolation cham(cid:173)\nber at 44. 1kHz sampling rate directly from  audio compact disk, while the animal sat \npassively  in  the  sound  field  or  actively  performed  an auditory  discrimination  task, \nreceiving  occasional juice rewards.  The complete characterization set lasted for  ten \nminutes, thereby including 30,000 individual chords. \n\nSpike trains were collected from mUltiple sites in the cortex simultaneously during the \npresentation of our characterization stimulus set, and individually reverse correlated \nwith the times of onset  of each of the tonal stimuli.  The reverse correlation method \ncomputes the number of spikes from a neuron that were detected, on average, during \na  given  time  preceding,  during,  or following  a  particular tonal stimulus component \nfrom  our  set  of chords.  These values  are  presented  in  spikes/s for  all  of the tones \nin  the  stimulus  set,  and  for  some  range  of time  shifts.  This  method  is  somewhat \nanalogous  in  intention  to  a  method  developed  earlier  for  deriving  spectrotemporal \nreceptive fields for  auditory midbrain neurons  (Eggermont et al.,  1983), but previous \nmethods have not been effective in the auditory cortex. \n\n3  Results \n\nFigure 2 shows the spectrotemporal responses of neurons from  four  locations in the \nprimary auditory cortex.  In each panel, the time in milliseconds between the onset of \na  particular stimulus component  and a  neuronal spike is  shown  along the horizontal \naxis.  Progressively greater negative time shifts indicate progressively longer latencies \nfrom  the  onset  of a  stimulus  component  until  the  neuronal  spikes.  The  frequency \nof the stimulus  component  is  shown  along  the vertical  axis,  in  octave spacing from \na  110Hz  standard, with  twelve  steps per octave.  The brightness corresponds to the \naverage  rate  of the  neuron,  in  spk/s,  driven  by  a  particular  stimulus  component. \nThe reverse-correlogram is thus presented as a stimulus triggered spike rate average, \nanalogous  to  a  standard  peristimulus  time  histogram  but  reversed  in  time,  and  is \nidentical  to the spectrogram of the estimated  optimal stimulus for  the cell  (a spike \ntriggered stimulus average which would be in units of mean stimulus denSity). \n\nA  minority  of neurons  in  the  primary  auditory  cortex  have  spectrotemporal recep(cid:173)\ntive fields  that show only a single region of increased rate, which  corresponds to the \ntraditional characteristic frequency of the neuron, and no inhibitory region.  We have \nfound  that cells  of this type  (less  than 10%,  not shown)  are  less  common than cells \nwith multimodal  receptive field  structure.  More commonly,  neurons have regions  of \nboth increased and decreased  firing  rate relative to their mean rate within their re(cid:173)\nceptive fields.  For terminological convemence, these will  be referred to as excitatory \nand inhibitory regions,  though these changes in rate are not  diagnostic of an under(cid:173)\nlying  mechanism.  Neurons  with  receptive fields  of this  type can serve  as  detectors \nof stimulus edges in both frequency  space,  and in time.  The neuron shown in figure \n2a has a  receptive field  structure indicative of lateral inhibition in frequency  space. \nThis cell  prefers  a  very  narrow range of frequencies,  and decreases its firing  rate for \nnearby frequencies, giving the characteristic of a sharply-tuned bandpass filter.  This \n\n\fCharacterizing Auditory Cortical Neurons  Using Reverse Correlation \n\n127 \n\na) \n\n3 \n\n2.5 \n\n..  2 \n> .. \ng 1.5 \n\n0.5 \n\n-100 \n\nc) \n3.5 \n3 \n\n2.5 \n\n., \n.. \n2 \n> \ng 1.5 \n\n40 \n\n30 \n\n20 \n\n10 \n\n10 \n5 \n0 \n-5 \n-10 \n\n-50 \n\n0 \n\nmsec \n\n3 \n\n15 \n\n-SO \n\nmsec \n\n0 \n\nd) \n\n3 \n\n2.5 \n\nb) \n\n3.5 \n\n3 \n!: 2.5 \n!!! \n8  2 \n1.5 \n\nN \nJ: \n0 \n)!  10;::: \n'\" \n> \n.8 \n< \n~ \nCD \nU \n0 \n\n2 \n5  .,  1.5 \n\n-100 \n\n-50 \nmsec \n\n0 \n\n~O  -40 \n\n-20 \n\nmsec \n\nFigure 2:  Spectrotemporal receptive fields  of neurons in the primary auditory cortex \nof the awake primate.  These receptive fields  are computed as  described in methods. \nReceptive field structures read from left to right correspond to a preferred stimulus for \nthe neuron, with light shading indicating more probable stimulus components to evoke \na spike, and dark shading indicating less probable components.  Receptive fields  read \nfrom right to left indicate the response of the neuron in time to a particular stimulus \ncomponent.  The colorbars  correspond to the  average firing  rates  of the  neurons  in \nHz at a given time preceding, during, or following  a particular stimulus component. \n\ntype of response is the auditory analog of a visual or tactile edge detector with lateral \ninhibition.  Simple  cells  in  the primary visual cortex typically show  similar patterns \nof center  excitation  along  a  short  linear  segment,  surrounded  by  inhibition  (Jones \nand Palmer, 1987;\u00b7Reid and Alonso,  1995; Reid et al.,  1991).  The neuron shown in \nfigure  2b  shows  a  decrease in firing  rate caused by a  stimulus frequency  which  at a \nlater time causes  an increase in rate.  This  receptive field  structure is  ideally  suited \nto detect stimulus transients; and can be thought of as a detector of temporal edges. \nNeurons in the auditory cortex typically prefer this type of stimulus, which is initially \nsoft  or  silent  and later loud.  This  corresponds to a  neuronal  response  which  shows \nan increase followed  by a decrease in firing rate.  This is again analogous to neuronal \nresponses  in  the  primary  visual  cortex,  which  also  typically  show  a  firing  rate pat(cid:173)\ntern to an optimal  stimulus  of excitation  followed  by  inhibition,  and  preference  for \nstimulus transients such as when a stimulus is  first off and then comes on. \nThe neuron shown in figures  2c shows an example which has complex receptive field \nstructure,  with  multiple regions.  Cells  of this  type would be indicative  of selectiv(cid:173)\nity for  feature  conjunctions  or  quite  complex  stimuli,  perhaps  related  to sounds  in \nthe animal's learned environment.  Cells with complex  receptive field  structures are \ncommon  in the awake auditory cortex, and we  are in the process of quantifying  the \npercentages of cells that fit  within these different  categories. \nNeurons were observed which respond with increased rate to one frequency  range at \none time,  and a  different frequency  range at a later time, indicative of selectivity for \nfrequency modulations(Suga, 1965).  Regions of decreased firing rate can show similar \npatterns.  The neuron  shown in  figure  2d is  an example  of this  type.  This pattern \nis  strongly analogous  to motion energy detectors in the visual system  (Adelson and \nBergen, 1985), which detect stimuli moving in space, and these cells  are selective for \nchanges in frequency. \n\n\f128 \n\nR.  C.  deCharms and M  M.  Merzenich \n\n2 oct/sec  6 oct/sec  10 oct/sec  14 oct/sec  30 oct/sec  100 oct/sec \n\n\u00b72 oct/sec  \u00b76 oct/sec  \u00b710 oct/sec \u00b714 oct/sec \u00b730 oct/sec \u00b7100 oct/sec \n\nFigure 3:  Parametric stimulus set used to explore neuronal responses to continuously \nchanging stimulus frequency.  Images axe spectrograms of stimuli from left to right in \ntime,  and spanning seven octaves of frequency from  bottom to top.  Each stimulus is \none second.  Numbers indicate the sweep rate of the stimuli in octaves per second. \n\nBased  on  the responses shown,  we  wondered  whether  we  could  find  a  more optimal \nclass of stimuli for these neuron, analogous to the use of drifting bars or gratings in the \nprimary visual cortex.  We have created auditory stimuli which correspond exactly to \nthe preferred stimulus computed for  a  paxticulax cell  from  the cell's spectrotemporal \nreceptive field  (manuscript in prepaxation),  and we  have  also  designed a  paxametric \nclass  of stimuli  which  are designed  to be particularly effective for  neurons  selective \nfor  stimuli of changing amplitude or frequency, which are presented here.  The stimuli \nshown  in  figure  3  are  auditory  analogous  of  visual  drifting  grating  stimuli.  The \nstimuli axe shown as spectrograms, where time is along the horizontal axis, frequency \ncontent on an octave scale is along the vertical axis, and brightness corresponds to the \nintensity of the signal.  These stimuli contain frequencies that change in time along an \noctave frequency scale so that they repeatedly pass approximately linearly through a \nneurons receptive field,  just as  a  drifting grating would  pass repeatedly through the \nreceptive field  of a  visual neuron.  These stimuli axe  somewhat analogous to drifting \nripple stimuli which  have recently been  used  by  Kowalski,  et.al.  to characterize the \nlinearity of responses of neurons in the anesthetized ferret auditory cortex (Kowalski \net al.,  1996a; Kowalski et al., 1996b). \n\nNeurons in the auditory cortex typically respond to tonal stimuli with a brisk onset \nresponse at the stimulus transient, but show sustained rates that axe far smaller than \nfound  in  the  visual  or  somatosensory  systems  (deCharms  and  Merzenich,  1996a). \nWe have found  neurons in the awake animal that respond with high  firing  rates and \nsignificant selectivity to the class of moving stimuli shown in figure 3.  An outstanding \nexample of this is shown in figure 4.  The neuron in this example showed a  very high \nsustained firing rate to the optimal drifting stimulus,  as high as 60 Hz\u00b7 for one second. \nThe neuron shown in this example also  showed  considerable selectivity for  stimulus \nvelocity,  as well as some selectivity for  stimulus direction. \n\n4  Conclusions \n\nThese  stimuli  enable  us  to  efficiently  quantify  the  response  characteristics  of neu(cid:173)\nrons  in  the awake  primaxy auditory cortex,  as  well  as producing optimal stimuli for \nparticular neurons.  The data that we  have gathered thus  far  extend our knowledge \nabout  the complex  receptive field  structure of cells  in  the primary auditory cortex, \n\n\fCha racterizing Auditory Cortical Neurons  Using Reverse Correlation \n\n129 \n\n2 oct/sec \n\n6 oct/sec \n\n10 octIsec \n\n14 oct/sec \n\n30 oct/sec \n\n100 oct/sec \n\n-2 oct/sec \n\n-6 oct/sec \n\n-10 oct/sec \n\n-14 oct/sec \n\n-30 oct/sec \n\n-100 oct/sec \n\nFigure  4:  Responses  of a  neuron  in the  primary  auditory  cortex of the  awake  pri(cid:173)\nmate to example stimuli take form  our characterization set,  as  shown in figure  3.  In \neach panel, the average response rate histogram in spikes per second is shown below \nrastergrams showing the individual action potentials elicited on,each of twenty trials. \n\nand show  some  considerable analogy with  neurons  in  the primary visual  cortex.  In \naddition, they indicate that it is possible to drive auditory cortical cells to high rates \nof sustained firing,  as in the visual cortex.  This method will allow a number of future \nquestions to be addressed.  Since we have recorded many neurons simultaneously, we \nare interested in the interactions among large populations of neurons and how these \nrelate to stimuli.  We are also recording responses to these stimuli while monkeys are \nperforming  cognitive  tasks  involving  attention and learning,  and  we  hope  that this \nwill give us insight into the effects on cell selectivity of the context provided by other \nstimuli,  the animal's behavioral state or awareness  of the stimuli,  and  the animal's \nprior learning of stimulus sets. \n\n5  References \n\nAdelson EH,  Bergen JR (1985)  Spatiotemporal energy models for  the perception of \nmotion.  J. Opt.  Soc.  Am.  A,  2,  284-299. \nBrosch M, Schreiner CE (1997)  Time course of forward masking tuning curves in cat \nprimary auditory cortex.  J  Neurophysiol, 77,  923-43. \ndeCharms Re, Merzenich  MM  (1996a)  Primary cortical representation of sounds by \nthe coordination of action-potential timing.  Nature, 381, 610-3. \ndeCharms RC, Merzenich MM  (1996b)  Primary cortical representation of sounds by \nthe coordination of action-potential timing.  Nature, 381, 610-613. \nEggeI1I).ont  JJ,  Aertsen  AM,  Johannesma  PI  (1983)  Quantitative  characterisation \nprocedure for  auditory neurons based  on the spectro-temporal receptive field.  Hear \nRes,  10,  167-90. \nHubel  DH,  Wiesel  TN  (1962)  Receptive  fields,  binocular interaction and  functional \narchtecture in the cat's visual cortex.  J. Physiol., 160,  106-154. \nJones JP, Palmer LA (1987) The two-dimensional spatial structure of simple receptive \n\n\f130 \n\nR.  C.  deCharms and M  M. Merzenich \n\nfields  in cat striate cortex.  J  Neurophysiol, 58,  1187-211. \nKowalski N,  Depireux DA, Shamma SA  (1996a) Analysis of dynamic spectra in ferret \nprimary auditory cortex.  I.  Characteristics of single-unit responses to moving ripple \nspectra.  J  Neurophysiol,  76,  3503-23. \nKowalski  N,  Depireux  DA,  Shamma SA  (1996b)  Analysis of dynamic spectra in fer(cid:173)\nret  primary auditory  cortex.  II.  Prediction of unit  responses  to  arbitrary dynamic \nspectra.  J  Neurophysiol,  76,  3524-34. \nMerzenich MM,  Jenkins WM, Middlebrooks JC (1985)  Observations and hypotheses \non  special  organizational features  of the  central  auditory nervous  system.  In:  Dy(cid:173)\nnamic  Aspects  of Neocortical  Function.  Edited by E.  G.  a.  W.  M.  C.  G.  Edelman. \nNew  York:  Wiley,  pp.  397-423. \nMerzenich  MM,  Knight  PL,  Roth  GL  (1973)  Cochleotopic  organization of primary \nauditory cortex in the cat.  Brain Res,  63,  343-6. \nNelken  I,  Prut  Y,  Vaadia  E,  Abeles  M  (1994)  In search  of  the  best  stimulus:  an \noptimization procedure for  finding  efficient  stimuli in the\u00b7 cat auditory cortex.  Hear \nRes,  72,  237-53. \nPhillips DP, Cynader MS (1985) Some neural mechanisms in the cat's auditory cortex \nunderlying sensitivity to combined tone and wide-spectrum noise stimuli.  Hear Res, \n18,  87-102. \nReid  RC,  Alonso  JM  (1995)  Specificity of monosynaptic  connections from  thalamus \nto visual cortex.  Nature,  378,281-4. \nReid RC, Soodak RE, Shapley RM  (1991)  Directional selectivity and spatiotemporal \nstructure of receptive fields  of simple cells in cat striate cortex.  J  Neurophysiol,  66, \n505-29. \n\nRingach  DL,  Hawken  MJ,  Shapley  R  (1997)  Dynamics  of  orientation  tuning  in \nmacaque primary visual cortex.  Nature, 387,  281-4. \n\nSchreiner  CE,  Mendelson  JR (1990)  Functional topography of cat primary auditory \ncortex:  distribution of integrated excitation.  J  Neurophysiol,  64,  1442-59. \nSchreiner  CE,  Urbas  JV  (1988)  Representation of amplitude in the auditory cortex \nof the cat.  II.  Comparison between cortical fields.  Hear.  Res., 32, 49-64. \n\nShamma  SA,  Symmes  D  (1985)  Patterns  of inhibition  in  auditory  cortical  cells  in \nawake squirrel monkeys.  Hear Res,  19,  1-13. \nSuga N (1965) Responses of cortical auditory neurones to frequency modulated sounds \nin echo-locating bats.  Nature,  206,  890-l. \n\nSutter ML,  Schreiner CE  (1995)  Topography of intensity  tuning in cat primary au(cid:173)\nditory  cortex:  single-neuron versus  multiple-neuron recordings.  J  Neurophysiol,  73, \n190-204. \n\nWang X,  Merzenich MM,  Beitel R,  Schreiner CE (1995)  Representation of a species(cid:173)\nspecific  vocalization in the primary auditory cortex of the common  marmoset:  tem(cid:173)\nporal and spectral characteristics.  J  Neurophysiol,  74,  2685-706. \nWollberg  Z,  Newman  JD  (1972)  Auditory cortex of squirrel monkey:  response pat(cid:173)\nterns of single cells to species-specific vocalizations.  Science,  175,  212-214. \n\n\f\f", "award": [], "sourceid": 1462, "authors": [{"given_name": "R.", "family_name": "DeCharms", "institution": null}, {"given_name": "Michael", "family_name": "Merzenich", "institution": null}]}