To  this  title. Because  I  promised  Garrett  that  I  would come  up  with  something  really  pompous. That's  about  as  pompous  as  it  gets. I'm  going  to  try  to  break  it  down  into English  terms  for  you  starting  now. The  reason  I  study  the  sense  of  taste  is  because  of the  inevitable  sensory  motor  transformation of  taste  information. You  can  take  your  rodent  or whatever  and  give  it  a  visual  stimulus. And  there  are  a  lot  of  visual  stimuli  that, that  rodents  brain  will  respond  to, but  that  are  totally  artificial  and  have no  impact  on  that  rodent  that  are  arbitrary. Same  with  auditory  stimulus. If  you  want  to  go  simple, you  can  study  the  response  to  a  sine  wave. If  you  want  to  get  fancy,  you  can  do  a  fancy  torque. But  the  fact  is  that  the  animal doesn't  give  a  hit  about  those  stimuli. And  while  its  brain  responds, it  doesn't  have  anything  to  do  with  it. That's  never  the  case  with  a  taste. When  you  have  a  and you  or  your  rodent  are  experiencing  that  taste, the  stimulus  is  inside your  body  and  it  is  required  that  you  have  a  reaction, that  you  actually  produce  a  behavioral  response. And  those  behavioral  responses basically  break  down  into  a  binary  decision. On  the  one  hand,  if you  have  something  sweet  put  in  your  mouth, which  is,  this  is  the  work  of  my  very first  graduate  student,  Steve  Grossman. And  so  you  can  see  on  this  female  rodent, you  can  see  the  fore  paws, you  can  see  the  snout  and  the  all  important  mouth. And  what  you  can't  see  is  that  Steve  has  just put  a  little  bit  of  sugar  water  into  her  mouth. And  what  she  does  is  she  produces  the  rat  yum  face. And  this  is  the  tongue  is  sweeping  around  inside the  mouth  to  pull  the  fluid  back in  in  preparation  for  swallowing  it. The  rat  has  made  a  decision  that  it  likes. This  contrast  that  with  what  happens  to the  same  rat  if  what  Steve  puts  in  her  mouth  is  quinine. Now,  quinine  is  an  incredibly  bitter  taste, although  it  is  delightful  with  gin. And  when  you  put  that  in  the  animal's  mouth, you  get  an  entirely different,  different  behavioral  response. The  most  notable  one  is  referred  to  as  a  gape. It  is  a  large,  yawning  face. You'll  see  the  teeth  exposed. What  you  won't  be  able  to  see  is  the  tongue  rolling forward  to  expel  the  offensive  fluid  from  the  mouth. But  it  looks  basically  like  that. There's  another  one  coming  up. This  is  very  slowed  down. This  happens  on  about  four  Hertz  rhythm. And  then  you  could  see  there  was a  little  limb  flail  as  well. That's  the  rat  yuck  face. And  I'm  going  to  focus  most  of  my  talk on  this  because  this  is  what  I  am. I'm  a  behavioral  neuroscientist, and  this  is  the  behavior  I'm  going  to focus  on  for  the  purposes  of  this  talk. Why?  Because  it's  a  decision  that  the  animal  has  made. And  I  didn't  have  to  train  the animal  to  make  this  decision, although  I  can  train  the  animal  to  change  its  decision. Furthermore,  it  happens  on average  about  a  second  after  the  taste  is  in  the  mouth. Which  means  that  there's  plenty  of  time  to analyze  neuroactivity  leading  up  to  that. You're  going  to  see  though  that  there's a  huge  variability  of  that  because this  is  an  animal,  not  a  machine. But  this  is  what  we're  going  to  be studying  and  we're  going  to talk  about  how  you  get  from  taste  to  that. Just  so  you  know,  this  is  not  a  one  species  thing. This  is  a  human  infant  that  just put  a  sour  lemon  into  his  mouth. And  not  unlike  my  rodents, this  human  is  now  making  a  gape. I'm  often  asked  what's the  difference  between  rodents  and  humans? And  the  answer  is  not  very  much. But  one  difference  is  that my  rats  are  smarter  than  this  human. As  soon  as  the  cameras  clicked, I  imagine  that  this  infant  put  the  lemon back  in  his  mouth,  just  so  you  know. It  goes  all  the  way  across  to other  parts  of  the  phylogenetic  tree. Here  is  a  video  from PBS  on  the  Kalahari  Desert  about  toads. And  a  toad  has  a  very  tiny  brain. Although  it  is  a  vertebrate, you  think  of  them  as  very  innocuous  critters, The  worst  thing  they  might  do  is  pee  in  your  hand. But  if  you're  in  the  Kalahari desert  and  you  are  a  beetle, these  are  dangerous,  rapacious  hunters. The  beetles  in  the  Kalahari  have  in  general, evolved  wonderful  defense  mechanisms. The  hero  of  this  video has  developed  this  defense  mechanism. It  tastes  like  crab. Watch  what  happens  when  that  beetle  rolls  along. This  toad  tries  to  swallow  him  and  makes  a  toad  game. Even  any  animal  that  has  a  tongue  does  these  behaviors. These  are  not  only  ubiquitous  to  the  stimulus  presented, they  are  ubiquitous  across  species. That's  what  we  study.  We  are electrophysiologists  in  my  lab, and  I'm  happy  to  have  debates about  electrophysiology  versus  imaging  if  you'd  like. I  think  what  I  will  hopefully prove  to  you  in  the  course  of this  talk  is  that  at  least  for  what  we're  looking  at, imaging  isn't  fast  enough  to  capture  what  happens. We  do  most  of  our  recordings. And  most  of  the  talk  is  going  to focus  on  gustatory  cortex, primary  sensory  cortex,  and  taste. And  we're  going  to  look  at how  this  process  of  getting  from taste  on  the  tongue  to  making a  response  unfolds  through  time. What  we  believe  in  my  lab  and  what  I'm going  to  show  you  in  this  talk  is, for  starters,  time  matters  in  neural  taste  processing. Now,  everything  that  I'm  going  to  tell  you  is stuff  that  you  all  already  know, and  yet  we  behave  as  scientists, as  if  they  aren't  true. So  for  instance,  if  you  have just  a  neuron  and  it  fires  an  action  potential, and  it  fires  an  action  direction  potential, then  there's  a  pause  and  it fires  another  action  potential. That  means  something. But  the  way  we  tend  to  analyze  it  is  we just  say,  ah,  that's  three. How  does  the  thereon  respond  to  this  stimulus? Three,  We're  going  to  tell  you  that that's  the  wrong  way  to  look  at  it. And  that  in  fact,  what  I'm  going to  show  you  specifically  is  that cortical  taste  responses  evolve across  a  second  to  a  second  and  a  half. That  evolution  is  meaningful, you  can  see  the  process occurring  by  paying  attention  to  it. I'm  going  to  then  move  on to  another  thing  that  you  already  know, which  is  neural  ensembles work  together  to  process  tastes. But  I'm  going  to  get  very  specific  about  it. I'm  going  to  show  you  that  single  trial  dynamics, single  trial  responses  to  tastes  are coherent  across  ensembles  and  they  are  non  linear. And  one  of  the  great  things  about  it  that  I'm going  to  show  you  is  being  able  to  get ahold  of  that  single  trial  dynamic  enables  us  to show  that  the  coherent  non  linearities drive  behaviors  in  real  time. Finally,  it's  not  just  that neural  ensembles  work  together  to  process  taste, neural  ensembles  between  brain  regions  work  together. I'm  going  to  show  you  some  new  data  showing  that  amigdla, in  particular  basilateral  amigula, couples  with  cortex  into  a  single  processing  unit, enabling  those  nonlinear  dynamics. That's  the  whole  talk.  I'm  going  to  spend only  a  very  little  time  trying  to  convince  you  of  this, although  to  be  honest  with  you, number  one  is  basically  the  entire  talk  as  well. Let  me  get  that  out  of  the  way  by showing  you  what  a cortical  taste  response  actually  looks  like. This  is  the  over  lane  Perry  stimulus  time  histograms, or  SDH,  meaning  Perry  stimulus. Around  the  time  of  the  delivery  of  the  stimulus. Time  going  through  time  and  histogram  of  firing  rates. We've  taken  the  responses  to  all  of  these  tastes  for this  one  particular  neuron and  just  laid  them  on  top  of  one  another. And  what  you  can  see  from  looking  at this  right  away  is  that  it's  not  as simple  as  the  neuron  turns on  or  the  neuron  doesn't  turn  on. It's  not  even  as  simple  as  the  neuron  turns on  to  this  degree  or  to  this  degree,  or  to  this  degree. In  fact,  way  back  in 2,001.1  of  the  things  I  love  about  presenting  this  slide is  it  allows  me  to  take  my  five  year  post  doc  at Duke  University  and  reduce  it  to  a  single  slide. What  you  find  when  you  do  a  whole  lot  of analysis  on  this  is  that  in  fact, these  responses  basically  have  three  parts, there  are  three  phases  to  them. The  initial  thing  that  happens  when  the  taste  hits the  tongue  is  that  some  neurons  do  nothing  initially. But  those  neurons  that  do  something  fire  a  burst  of action  potentials  that  is completely  non  distinctive  for  taste. We  refer  to  this  as  the  detection  period. We  are  not  making  claims  at  this  moment  about  what the  rat  detects  at  that  point,  but  we, as  scientists  can  look  at  a  neural  ensemble, we  can  see  that  the  taste  has  hit the  tongue  because  of  this  detection  response. And  you  can't  tell  anything else  from  that  detection  response. But  you  can  see  that  over  the  course of  just  the  first  couple  hundred  milliseconds, these  responses  diverge  from  one  another. And  then  you  get  into  a  new  period  that we  refer  to  as  the  identification  period, where  we,  with  even  very  simple  analyses, can  tell  you  what  the  taste  on  the  tongue  is. We  can  look  at  this  activity, sometimes  we  need  a  few  neurons  to  do  it  really  well. We  can  say  which  taste  is  on the  tongue  on  the  basis  of  what the  firing  rates  are  during  that  period. But  then  you  can  see  that  the  responses  change  again. And  now  you  get  into  a  period  where the  responses  are  not  only  taste  specific, they  are  in  fact  palatability  related. Which  is  to  say,  you  can  now  answer the  question  from  looking  at  these  data, what's  the  animal  going  to  do  about  it? Does  the  animal  want  to  eat  it  or  not? With  this  particular  iron,  you  can  see  that the  strongest  response  out  during this  period  is  to  the  bitter  quinine, and  the  weakest  response  is  to  the  yummy  sucrose  and the  different  sodium  chloride stimuli  layout  in  the  right  way, we  sometimes  see  the  opposite  pattern. But  either  way,  a  very  simple  analysis  you  can  do. You  can  go  ahead  and  you  can  correlate the  firing  at  any  particular  moment with  the  palatability  of  those  tastes. We  can  take  the  array  of  firing  rates  at  this  moment  and compare  it  to  the  array  of palatability  of  the  taste which  the  animal  likes  the  most, which  the  animal  likes  the  least. When  you  do  that,  you  can see  that  for  the  first  half,  second, even  a  little  bit  more,  there's  nothing  neurons, the  neuron  is  responding  in  a  taste  specific  way  there. But  there's  no  correlation  to  palatability. But  then  as  you  go  out  past the  second  mark,  this  correlation  increases. That  shows  you  that  it's a  palatability  related  response  for  this  neuron. It  becomes  significant  at  about  0.7  seconds. All  right,  in  case this  doesn't  convince  you  that  what's  going  on  here is  that  these  neurons  are  processing  something  out to  get  to  a  palatability  related  response,  recent  data. I  love  putting  unpublished  data  in  my  talk, and  this  is  unpublished  data  from my  senior  graduate  student,  Kathleen  Megler. The  unpublished  data  is  a  little  rockier, but  basically  what  she  did  was she  noted  the  fact  that  rats, like  humans,  have individual  differences  in  what  they  like. For  instance,  she  gave  four  different  rats  both sucrose  and  sodium  chloride,  both  delicious. And  what  she  saw  is  that  one  rat much  preferred  sucrose  to  sodium  chloride. Another  rat  preferred  sodium  chloride  to  sucrose. Two  others  were  more  hovered  around  liking  them  the  same. She  also  took  rats  and  gave them  both  quinine  and  citric  acid. Bitter  and  sour  both  tastes  the  animal  doesn't  like. Again,  while  it's  pretty  much the  case  that  quinine  is  the  worst, there  are  some  animals  that  dislike  them equally  and  there  are  some  animals  that only  dislike  quinine  a  little  bit  more. All  that  Kathleen  did  is something  that  we  should  have  done  20  years  ago. She  actually  took  these  animals  that  she'd  gotten the  individual  preferences  from and  then  did  the  recordings. What  she  found  is  that  here  is  the  growth of  palatability  coating  across  from, again  from  about  500  milliseconds, peaking  out  a  second. This  is  when  we  use  the  canonical  ranks, meaning  sugar  best  less,  and  bitter  worst. When  she  instead  uses  the  animal's  own  performance, she  gets  better  prediction and  to  take  the  whole  data  set. When  you  get  out  to that  late  Epic  that  codes  palatability, you  get  an  improved  performance from  using  the  animal's  own  data. This  late  period  that  we  get out  to  doesn't  just  code  palability. It  codes  what  is  palatable  for  the  individual  rat. When  does  the  motor  response  starts? I'm  going  to  talk  a  lot  about  that. The  motor  response  typically  starts  around  here, but  I'm  going  to  give  you  much  more  specific  data to  that  in  just  a  little  while,  okay? So  that  is  all  I'm  going  to  do, just  plain  and  simple  on  temporal  coding, although  again,  temporal  coding plays  a  role  in  everything  else. What  I  want  to  focus  on  next  is  another  thing that  in  addition  to  not  using  individual  animals  data, behavioral  data  is  another  mistake  we  were  making. Which  was,  we  were  looking  at  this  and  saying,  oh, isn't  that  a  beautiful  ramp,  isn't  that  beautiful? How  palatability  emerges  across  a  half, a  second,  a  little  bit  more. And  then  we  decided  to  look  at  things  a  different  way, and  we  discovered  that  that  was  entirely  wrong. Different  thing  that  we  did  was  we  decided  to take  ensemble  coding  very  seriously. What  I'm  going  to  show  you  now, PSDH's  are  basically  a  bunch of  trials  of  a  taste all  collapsed  together  for  a  single  neuron. What  we're  going  to  look  at  now,  instead, is  a  single  trial, where  we're  looking  at  an  ensemble  of  neurons. Here's  an  ensemble  of  ten  neurons, and  here  is  their  response  to  a  single  trial  of  quinine. Each  one  of  these  hash  marks  is an  action  potential  fired  by  that  neuron. Now,  I  would  imagine  that  to  most  of  you, this  picture  right  now  looks  just  like  noise. And  I'm  going  to  help  you  with  the  way our  analysis  helped  us  and tell  you  that  right  about  there, which  is  by  the  way,  around  the  middle  of  the  time  period when  supposedly  palatability  is coming  on,  there  is  a  change. And  it  turns  out  that  that  change  is  sudden and  coherent.  It's  so  coherent. Here's  a  neuron  that  basically  only started  firing  immediately  after  that  moment. Here's  another  neuron  that  also started  firing  just  after  that  moment. Here's  another  one.  Here's  another  one. Here's  another  15  of these  ten  neurons  all at  this  time  point  suddenly  increased  their  firing. Suddenly,  by  the  way, if  that's  not  amazing  enough, this  neuron  suddenly  decreased its  firing  at  that  time  point. And  so  did  this  neuron,  the  majority of  these  neurons  that  we  were  recording. This  was  back  when  we  could  only  record  about  a  dozen  at a  time  change  their  firing. Suddenly  at  this  time  point, suddenly  no  ramping, a  very  sudden  change  to  really  analyze  this. Because  I  didn't  just  put  this  line  here  by  eye. We  actually  did  an  analysis that  is  referred  to  as  Hidden  Markov  Modeling. At  this  point,  lots  of  people  know about  Hidden  Markov  modeling. Let  me  tell  you  when  we  first tried  to  publish  on  it,  they  did  not. Hidden  Markov  modeling  is  designed to  find  sequences  of  states  in  your  data. Up  until  very  recently, if  you  talked  on  the  phone  to a  machine  and  it  was  interpreting  your  voice, you  were  probably  talking  to  a hidden  Markov  model  that  was trying  to  say  where the  syllables  and  syllable  boundaries  were. And  what  you  were  saying,  in  the  case  of  our  neurons, what  they  show  us  is  that  essentially  there  is a  state  here  and  a  state  there, the  state  defined  by  the  population  of  firing  rates. In  between  the  two,  there  is  a  period in  which  the  analysis  is  uncertain, but  that's  where  the  transition  happens. In  fact,  in  this  analysis, you  can  basically  pull  out  what  amounts to  the  calculated  confidence  that  say  this  state  exists, which  is  zero  here, rises  really  quickly  and  is  100%  there. So  that's  what  hidden  mark  off  modeling  does  for  us. So  I'm  going  to  use  that  now  for  a  lot  of  stuff. I'm  going  to  actually  talk  to  you about  what  those  states  are. But  first,  just  to  get  it  out  of  the  way, let's  reconcile  the  PSTH  and  HMM  based  analysis of  taste  responses  when  we  take. A  random  animal  and  a  random  ensemble  of  neurons. And  we  take,  say, ten  trials  of  quinine. And  we  throw  up  on  there the  likelihood  of  the  state that  is  dominant  somewhere  out  here. And  this  is  the  state  which  we're  going  to  refer  to again  as  the  putative  palatability  state. Because  it  occurs  out  there. When  we  already  know  that palatabibility  information  is  high, we  just  pick,  I  think  there  are  ten  trials. And  said,  how  does  the  likelihood  of  that  state  evolve? And  what  you  can  see  is  that  on every  single  trial,  it  happens  suddenly. In  every  single  trial,  the  probability of  that  state  emerges,  all  of  a  sudden. But  with  a  wide  variability  between  trials. And  exactly  when  it  happens, If  you  just  do  something simple  like  average  across  those  trials, you  get  that  that  is a  ramp  that  looks  decidedly  like  this  ramp. We  began  thinking  that  perhaps the  idea  that  palatability  coating emerges  across  this  half  second  is an  artifact  of  collapsing  across  trials. In  which  palatability  coating  emerges  suddenly, but  at  different  times  on  each  trial. Now  I'm  going  to  prove  that  to  you. This  is  the  work  of  ex  graduate  student  Brian  Sedaka. What  he  did  is  he  collected a  new  dataset  from  gustatory  cortex. First  thing  he  did  was  he just  ran  it  through  the  typical  analysis. So  that  you  can  see  that  like  I've  shown  you  before, across  the  population  of  neurons, you  go  from  being  a  low  correlation  with palatability  at  around  0.5  seconds, a  time  at  which  the  neurons  are  coding tastes  and  it  gets  high  by  a  little  after  1  second. So  this  looks  exactly  like  the data  we've  shown  you  before. And  I  will  say  I've  been  doing  this  now  for  22  years. I  still  breathe  a  little  sigh  of  relief every  time  a  dataset  comes  in  and  it  looks  like  this. And  so  then  Brian  decided to  do  just  very  simple  analysis. He  took  every  trial  that  was  used  to  make  this. And  what  he  did  was  he  did hidden  mark  off  modeling  on  it. This  is  a  schematic  of  the  Hidden  Markoff  model  solution. Each  one  of  these  lines  is  the  probability  of an  individual  state  in  the  cross  time. Basically,  this  is  based  on  the  ensemble  of  data. Then  he  made  a  simple  assumption. The  time  period  out  here  is  where  palatability  is  high. And  I'm  going  to  call  the  state that  is  most  likely  now  in  that  chin  period, in  this  case,  the  one  in  blue. That  is  my  putative  palatability  state. Everyone's  still  with  me.  And  then  he  said, okay,  let's  just  do  something  simple. Let's  find  the  time  point  at  which that  state  in  each  trial  becomes  maximal. I  should  tell  you  that  everything that  he  does  did  this  way. We  also  try  it  with  the  75th  percent, with  the  0.75  and  the  0.8  It  always  comes  out  the  same. Then  he  did  a  very  simple  manipulation. He  just  took  these  trials  and  he  nudged  them  over. Instead  of  being  oriented on  the  zero  time  point  when  the  taste  was  delivered, they're  now  oriented  on  the  time  point  that that  state  became  maximal. And  so,  same  dataset, the  average  movement  of  these  firing  rates was  about  150  milliseconds. Although  some  of  them  have  neurons  had  to  move, some  trials  had  to  move  a  lot. But  now,  when  we  do  the  exact  same  analysis, we've  got  this  time  point  centered  in  the  analysis. The  time  point  at  which  the  state  comes  on. And  when  you  do  that,  suddenly your  transition  to  palatability  is  super  fast. This  transition  happens  fast  at  zero  time  point. We  should  tell  you  that  there's a  lot  of  stuff  that  goes  in  here. And  in  old  talk,  I  spent  a  good  deal  of time  pulling  apart  this  result. I'm  not  going  to  do  that  now  because  I  have a  lot  of  new  data  that  I  want  to  get  you  to. I'm  going  to  ask  you  for  now, take  on  faith  that we've  done  a  ****  ton  of  analysis  on  this. And  it's  not  just  that  that  transition  is  fast, that  transition  is  effectively  instantaneous. We  cannot  figure  out  a  way  to  make  it  faster. In  fact,  we  did  dummy  data, simulated  firing,  that  was  made  from  our  original  data. And  the  only  difference  is  we  actually made  sure  that  the  transitions  were  instantaneous. We  got  exactly  the  same  curve because  you  do  get  some  smearing from  the  way  you  do  analysis. As  far  as  we  can  tell,  halidability  comes  on  in  a  moment. The  great  thing  about  that  for  us  is that  now  we  have  a  single  trial  measure. We  can  use  HMM  on  single  trials  and  we  can tell  you  when  that  transition  occurs  on  single  trials. Which  means  we  can  get  to  the  question  that  was  asked, which  is  how  does  that  relate  to  individual  behaviors? Here's  how  it  relates.  We  basically  did. I'm  going  to  reduce  this  part  of the  data  to  the  single  most  important  plot, which  is  a  scatter  plot  on which  one  axis  is  the  latency  to gaping  on  the  other  axis  is the  time  of  that  transition to  the  palatability  related  state. Yes.  Yes  sir.  I'm  happy  to  talk  about  this  at  length. We,  at  this  point,  are  doing  this  with about  75  neurons  per  ensemble. We've  checked,  yeah,  but  we've  tested  this. We  can  do  this  with  six  neurons. We  can  do  the  same  thing  with  six  neurons. Again,  I'm  happy  to  debate  about this  when  you  hear  about  how  important  it  is. We've  got  to  get  more  neurons,  and more  neurons,  and  more  neurons. This  activity,  this  dynamic that  I'm  talking  about,  is  not  sparse. This  is  what  the  brain  is  doing. We  would  argue  you  don't need  a  bunch  of  neurons  to  see  it. We're  recording  from  a  bunch  of  neurons now  because  everyone's  really  impressed. But  you  can  see  exactly  this  pattern  with  six  neurons, and  it  doesn't  change  when  you  have  75.  What  is  that? Here  we  have  the  diagonal line  in  the  middle  of  this  plot, which  means  that  anything  that  falls  on  this  line, the  behavior  occurs  at exactly  the  time  of  neural  transition. I  should  mention  that  we're  using EMG  electromyography  to  report  the  behavior  time, because  with  a  freely  moving  rat, it's  hard  to  actually  tell  when  the  gap  starts. Any  dots  that  occur  above the  diagonal  line  means  the  neural  transition  happened. Any  dots  that  happen  below the  line  means  that  behavior  happen  first. And  here's  a  full  dataset. And  so  what  you  can  see  is  that  with the  exception  of  a  couple  of  straight  trials, which  we  probably  just  got something  wrong  in  our  analysis, the  vast  majority  of  them  show  that  not  on  that. First  of  all,  there  is  a  huge  variability in  the  behavioral  latency  and a  huge  variability  of  the  time  of neural  transition  that  the  two  are  linked. And  that  in  fact,  reliably, the  neural  transition  happens a  couple  hundred  milliseconds  before  the  animal  gapes. And  I  can  show  you  I  took  out  of the  talk  again  for  revity's  sake, just  individual  trial  data. It  happens  again  and  again. No  matter  how  early  that  transition  is, a  few  hundred  milliseconds later  than  that,  the  animal  starts  gaping. Which  means  that  with  these  data  and  this  analysis, we  can  not  only  tell  you  what behavior  the  animal  is  going  to  do, we  can  tell  you  when  the  animal's  going  to  do  it. Okay,  but  that's  all  phenomenology. And  so  we  moved  from  there  on  to  perturbation  studies. Being  psychologists,  we  started  with a  behavioral  perturbation  study  in  which my  post  doc  Jen  Lee  manipulated  the  behavior  by  Qing. What  she  did  was  very  simple. She  cut  down  the  dataset  she  used  so  that  she  was  just using  sucrose  and  quinine and  just  strong  versions  and  dilute  versions. Really,  none  of  that  matters. All  you  need  to  know  is  that the  only  thing  that  caused  gap  is  a  very  strong  quinine. What  she  did  is  every  time before  she  delivered  strong  quinine, she  played  a  tone  across  the  session. The  rats  learned  when  strong  klinine was  coming  and  that  caused  them  to  change  their  behavior. From  the  first  half  to  the  second  half  of  the  session, the  latency  to  gaping  reduced  by  about  150  milliseconds. Just  so  you  know, when  she  didn't  use  the  cue, there  was  no  change  from the  first  half  to  the  second  half  of  the  session. Then  she  said  quite  simply, okay,  what  happens  to  the  neuroactivity? What  happens  to  the  timing  of  this  transition? And  it  was  almost  magical, basically  a  cough  from  the  first, second  half  of  the  session. The  latency  to  that  transition also  moved  by  150  milliseconds. When  you  perturb  the  behavior  itself, the  neural  activity  is  perturbed  with  it. But  of  course,  what  everyone  also  wants  to know  is  what  happens  if  you  perturb  the  neural  activity. My  graduate  student,  Norendramucerg, decided  to  look  at  this  and  he looked  at  it  with  optogenetics. And  this  is  just  a  bit  of  a  brain  tissue. This  is  showing  where the  injection  of  the  virus  with  ArchT, which  is  an  optical  silencer,  and  GFP, which  is  of  course  fluorescence, were  put  into  gustatory  cortex. And  then  he  let  the  animal  recover  stuck  in a  fiberoptic  rather  than  doing  the  standard  thing. Since  we  seem  to  be  constitutionally  incapable  of  doing, we  decided  to  do  it  a  little  differently. He  delivered  four  different  trial  types. In  one  of  them,  there  was  no  laser  at  all. In  another  one,  the  laser  was  on  for half  a  second  at  the  beginning  of  the  trial. Then  there  was  some  trials  in which  the  laser  was  turned  on for  half  a  second  in  the  middle  of  the  trial, and  some  trials  in  which  the  laser  was turned  on  for  half  a  second  late  in  the  trial. This,  by  the  way,  is  a  good  way  to  avoid  needing an  empty  virus  control  because  we're  just comparing  different  kinds  of different  kinds  of  laser  delivery and  not  laser  to  no  laser. There's  a  lot  of  really  cool  data  in  this  study, but  I  really  just  want  to  focus  in  on  this  middle  one. Because  the  range  of  transition  times  and  gaping  times, if  you  recall,  is  somewhere  around  there. That's  the  huge  variability. And  when  the  animal  gapes, looking  at  narendras  data, with  just  looking  at the  time  of  GC  perturbation  being  from 7.7  to  1.2  again  half  a  second. This  is  a  really  short  perturbation  in  his  data  set. The  average  time  of  gaping  was  about  900  milliseconds. But  there  was,  of  course,  a  big  variability  in  that which  left  us  with  what  we  thought  was a  problem  and  turned  out  to  be  a  wonderful  thing. Which  is  that  sometimes  on  some  trials, remember  all  of  these  trials, as  far  as  Narendra  and  I  were  concerned,  were  the  same. We  turned  on  the  laser,  we gave  the  animal  quinine  at  zero, turned  on  the  laser  at  700  milliseconds, turned  off  the  laser  at  1,200 milliseconds  for  every  one  of  these  trials. I'm  going  to  talk  about  on  some  of  the  trials, the  neural  transition  that  I'm talking  about  hadn't  happened  yet  at  that  time  point. Because  there's  this  big  variability we  were  doing  recording and  narendra  could  identify  after  the  fact, the  trials  in  which  the  animal had  not  made  that  transition, which  GC  was  not  yet  doing, palability  related  firing  on  those  trials gaping  was  massively  delayed  by  over  half  a  second. Basically,  the  animal,  for  all  intents  and  purposes, wasn't  able  to  gape  until  we  turned  the  laser  off. Then  there  were  a  few  trials  in  which  we  missed  it, in  which  the  animal  had  the  GC  had  already made  the  neural  transition  into  palatability  coating. And  it  turns  out  that  on  those  trials, our  laser  had  zero  effect. This  looks  like  the  laser  had  a  negative  effect, but  that's  just  because  we're  only looking  at  the  most  early  trials  here. Basically,  this  had  no  effect  and  we did  our  best  to  analyze  this  as  carefully  as  we  could. If  the  laser  turned  on 25  milliseconds  after  that  transition,  we  had  no  effect. What  that  means,  first  of  all, is  an  aside,  but  I'd  like  to  bring  it  up. What  happens  when  you  inhibit  GC? You  might  ask,  the  answer  to that  question  is,  well,  it  depends. In  fact,  we  write, and  in  fact  we  did  write  in  an  opinion  piece, first  authored  by  my  post  doc  Jean  Len, that  the  idea  that  you  figure  out  what  neurons do  by  knocking  them  out  is  simplistic  and  wrong. That  what  neurons  do  is  determined by  exactly  when  you  ask them  and  not  just  when  you  ask  them, but  what  they've  been  doing  just  before. But  for  the  purposes  of  this  talk, another  point  that  carries  us  to the  next  level  is  that  gaping  did  eventually  happen. We  were  hoping  that  by  doing  this  we'd  basically just  blow  the  animals  trials or  smithereens  and  he  wouldn't  gape. She  wouldn't.  When  we turned  the  laser  on,  that  did  not  happen. We  delayed  it  massively, but  once  we  turned  the  laser  off, the  animal  gaped,  which  could  mean  a  lot  of  things, but  it  probably  means  that  something  else  is  involved. Knocking  out  GC  wasn't  enough  because  there's a  system  that's  actually  moving  you  toward  gaping. Once  you  let  GC  go, that  system  is  still  able  to  communicate  with  it  and the  fact  of  the  rest  of  the  system drives  the  behavior  home, which  suggests  that  GC  is  maybe  the  output  station. That  signal,  that  is  this  neural  transition  is  then  that, okay,  we've  reached  the  decision  to  gap, but  that  is  not  making  that  decision  alone. Yes,  the  stimulus  is  still  in  the  mouth, although  the  animal  is  engaged  in  massive  removal of  it  and  it's  also  salivating. We  find  that  while  you can  still  detect  that the  animal  has  not  gotten  back  to  ground  zero, certainly  you're  not  getting  much  in  the  way  of  behavior. Neural  activity  is  receding down  to  nothing  by  about  a  second  and  a  half. But  if  GC  isn't  doing  this  alone,  then  what's  going  on? We  focused  on  the  basilateral  amigula, although  at  this  point  we've  got about  four  different  projects going  on  the  lab  having  to  do  with four  different  other  brain  areas. But  why  do  we  focus  first  on  the  basilateral  amigula? Well,  for  starters,  it's  reciprocally  connected  to  GC. In  addition,  base  lateral  amigo  neurons are  responsive  to  tastes  and not  just  responsive  to  taste, they're  responsive  to  taste  on  the  same  clock  as  GC. When  you  look  at  BLA  neurons  of  which  this  is  one. And  this  is,  again,  the  formatting is  a  little  different  because  this  is the  work  of  my  post  doc  Abu  Mahmoud  recently. But  we  did  this  way  back  in  2009  as  well. You  can  see  that  there  is  an  early  epic, which  in  this  case  of  this  neuron,  there's  no  firing. And  then  you  get  a  middle  epic. And  then  you  get  a  late  epic. And  you  can  see  the  responses  change going  into  each  of  those  epics. Remember  that  in  GC,  that middle  epic  is  the  identity  epic, and  that  late  epic  is  the  palatability  epic. It  turns  out  it's  a  little  different. In  BLA,  it's  the  same  clock, but  in  the  amigula, the  ability  to  tell  what  the  taste  is comes  on  with  the  identity  epic,  but  then  goes  off  again. Basically,  it's  only  briefly  taste  responsive. Furthermore,  that  brief  taste  responsivity is  intrinsically  palatability  related. When  you  do  palatability  correlation, you  see  that  this  middle  epic  in basilateral  emigla  neurons  is already  giving  us  palatability. Is  already  giving  palatability related  firing.  I  have  to  be  careful  with  that. My  closest  collaborator  is a  mathematician  and  he  always  tells me  information  is  one  thing,  relatedness  is  another. What  you've  got  here  is  you've  got palatability  related  firing  right  off  the  bat, right  in  that  middle  epic.  In  addition. Finally,  and  I  don't  have  time  to  go  into all  of  the  work  that's  been done  by  a  lot  of  my  colleagues. The  base  lateral  is  known  to  be involved  in  driving  emotion  laden  decisions. And  learning  the  decision  that  you  hate, what's  in  your  mouth  is intrinsically  an  emotional  behavior. And  the  Amigla  has  been  shown  to  be  involved  in  those. That's  why  we  decided  to  look  at  BLA. My  post  doc  Jan  Yulin  decided  to  ask  to  the  degree  to which  BLA  is  important for  the  dynamics  we  see  and  what  we  see  in  GC. What  he  did  was  he  infected  BLA  with  RT. The  BLA  neurons  produce  that  channel, so  did  their  axons. And  then  what  Jean  Yu  did  was  he  put his  optotrodes  electrodes  plus  fiber  optics  just  into  GC. The  idea  was  that  with  that  optical  stimulation, he  could  perturb  activity  not  in BLA  by  itself  and  not  in  GC  which  is  not  infected, but  within  the  axons  that  are  projecting  to  GC. I'll  admit  didn't  think  this  was  going  to  work. We've  done  a  lot  of  controls, can  tell  you  that  not  only  does  it  work, I'm  going  to  show  you  the  impact  of  it, but  we've  also  done  the  controls  to show  that  it's  not  actually messing  very  much  with  activity back  in  BLA  which  we  were  worried  about. The  important  thing  I'm  going  to  show you  is  that  when  you  do  that, you  mess  with  palatability. And  GC,  about  50%  of  the  neurons  in  GC  that  had palatability  related  firing  lose it  when  you  knock  out  that  input  pathway. You  come  back  to  it  in  a  second that  some  neurons  that  didn't  have  it  develop  it. We're  going  to  come  back  to  that  in  just  a  second. Point  out  one  of the  reasons  we're  pleased  about  this  is  that  it  actually replicates  earlier  work  done  with actual  inactivation  of  the  Migla  itself. These  data  look  remarkably  like  that, but  we  weren't  super  impressed  because  again, it's  not  just  that  there's  some  palatability information  that  remains  in  GC. There  are  some  GC  neurons  that  when  you  block  BLA  input, get  palatability  related  information. We  don't  really  have  a  good  answer  yet  for exactly  where  that  information is  coming  from  and  why  it's  there. But  I  can  tell  you  that  it  made  us  say,  okay, what  really  is  BLA  doing  then if  it's  not  just blocking  palatability  from  getting  into  cortex? The  answer  is  actually  a  little  more  interesting. It  has  to  do  with  the  dynamics. When  Janu,  in  his  own  dataset, looked  at  the  onset of  palatability  response  in  relation  to  that  transition, he  saw  what  we've  seen  before, which  is  a  very  sharp  rise  again as  you  go  into  that  palatability  related  ensemble  state, you  get  palatability  information. And  then  he's  asked  what  happens  when  the  laser  is  on. The  answer  is  that,  in  fact, in  looking  at  the  entire  dataset, the  slope  of  that  rise  of palatability  related  information  is  cut more  than  a  half  by  the  loss  of  BLA  LA. Yeah,  it's  involved  in  having palatability  related  information  cortex, but  what  it's  more  robustly  involved in  is  making  those  dynamics  happen. Without  the  BLA,  you  don't  have those  sharp  dynamics  which  we've  related  to  the  behavior. You  might  ask,  what  happens  to the  prediction  of  behavior?  It  goes  to  hell. Now  does  it  go  to  hell?  The  behavior  goes  to  hell. Or  our  ability  to use  HMM  goes  to  hell.  We're  not  entirely  sure. But  my  other  postdoc, one  of  my  other  postdocs  abu decided,  well,  if  that's  the  case, then  it  becomes  reasonable  to  ask  if  you  need  BLA  to  make these  dynamics  happen  and  we  already  know that  BLA  works  on  the  same  clock. Is  it  even  reasonable  to  be  talking about  BLA  and  GC  separately? Are  they  really  part  of  the  same  unit? All  he  did  was  he  put  electrode  bundles down  into  both  structures  simultaneously. And  then  he  just  did  another  recording  session. What  we  did  then  is  we  looked  at HMM  solutions  to  the  firing  in  both  regions. What  we  saw  is  that  in  general, we  saw  that  when the  GC  neurons  transitioned  into  the  late  state, so  to  the  BLA  neurons, when  on  a  different  trial,  that  transition happened  earlier,  it  happened  earlier. For  both,  When  it  happened later,  it  happened  later,  for  both, we  went  and  did  the  full  analysis  on  the  entire  dataset. Thinking  that  transition  one  is  going from  the  detection  to  the  identity  phase  in  GC. Transition  two  is  going  from identity  palability  transition  three, which  we  generally  don't  even. Do  much  with  yet  is coming  out  of  the  palatability  related  state. And  what  he  found  is  that  there  is  a  synchronization, a  correlation  between  when  this  happens, but  only  for  the  transitions  in  and  out  of  palatability. What  he  found  is  that  GC  and  BLA  are  in  fact coupling  for  purposes  of  doing  that,  palatability  coding. This  is  a  paper  that  got  published  just this  past  year  in  work  that  he's  still  doing  now. He's  gone  beyond  this  because  of  course, this  correlation  doesn't  necessarily mean  exactly  at  the  same  time, it  could  mean  one  leading  the  other. In  fact,  when  he  does  an  analysis  of exactly  the  relationship  between transitions  in  the  two  in  a  pot  like  this, any  example  that's  out  here means  the  BLA  transitions  first. On  this  side  means  the  GC  transitions first  for  that  first  transition, which  remember,  is  distinctly  uncoupled,  BLA  comes  first. What  we  got  at  the  beginning  of this  response  is  GC  and  BLA  are  doing  their  own  thing. They're  both  getting  taste  related  information. It's  getting  to  BLA  first. But  then  look  at  the  next  two  transitions. Somehow  in  between  transition  one  and  transition  two, the  two  not  only  become  coupled in  that  there's  correlation, they  become  coupled  in  that  they're synchronous  in  that  by  the  time  you  get  out  to  here, now  BLA  and  GC  have effectively  become  a  single  processing  unit. They  are  now  working  together. They  click  into  alignment  mind. I'm  almost  done  now. This  is,  all  the  stuff  that  I'm showing  you  now  is  also  unpublished. I'm  happy  to  talk  about  it. Be  lying  if  I  said  I  understood  at  all. But  that's  where  we  are  now. With  one  exception  from  this. Abou  said,  okay. What  I  think  is  going  on  Don, is  that  early  in  this  response  before  they  synchronize, remember  that  you've  got  palatability already  happening  in  Amigla. I  think  that  the  Amigla  is pushing  the  system  into  a  coupled  mode. And  what's  going  on  is  that information  from  the  Amigla  is  what's  being  used to  drive  the  system into  a  particular  coherent  global  state. He  went  ahead  and  did some  analysis  using  Granger  causality, which  for  those  of  you  who  don't  know, is  a  rough  and  ready  first  pass  at  an  analysis  that  says, not  just  is  there  coherence  between  these  two  regions, but  is  there  directional  coherence. Is  it  fair  to  say  that  one  region is  influencing  the  other? And  what  he  found  is  when  he  looked at  the  BLA  to  GC  connection, with  the  influence  going  from  strong  to  weak. What  he  found  was  that  the  only  notable  coupling that  happened  after  the  taste  hits the  tongue  is  in  a  very  low  frequency  range. And  in  fact,  it  happens  exactly when  we're  going  between transition  one  and  transition  two, that  during  the  period  when BLA  is  coding  in  a  palatability  related  fashion, before  GC  is  BLA, information  flow  is  going  from  the  BLA  to  GC. In  the  meantime,  by  the  way,  this is  not  something  that  is  reciprocal. In  fact,  C,  the  influence  of  GC, BL  is  T  is  quite  a  bit  stronger  and  it  is  fairly  stable. If  anything,  it  looks  like  there  might  be a  little  bit  of  a  decrease  during that  period,  then  reinvigoration. Although  again,  these  are  very  preliminary  data. Obviously,  we  have  a  lot  of  analysis  to  do before  we  can  convince  ourselves. But  it  seems  like  there  may  be  in  fact  a  situation  where it  seems  like  the  BLA  to  GC  flow  is  driving  coupling. At  which  point  the  system  emits a  signal  and  the  GC  then  communicates  out  and  says, okay,  it's  time  to  gap. That's  all  I've  got  for  you.  And  data  again, happy  to  talk  about  this,  particularly because  we  haven't  finished  it  yet. Yes,  I  think  this  preliminary  analysis has  ruled  out  a  third  area  that  projects  to  vote. It  clearly  has  it  by  the  very  nature  has  not. We  are  now  doing  experiments  in  which  we  knock out  other  regions  at different  time  points  and  that  will  give  the  proof. We  think  it's  fairly  safe  to  say  that  this  is  probably not  from  a  common  source,  but  we  don't  know  for  sure. I  obiligatory  summary slides  to  tell  you  what  I've  told  you. Now  that  I've  told  you  GC  taste  responses reflect  taste  processing  across  successive  epics. Which  means  those  of  you  who  have  been  wowed  by simple  stories  of  just  neurons. Sugar  neuron  neuron,  bitter  neuron.  It's  just  wrong. In  fact,  ensemble  analysis  of  single  trials  reveals  that the  onset  of  palatability  coating is  a  sudden  ensemble  transition. And  that  this  transition  in  turn  drives  behavior, probably  by  modulating  a  brain  stem,  CPG. Another  set  of  experiments  that we're  doing  are  looking  into  that. Right  now,  I  would  like  to  do  a  shout  out  in  particular, there's  a  lot  of  research  from colleagues  that  I  haven't  really been  able  to  unpack  for  you. But  I  would  like  to  point  out  that  one of  my  mentors  is  Eve  Mart. And  she  has  spent  her  illustrious  career  which  has  most recently  resulted  in  her  getting a  big  medal  from  the  President. Looking  at  a  central  pattern  generator in  the  lobster  and  crab. And  most  of  the  work  that  she's done  on  that  has  been  on  the  fact  that  there  is a  whole  bunch  of  top  down  modulation  which helps  to  determine  exactly  what  that  CPG  does. We  are  actually  thinking  of the  four  brain  in  those  terms  that  what  it's  there  for. And  we'd  like  to  help  Eve  rescue  modulation from  the  second  class  status  to  which  it  has been  resigned  that  the  four  brain is  essentially  modulating  this  brain  stem  CGG, in  a  way  that  allows  the  four  brain  to  say, okay,  you  could  go  into  gaping  rhythm. Now  finally, the  mechanism  of  construction  of  this  transition involves  coupling  between  at  least  the  GC  and  BLA  into a  functional  unit  where  it  basically  where it  transition  synchronously  in and  out  of  the  important  state. That's  where  we've  gotten  into  it  now. And  so  I  send  you  greetings  from the  Behavior  Learning  and  Electrophysiology of  Chemo  Sensation  Lab at  Brandes  University. This  is,  I  think,  a  complete  list  of  my  people, I'm  not  entirely  sure, but  every  single  person  who  is  underlined here  contributed  a  first  author  paper to  the  talk  I  just  gave  you. I  should  also  mention  a  shout out  to  my  most  frequent  of  many  collaborators. Brandeis  is  a  very  incestuous  place, we're  all  working  in  one  other  lab, but  most  frequent  collaborator  is  Paul  Miller. Paul  is  a  recovering  physicist  and  a  brilliant  modeler, and  I  can  barely  add, he  is  the  person  who  makes  sure  that  all  of our  computational  work  can  be  trusted. I  would  also  like  to  thank, of  course,  funding  from  the  NIH, the  NSF,  and  the  Schwartz Foundation  for  Computational  Neuroscience. Thank  you  for  listening.  I'm  happy  to  take  any  questions. There's  a  good  time  here  for questions  discussion.  Thank  you  so  much. So  your  focus  on  whether  it's sudden  or  gradual  just  requires  as  well,  shall. Yeah,  I  want  to  there, of  course  they  want  to  say  whether  it's an  accumulation  of  what  does  it  mean. Right.  That's  a  great  question. The  paper  at  all. 16  fairly  simple  paper. You  oh,  I'm  sorry. The  question  was  that  it  was, it  was  noticed  that  when  we talk  about  sudden  transitions, we  are  sort  of  evoke. Mike  Shadlin  is  suddenly  in  the  room  who talked  about  the  neural  basis of  decision  making  and  primates, And  has  talked  for  years  about a  biased  random  walk  and  evidence  accumulation, which  necessarily  leads  to  a  ramping  of  activity. And  he's  talked  about  that  at  length. And  so  the  2016  paper that  I  was  referring  to  at  that  point took  a  year  and  a  half extra  to  get  published  than  it  was  supposed  to. I  should,  as  I've  mentioned  before, I  don't  send  papers  to  neuron  because I  got  no  interest  in  a  two  year  turnaround. That's  one  of  the  reasons.  Usually  my  papers between  my  initial  submission  and the  publication  of  that  paper  is  six  to  eight  months. This  took  a  long  time  and  the  reason  was  because  we  ran afoul  of  the  Shadlin  people  in that  paper  Stuff  that  I'm  not  talking  to you  about  to  you  this  time  is  we  went to  the  wall  and  beyond  to  show  that you  cannot  explain  this  with  accumulation  of  information. That  this  is  an intrinsically  attractor  network  model  that  explains  this. We  also  have  a  Paul  and  I  wrote a  50  page  computational  paper. If  you're  having  trouble  sleeping, I'll  send  you  a  copy  which  we  show  that  the  Shad  model doesn't  always  give  you the  best  result  in  the  attractor  model  that  we  use. That's  noise  driven  can  be  more  optimal. But  I  will  say  that  when  I  gave  this  talk  at  Columbia, Mike  and  I  spent  about  an  hour  and  a half  in  our  half  hour  meeting talking  about  different  behaviors that  might  call  for  different  types  of  decisions. I'm  pretty  comfortable  with  that  answer. That  in  fact,  the  kind  of  task  that Shadlin  does  is  a  task  that benefits  from  accumulation  of  information. Not  so  much  this  one.  Having  said  that, it's  worth  noting  that  I  also  gave  this  talk. Way  before  we  got  it  published  at a  computational  meeting  where Jonathan  Pillow  was  hanging  out. And  Pillow  went  back  to  his  collaborator  Huck, who  used  to  work  for  Shadlin. And  the  two  of  them  took  apart Shadlin's  own  data  and  they concluded  that  you  don't see  actually  a  random  walk  there  either. And  so  we  think  that  at  least  in  our  system, it's  an  entirely  different  network. Yes. In  fact,  we  can  only  say  that  recently because  here  is  a  place  where  more  neurons  mattered. Michel  is  a  harder  hit  for  those  of you  who  are  forced  to  try  to  do  it. And  we've  only  recently  been  able  to  get enough  neurons  to  say  that,  yes,  the  same  is  true  there. Yes.  You  talk  a  little  bit  about  how  many  neurons  do  you need  to  make  this  estimate  depends  on how  many  neurons  are  prensa  together,  right? I  think  you  just  were  saying  that  in the  amigula's  not  the  same  scales, you  need  more  neurotme  problem was  that  we  were  generally  getting  fewer  neurons, always  from  amigulaan  cortex. And  it  was  I'm  a  technophobe  personally, and  it  was  only  with  my  graduate  students forcing  me  into  the  modern  age  that  we  got to  a  point  where  we  were  actually  getting a  substantial  number  of  neuron simultaneously  from  the  migula. It  still  takes  more because  one  of  the  things  that  matters  is  firing  rates. The  amigla  responses  neurons tend  to  be  quieter  than  cortical  neurons, at  least  to  L  neurons  in  the  awake  animal. The  fire  rates  that  we  were  working  with, the  fire  rate  transitions  were  smaller  transitions. It  took  more  neurons  there  for  that  reason  as  well. Yes, it's  a  mix. We  actually,  with  our  technique, we  cannot  definitively  show what  neuron  types  we're  recording  from. We  can  only  do  putative, We  could  go  in  and  infect  interneurons  and  do  that. And  to  be  honest  with  you,  we  just haven't  felt  it  was  urgent  enough  to  do  that. But  what  we  have  done, putameal  cells  and  interneuron, which  you  can  pretty  well  identify  on  the  basis  of basal  firing  rates  and  the  width  of  spikes. And  what  we  find  is  that  we  do  have neurons  in  which  the  transition is  down  and  neurons  in  which  the  transition  is  up. And  that  that  does  not  obey  neuron  identification  at  all. That  we  get  both  inner  neurons going  up  and  parameter  cells  going  up. The  reason  I  was  asking  is  it  get neurons  all  of  a  sudden  suddenly  transition from  silence  to  firing, wondering  if  it's  something  local  where  there's like  urscitation  or  release  something. St,  Yeah,  we  don't  know. The  honest  truth  is  just  simply  that  we  haven't  been doing  experiments  that  are designed  to  get  at  that  level  of  analysis. I  can  tell  you  that  we  see  them  a  lot  and these  sudden  transitions  into high  firing  rates  are  the  most  common  thing  we  see. Wouldn't  surprise  me  at  all  if  it's  happening  at  the, because  of  the  removal  of  inhibitory  neurons  don't  know. Have  to  do  a  kind  of  experiment  that  we  generally don't  do  in  our  lab  to  really  know  the  answer.  Yes. Ranger  causality,  preliminary  data. Slow  what  you  call  slower  oscillations. Those  to  me  looked  kind  of  theta  is  O  is  theta. Seems  playing  a  role  here. I  mean,  obviously  it's  playing  a  role  in a  faction  and  the  sampling  of  smells  I  don't  know  about. Yeah,  there's  a  lot  of  rhythms  going  on  up  there. It's  a  little  tough  to  pull  them apart  because  some  of  them  are  driven  by  behavior. But  we  definitely  have  seen  high  theta  in  our  activity. And  these  data  just  came  in, haven't  really  taken  that  apart  at  all. But  certainly  it  looks  like  what  we're seeing  from  BLA  to  GC  is  either  high  theta  or  low  alpha. Yes.  Okay.  I  was  interested  in the  variability  in  the  time like  onset  of  the  palatability  phase and  also  the  data  you  were  showing  about  the  animal  to animal  variability  and  whether or  not  they  even  gave  or  not. And  I  was  curious  whether  or  not  there's anything  in  the  early  part,  you  know, the  earlier  kind  of  two  phases  that  may  be predictive  of  either  the  onset  of timing  of  that  palatability  phase or  the  likelihood  that  they  gave. We  have  not  found  anything. So  the  question,  the  question  is, can  we  tell  in  a  particular  trial  at  the  beginning? Whether  it's  going  to  be an  early  transition  to  Palabllor, late  transition  to  Palbll. We've  been  trying  everything  we  can think  of  and  we  can't  find  anything  yet. Now  I'm  sure  it's  there  actually. It  doesn't  necessarily  have  to  be  there. One  of  the  things  that  can  drive a  system  like  this  is  noise. In  fact,  I  will  say  that  that's  one  of the  things  we  like  about  our  system, about  this  network  better than  an  evidence  accumulation  network, is  because  those  networks, basically  any  noise  is  bad  and these  networks  actually  thrive on  a  certain  level  of  noise. But  we've  been  looking  like crazy  for  that  stuff  and  we  haven't  found  it  yet. Having  said  that,  I  don't  want  to  say  it's  not  there. I  think  that  we  haven't  had enough  creativity  and  insight  yet. I  got  a  brand  new  post  doc  whose  entire  job is  to  search  for  that  check  with  us  in  about  a  year. Yes,  question  about  the  sensory  themselves. If  you  can  record  their  dynamics  when  you  get  a similar  by  stable  state  speech, something,  I  can  imagine  that  there's a  commitment  in  the  tongue, right,  to  taste  more  tasteless. It's  a  great  question  and I  don't  even  ever  present  the  taste  circuit  anymore. But  the  primary  gustatory  cortex is  actually  several  synapses  north  of  the  tongue. You  go  from  the  tongue  to  a  medullary  nucleus, to  a  pontine  nucleus, to  the  thalamus,  to  the  cortex. We  haven't  gone  back  all  the  way  to  the  sensory  neurons. We  have,  however,  gone  back  to  the  very  second  relay, which  is  the  pontine. I  have  a  postdoc  who  nearly  destroyed  herself. Getting  good  neurons  down  there  in  the  awake  animal. And  what  she  found  was  almost exactly  what  we  see  in  cortex. Now  we  need  to  do  a  lot  of  work  on  that. She  basically  just  came back  this  close  to  quitting  science. We  haven't  been  back.  But  what I  can  tell  you  for  sure  is  that that  processing  engaged  in that  is  at  least  as  early  as  the  pontine  second  relay? Yes.  You  said  you  think  imaging  is  too slow  for  these  kind  of  get  the  data  you  need. Would  you  include  like  newer voltage  indicators  in  that  camp? The  question  is  in  my  callous  dismissal  of  imaging, am  I  ignoring  the  new  stuff? And  the  answer  is  I  pretty  much  am. But  it's  important  to note  at  this  point  that  a  couple  of my  best  friends  are  imagers. And  in  particular,  I'm  very  close  with an  olfactory  imager  named Matt  Wakoyak  at  the  University  of  Utah, who's  amongst  other  things,  a  brilliant  technician. And  he  is  one  of  the  people  working  on making  the  newest  generation  of  voltage  indicators  work. And  no  joke,  if  it gets  to  the  point  where  the  details  have  been  ironed  out. Again,  I'm  a  technophobe, I  don't  want  to  develop  anything. I  want  to  use  things  to  answer  questions. As  soon  as  they  got  to  a  point  where  I  can see  this  time  of  activity  in  the  firing, I'm  going  to  drop  electrodes  like a  hot  potato  and  go  to  imaging. Mostly  I'm  talking  about  calcium  imaging  and  I've had  friends  who  are  doing  calcium  imaging  and  Tex, it's  pretty  clear  that  they  can  get  a  shadow. But  then  between  the  signal  noise  ratio  being  relatively low  and  the  slowness  of  the  calcium  changes  themselves, they  can't  really  get  single  trial  resolution  yet  yet. So  we  still  officially  have  1  minute  I'd  like  to  ask. You  might  be  a  pretty  open  question. The  behaviors  you're  describing  are  based on  the  gaping  mostly  in  this  talk,  huh? Just  curious  if  the  animals  were, this  may  not  be  as  naturalists  that, but  if  the  animals  were  trained  to  discriminate between  very  subtly  different,  you  imagine. Could  you  imagine  forcing decision  making  in  this  time  frame to  really  get  at  the  question  of  when, when  the  information  is  ava. That's  a  great  question. We've  devoted  a  couple  of studies  to  that,  that  I  haven't  talked  about. In  one  of  them  we  did  almost  exactly  what  you  described. One  of  the  tried  and  true  tasks  of the  behavioral  neuroscience  is the  two  alternative  force  choice  task. We  did  a  four  force  choice  tasks  so that  we  could  get  rid  of  palatability  as  a  signal, a  palatable  and  unpalable  taste  respond. Did  this  lever  over  here,  pala, different,  palable,  unpalable  taste  respond  over  here. What  we  saw  is  basically  pushed the  decision  variable  way  up  into  the  identity  epic. At  the  same  time,  we've  also done  conditioned  taste  diversion, in  which  we  basically  swap  a  taste from  being  palatable  to  being unpalatable  in  a  single  session. With  that,  we  see  nothing  in  that  middle  epic, only  changes  in  that  late  epic. We  can  do  it  probably we  should  be  doing  a  lot  more  of  that, but  that's  all  we've  got  so  far. All  right,  let's  thank  God  for  a  great  talk. Thank.  All  right, the  sandwiches  were  stealthy  brought  in. Please  partake