thanks Garrett thanks bill all for the
invitation it's great to be here my

first time in Atlanta seems like a great
city

as Garrett mentioned my lab has been
interested in how the cortex represents

information and processes information in
the context of sensory information and

we do this in the whisker system of
rodents which is a really has a lot of

advantages to figure out some of the key
nuts and bolts for how the cortex is

organized and we do many different
things in this area as Garrett hinted

but actually what I want to talk about
today is a question that seems so simple

at its outset but yet it's really
complicated and interesting and that is

there's a map in somatosensory cortex
like in all primary cortices what is

mapped what is mapped in whisker
somatosensory cortex so that's what I

want to talk about today so in primary
sensory cortex in general I think the

field feels like it's been very clear
for years how this organization works

and what's mapped and the main idea
arises from work and visual cortex this

is a classic orientation map across a
little chunk of v1 where the colors show

patches of neurons that are tuned for
different orientations and within this

map the idea supported by many many
studies has been that individual neurons

are tuned for local sensory features
like some orientation that's produced by

some specific circuit and that there's a
neighborhood around each neuron that

contains cells that are coding for that
same feature right that's the classic

mount castle cortical column from 1957
and the idea is that those columns are

then topographically arranged to form
these nice maps and actually an s-1 the

idea has been even simpler because the
main map parameter that's thought to be

spread out across the cortex is simply a
place on the surface of the body

somatotopic location so the idea is that
there are columns and a simple somatic

topic map in s1 is this true while a
number of years ago some experiments

were done by clay reeds group which
showed that if you look more closely

there are more subtle and interesting
aspects of these maps that really I

think challenged us to understand what's
really going on and two of the key

experiments are shown here these are
two-photon calcium imaging experiments

each little blob here is a cell in cat
v1 or in rat v1 and what they're doing

is they're applying
different visual stimuli of different

orientations and then showing you in
color for every single cell at cellular

resolution what was the preferred
orientation for that cell and here's a

little chunk of v1 around an orientation
pinwheel and you can see that there's a

big blob of cells which like you know
this orientation a big blob over here

which like the opposite orientation that
is we see nice orientation columns just

like in the classic people and Wiesel
mount Castle type models but when they

did exactly the same experiments in rats
in rat v1 they found this neurons are

still beautifully tuned for orientation
but now they're completely intermixed in

a salt-and-pepper pattern and so this
idea of salt and pepper maps was first

discovered here but then was examined in
multiple studies in v1 in many studies

and a one where it's still a bit
controversial but there's still there's

very strong evidence for salt and pepper
organization and as I'll show you an s-1

and it really seems like a salt and
pepper organization is the way primary

cortex is put together in rodents okay
so how do we understand this

rodents fundamentally different from
these bigger animals how does this

organization work now one interesting
conceptual idea that came from studying

these intermixed dispersed rodent maps
is that if you think about all the cells

that are tuned for one particular
orientation like the yellow one they're

separated in cortex but beautiful
experiments by Tom MARSOC global and

others showed that those cells which
might be Co tuned for a particular

orientation that is representing a
particular input they're actually

semantically linked even though they're
spread apart so the idea has been that

in rodents you have ensembles of cells
that are tuned for particular sensory

features but they're just not clustered
together in a column they're spread

apart and what matters is the membership
of a cell and it's synaptic ensemble

that's what determines its tuning not
it's particularly loquacious and the

analogy which I think has been very
influential is if you imagine a circuit

board doing some useful computation in
your computer you could in principle

take the locations of all the individual
logical elements and you could scramble

them as long as you preserve the wiring
and that circuit would roughly maybe

even perfectly perform the same maybe
that's what's happened here in these

salt-and-pepper Maps cortex so the
questions that this raised for us is how

to understand this is this true first of
all our salt and pepper Maps an inherent

property
of rodent sensory cortex when you look

at a map like this does this map
completely lack columns which has been

the idea that is the Mount Castle idea
of code toon cells is that just wrong in

rodents and if so a columns irrelevant
for neurons sensory tuning a neuron

sitting here and a neuron sitting here
have very very different tuning even

though they're in the same neighborhood
is that really true does neighborhood

have no impact if so these are the
questions I want to get at

so the whisker system is a great place
to study this because the whiskers are

these mobile tactile devices ISM sure
you've heard but there's a beautiful

representation and s1 cortex of the
entire body map at a nice somatic topic

way including in rodents a clearly
definable anatomical column for each

whisker on the face and that's shown
here in a little schematic and we know

these are real because in layer four
there's actually a physical cluster of

cells called a barrel that sits in the
middle of a radial column and there's

one of those barrels one of those
columns for each whisker on the face and

a perfect map okay and here's just one
way that we know this is cytochrome

oxidase staining here's the arrangement
of whiskers on the face shown in those

barrels in layer four of s1 for example
for the d1 d2 d3 and d4 vibrissae

okay so it's a beautifully organized
place and many many old studies with

classic single electrode tungsten
electrode recordings showed that if you

look at the average tuning of cells in a
given column they look beautifully

precise that is if you wiggle the d1
whisker and ask where within this

overall anatomical map do cell spike
they spike pretty much in the d1 column

maybe a little bit around it if you
wiggle d2 cell spike in the d2 column

and just a little bit around it
reinforcing this one whisker one-column

kind of perfect topography model okay
and then the last thing I'll mention is

that there's an anatomical binding of
instead of them which reinforce the idea

that one whisker one column makes sense
and that's because the lamech input

representing those whisker deflections
arrives in layer four and those little

tight little clusters of cells and then
the intra columnar circuits that

originate in layer four are actually
quite largely radial these layer four

cells project beautifully and strongly
to

over lying cells in layer two three of
the same column and also to the deep

layers not shown here and there's lots
of interesting circuits but they're very

strongly linked within columns now that
idea I have to say ignores important

things it ignores for example in layer
two through the fact that cells get lots

of longer distance inputs they get
inputs from nearby columns they get

inputs from structures that don't map
the whiskers in a topographic way or

from top-down cognitive information so
really there's a lot of anatomical basis

to think that maybe there's some more
complicated mapping but what is it and

that's what I want to get at today when
we think about the whisker map did we

think that all the cells in the given
column are tuned for the features of how

that columns whisker is moving maybe
position or acceleration or velocity but

some kind of single whisker features or
should we think about a salt-and-pepper

map where those neurons are dispersed
like in the v1 gate okay so we did an

early study a number of years ago to get
at this this is a calcium imaging study

that looks at single-cell resolution at
the structure of that map in layer two

three okay and so Kelly Clancy who's a
grad student in the lab exposed region

over s1 here in a little cranial window
used intrinsic signal imaging to find

the column for one whisker like the d2
whisker on the face and then focused it

on that column loaded all the cells with
OGB which is a slightly old-fashioned

but still very good calcium dye and then
under anesthesia wiggled a large array

of whiskers in the pseudo-random kind of
sparse noise way to map the receptive

fields of all those cells okay and
here's what the this looks like actually

with a more modern dye this is G camp oh
this is really not projecting well but

there's lots of pyramidal cells there
we're wiggling the whiskers this is sped

up quite a bit and then you can see
different neurons flashing and

responsive different whiskers so we can
draw an analysis circle around one of

these cells and then show you here in
grayscale the fluorescence response of

that cell over many trials of wiggling
these nine different whiskers and so you

can see that this cell for example loves
the d1 whisker and doesn't like other

whisker so this would be a receptive
field map for this cell so

what is the tuning for cells look like
in one single column these are this is

an imaging field that's localized to the
center of the d2 whisker column cells

are color-coded by which whisker
actually drives them best and you can

see that just like in rat v1 there's a
salt-and-pepper intermixing of cells

tuned for lots of day and this is a very
very prominent thing I could point out

one example these are the five most
responsive cells don't number two in

this field which is in the d2 column
really does respond best this is Delta F

over f2 the DG whisker and not to the
other one so this cell is like correctly

tuned for the whisker for this column
but all these other cells are tuned for

different whiskers okay in the summary
of this study can be seen here where we

localized a bunch of the imaging fields
color coding the cells by their best

whisker as I showed you before and we're
localizing them to the actual anatomical

outlines of the columns that they came
from right and you can see that in each

case there's salt-and-pepper intermixing
but if you count carefully what you'll

see is that the largest number of cells
are always tuned for the correct whisker

for that column about 50% of the cells
the other cells are tuned for any of the

number immediate neighboring whiskers
and so what this means is that in layer

2 3 by this method
it looks like the map is salt-and-pepper

but it's correct on average just with
very high local scatter okay so there's

a number of important caveats to
interpreting studies like this one of

them is that that was done under
anesthesia maybe one animal wakes up and

the brain is more active maybe circuits
we organized a bit and maybe there's

beautiful responses that are perfectly
topographic all right so we wanted to

measure a receptive field map in awake
animals and the whiskers this is very

challenging because when animals wake up
and are interested in something they

actively move their whiskers around it
was great for them but as

experimentalist it means we can't apply
a calibrated deflection to each of those

whiskers like you need to to map a
receptive field so how do we do this so

we came up with a method to be able to
do this we took advantage of the fact

that mice and rats when they're in a
tight spatial position they will

actually rest their whiskers on nearby
objects and not move their whiskers but

you can show by delivering
like to a moving wall panel that they

can still detect and discriminate in
that case so there's a passive whisker

function that they can use even when
they're not whisking but they're

attending and so here we trained mice to
be head fixed under a two-photon

microscope we inserted nine different
whiskers into an array of different

independently controllable piezo x' and
then we would wiggle either deflections

of those nine single individual whiskers
and the animal was trained that when we

do this don't bother to lick you'll
never get a reward

so that's s- or no reward stimulus and
then occasionally we'll wiggle all

whiskers together and when we do that
the animals trained lick and get a

reward so the animal looks to that
stimulus we also throw in some other

stimuli like a couple of different tones
that are not rewarded and some blanks

that are not rewarded the animal learns
to do this task very well here's over a

bunch of training days the average
behavioral response the licking of a

bunch of mice to the whisk the all
whisker s plus stimulus the mice learns

to lick to those the mice rapidly learns
to to suppress their responses to the

single whisker stimuli here and then
they respond even less to the tones or

the blanks and the fact that they make
more incorrect responses false alarms to

the individual whiskers which they
shouldn't then two tones means to us

that they're actually paying attention
to the whiskers right that's an

important result so we have an animal
that's receiving all these stimuli and

its attentive to the whisker and what
we're going to do is the animal cares

about this stimulus but we are gonna
care about these stimuli because here

we're delivering calibrated stimuli the
animal is not licking which would

contaminate these responses and we're
going to map receptive fields to those

stimuli and what do we see so this is
results from a number of mice that are

all put into a common reference blank
frame the black circle here is the

boundaries of a reference column the
center of the column is in the middle

the position of each symbol is the
position of a cell okay and actually

these cells are all being collapsed onto
exactly the same reference field so the

black circle is the same column in all
cases but I'm just separating out the

cells the red cells in the middle are
the ones that were tuned for the whisker

corresponding to this column these are
the correctly tuned cells

right and there's lots of them but you
can see that intermixed within the same

population where many cells that were
tuned for the next whisker over in this

direction or the nest Whistler in this
direction or this one or this one

and in fact all the surround Wesker's
and if you add these up about 50% of

cells are tuned for the correct whisker
and the other ones are tuned for a

nearby whisker
so the salt-and-pepper intermixing

happens even in the awake attentive
animal and this is again in layer 2 3

now if you stand back a little bit and
say outside that one column is there

still correct average structure you see
beautifully the correct average

structure now our column of focus here
is just the small black circle in the

middle and cells that are located
outside it our position that they're

correct locations outside it and so the
cells that are responsive to this

central whisker these guys here are not
only found within that column but

they're also found in the neighboring
columns that's the salt and pepper map

and cells that were responsive to the
whisker that was one row closer to me on

the face tend to be located here the
cells one row in the other direction

located over here etc and so you can see
that the central focus of all these is

topographically orderly around the
central column that's the average map so

the average map is correct but there's
this tremendously high local scatter now

I'll throw in a little fact for you that
was all pyramidal cells we're doing cell

type-specific imaging it's not true for
every cell type one of the key

inhibitory cell types and cortex is the
parvalbumin or pv inter neuron and they

have many interesting properties which
we study in other projects but one cool

property about them is that they receive
input from virtually every pyramidal

cell within a hundred 150 microns around
them they provided inhibitory output to

those same cells and so they're in a
great position to average the local

activity of the nearby Network so you
might imagine that they're tuning might

actually be more on average correct
right thought that the pyramidal cells

are scattered salt and pepper but these
cells might be integrating and averaging

and in fact that's what happens when we
when we image in separate experiments in

pv cream ice from pv inter neurons those
neurons are actually quite precisely

localized to the correct columns which
with much less scatter across columns

okay so
the salt and pepper map seems real among

pyramidal cells now we also wanted to
know could it be modulated and I'm going

to show you two examples of how it's
going to be modulated if one is in the

case of attention and really we can
think about this as we measured this

tuning in a trained animal that animal
went through multiple days of training

and during that training it's learning
to pay attention to certain stimuli and

not others
maybe that's shaping the map right we're

imagining that we're measuring some
inherent property of this organization

but we also know that maps are plastic
maybe we've trained the map to look like

this so is that true so we did in
parallel with the study I just showed

you another version of a task in which
we applied to the awake Mouse exactly

the same stimuli exactly the only thing
that was different is that now the

animals trained to lick to what we're
presenting two tones one of the two

tones and not the other and it's not
looking to the all whisker stimulus so

we're training the animal attend away
from the whiskers a tend towards sound

and particularly tone a and the animals
do this they're not as good as for the

whiskers but they look beautifully to
the correct s plus tone they make a lot

of mistakes and they they lick to the
other incorrect tone as well but they

looked very very little to any of the
whisker stimuli or to the blanks and

again this tells us that we've succeeded
in the training these animals are

attending to the whisker a to the sound
the tones because that's the mistakes

they make and not to the whiskers so how
does this affect the map and here what

we find is that the topography of the
map is very very similar within each

column this was the whisker cute cat
task I showed you before here it is for

the sound cute version about half the
cells are tuned for the correct whisker

half the cells are tuned for another
whisker although actually if you notice

they're slightly more cells tuned to the
correct whisker in the sound cued task

and this suggests that the tuning of
cells within each column is a little bit

more similar to each other in the sound
cute case and we could analyze this in

more detail and we find that this is
absolutely true and that's what the

bottom row shows you here so what we're
computing here on the Left bar is tuning

similarity between pairs of neurons we
take two neurons are in our imaging

field we compute what's called a signal
correlation which is the similarity

the mean tuning curve that's plotted
here as a function of the distance

between those two cells and because of
the way the cortex is organized the

farther apart two cells are the less
similar the tuning is so there's an

overall slope but you'll notice that in
the sound cued case tuning similarity is

higher between cells in a column these
are all within a column then in the

whisker cute case and that's consistent
with this idea that more cells are tuned

for that columnar whisker if we look not
at risk or revoked activity but noise

correlations which are essentially
spontaneous activity that can't be

explained by tuning we see a very
similar phenotype that that falls off

with distance as well which was known
but it falls off more slowly in the

sound cute case than in the whisker cute
case and so what this tells us is that

both sensory evoked activity and
spontaneous uncontrolled activity right

they're both more similar between
neurons in a column when the animal is

not attending to the whiskers and
they're less similar within a column

when the animal is tuning it is
attending to the whiskers and that

actually makes perfect sense because
what it means is that when training and

attention has told the animal pay
attention and discriminate what's

happening to the whiskers bells are now
less correlated they're d correlated in

those columns right and that provides a
better discrimination signal between

those different inputs so the whisker
cube task D correlates neurons within

each column it preserves the overall
salt-and-pepper organization but it

changes these subtle correlations which
are in the right direction for improving

coding or whisker discrimination ok the
second example I want to tell you about

about Kenny's maps change is more of a
long-term plasticity example ok we

noticed and we were a little concerned
that in all of the experiments that had

documented salt and pepper maps ours and
s1 the v1 cases the a1 cases they were

all done in animals that were raised in
the normal rodent you know housing

environments plucked out one day and
then an imaging experiment was done but

we know that sensory cortex is highly
plastic the sensory experience and in

particularly r23 is very highly plastic
all these salt and pepper maps are

layer two three we also know that
standard rodent laboratory housing is

actually a deprived environment there's
not much going on and food falls out of

the sky right the animals don't have to
do very much

so maybe if the animal had richer
sensory experience maybe there would be

that experience would refine
developmentally refine these maps better

and maybe they would become less salt
and pepper and more columnar that is

maybe salt and pepper structure is
actually an artifact of how we raise our

animals it's an incompletely developed
map could that be so to test this

another student in the lab a mule in
Mussoorie a did a nice study where she

took animals at the age of weaning which
is just about a week after they begin to

whisk and separated litters into one
group that had tactile enrichment which

is just toys put in the cage every
couple days and the other animals that

were raised in standard environments and
she did this in two strains of mice that

allow us either to image layer two three
pyramidal cells or layer four excitatory

cells and we're going to use this to ask
what's the difference in topography of

the map between layer four and layer two
three and how does enrichment or this

extra experience change that structure
and this is all imaging from excitatory

cells from pyramidal so before I tell
you the enrichment effect let me just

tell you the difference between layer
four and layer two three I mentioned at

the outset that layer four cells are
clustered into these barrels and

actually the way that Phil amic
afferents innervate them is that a given

Afrin essentially innervates the correct
barrel and so actually you'd expect that

layer four might be topographically
precise even though layer two three of

salt and pepper so is that true and so
we normally have two animal she examined

exactly that here what we're plotting is
what fraction of cells are tuned for the

correct whisker for the column as a
function of what's the distance of that

cell from the center of that column
farther and farther away and what you

can see in layer two three in black is
that about 60 percent of cells in this

study we're tuned for the correct
whisker within the column and then that

fell off slowly over multiple columns
the columns are shown here until there's

very few cells once you get a few
columns away and that's consistent with

the salt-and-pepper map if you look in
layer four of those

Minh the same normally house conditions
what you find is at the center of the

column in layer four there's more cells
that are correctly tuned and that tuning

drops off much more sharply with spacing
in layer four and so what that means is

layer four is in fact more
topographically precise and then somehow

the signals are dispersed as they move
up into layer two three okay so now what

is enrichment - well in Richmond
actually increases the fraction of

correctly tuned cells in the center of
columns but it falls off approximately

the same right but the height of the
that tuning precision gets higher and

surprisingly the same thing even happens
in layer four and if you look at these

Lots in the top with just some example
fields I think what you can see is that

after enrichment there's still salt and
pepper organization but now things

actually look a little bit more
homogeneous okay so that's one measure

of how the maps change within Richmond
another measure is not thinking about

salt and pepper per se at all but you
can you can study the topography of the

map by asking what is the point
representation of a stimulus in the

world in the activity of cells across
that map so we wiggle one whisker and

we're gonna use imaging just to ask how
strong is the average response to that

stimulus at one location how does that
fall off that is what's the hill of

activity that you see in s1 in response
to that single point stimulus of the

periphery okay
and so what's plotted here is the mean

evoked response in Delta F over F as a
function of position away from the

center of that column and what you can
see is that in the normally has two

animals you wiggle a whisker there's the
strongest activity in the middle of the

corresponding column and then that falls
off gradually right and enrichment

causes the center response to strengthen
all right so the top of the hill gets

taller the flanks stay the same and so
the hill gets slightly sharper with it

Richmond okay that happens beautifully
in layer two three and not at all in

layer four and I think you can see this
here in the 2d version of these same

plots here these these are spatial bins
of cells around some reference column

and the color scale shows you this
strength of response to the reference

whisker and I think what you can see is
that when we wiggle the reference

whisker and the normally housed animals
yes you get activity within the column

but there's a substantial spread outside
the column and after enrichment that

hill
activity is much more focused within

that average column so enrichment is
sharpening up the columnar structure and

actually the strongest evidence we have
that the columnar structure is really

sharpened comes from those measures of
tuning similarity and similarity of

noise correlations that I showed you
before and actually in the interest of

time I'm going to skip one
quantification I'll just show you the

other quantification okay
these are spatial plots like I just

showed you before or each of these
histogram bins represents lots of cells

that were found in a particular spatial
bit and cortex relative to some

reference column whose outline is shown
there and the color scale in this case

on the top shows you the mean tuning
similarity the mean signal correlation

between cells in this bin and all the
cells that were found in that column so

the bluer you get the higher the tuning
similarity and so what you can see is

that in the normally housed case many
many cells over long distances have high

tuning similarity to the average cell
within that column and that's a

reflection of the very strong
salt-and-pepper intermixing that happens

in those normal normally housed animals
after enrichment what you can see is

that there still is a ring of similarity
of cells but now those are cells that

are much more immediately outside the
column in some distance away now the

tuning similarity really drops off so
you can see the development of this

columnar structure and tuning similarity
with enrichment and you can see that

particularly well if we take this same
data and instead of looking at in 2d

we're just going to collapse it to one
single distance dimension here along one

radial dimension here which is this
dimension but what's the average tuning

similarity as a function of distance to
that column center and here what you can

see is that in the normally house case
as you move in distance away from the

column Center and you cross that column
edge which we can see exactly we know

exactly where it is from the anatomy
there's very little drop-off in the

signal correlation meaning tooting
doesn't change much it doesn't respect

the column boundaries but after
enrichment tuning begins to drop off at

the column boundary and the same thing
is true even more strongly for noise

correlations noise correlations are
very strong over long distances in the

normally housed cortex become become
more focused after enrichment and if you

look in the average here in the normally
housed case again they're largely flat

across that column edge but after
enrichment they begin to fall off right

at the column edge and so enrichment
creates a tuning structure which aligns

to the edges of those column boundaries
okay so let's summarize this and I can

tell you how how we think about these
data okay so we started out with the

question of how is s1 organized is every
cell in each column simply encoding its

corresponding whisker on the face that
is their single whisker feature

detectors and there are no mistakes
within the map or is it some very

heterogeneous salt-and-pepper
organization where cells within the

column are intermixed for lots of
different and what I showed you is that

the salt and pepper organization is very
robust cost lots of different ways of

measuring it but it's also plastic right
and that the more experienced animal

gets you can get changes in the salt and
pepper organization that now begin to

finally respect those column boundaries
so the columns seem real but they're not

perfect they're real and their edges
have meanings but there still are many

cells which are pepper within the salt
and pepper map yeah okay and I should

say so the effect of enrichment how can
you think about this on circuit on the

circuit level we're very interested in
this but we haven't done a single

experiment yet but you can ask how is it
that cells within a column begin to

become more similarly tuned how is it
the noise correlations get actually

stronger within a column but fall off a
cross coat and one very simple

hypothesis is that circuits within a
column might strengthen with enrichment

whereas circuits that go across columns
might not or maybe even more inhibition

develops across columns right these
would both be very simple ways to

produce this kind of behavior and we're
very curious to know what actually

happened okay so I want to take the rest
of the time to answer I think a

fundamental question about this there's
more organization after tactile

enrichment but they're still salt and
pepper straw

so how do we think about that salt and
pepper structure and I want to boil this

down to these two linked questions is
the salt and pepper organization

essentially noise within a distributed
map that is what really matters is some

kind of tuning ensemble and not the
position of a cell right this cell tuned

you know to the wrong whisker within the
blue column was this a developmental

mistake that created that cell is that
self simply being averaged out and not

used or is there some other type of
organization that happens here and the

alternative that I would propose is that
maybe something else is being

represented that we're not even
measuring here and maybe that thing

really is coherent across calm what do I
mean by that

when you measure a receptive field in
the way that I've been describing you

wiggle one whisker at a time and you're
measuring the response of neurons to

individual whisker deflection you get
some tuning but fundamentally you're

only asking how do the cells respond to
single whiskers but we know in the

natural case that when the animal uses
all of its whiskers in the world on

objects it's never one whisker that's
deflected maybe there are more

complicated multi whisker features which
are being encoded here and maybe they're

encoded in some coherent way across
columns that were just missing with this

organization does that make sense okay
so and here's some indication that this

is not so crazy so of course multiple
whiskers move meet your Hartman's lab

did a really nice experiment a few years
ago they had essentially a vertical wall

a rat you know would amble over to the
wall and just whisk on it to explore it

and they had a very nice clever method
for being able to detect all of the

individual whisker contacts that were
made on to that wall by the whiskers

okay and so what you're seeing here is a
trace over time in black of the animals

nose location it started off kind of
where you guys are it would approach the

vertical wall which is where the screen
is it would hit the wall and then its

nose started off here and lured low on
the wall for a while then it started to

explore higher and higher on the wall
and in the meantime all of these colored

dots are individual contacts by all the
different whiskers on whiskers one color

hitting the object okay so lots and lots
of whiskers are contacting

and when they add when they looked at
the statistics of this which you might

imagine are very complicated they were
able to come away with some simple

underlying first-order description the
one first-order feature which is not

surprising is that when multiple
whiskers contact it's typically two

adjacent multiple whiskers which make
the closest cup right that makes sense

and then number two they can analyze
what's the time frame of that are the

whiskers hitting at exactly the same
time or at wildly different times and

what they found is that most of the
contacts were happening for adjacent

whiskers in a plus or minus 50
millisecond interval and again that

shouldn't be surprising because the
animals whisking back and forth and plus

or minus 50 milliseconds is basically
one contact time before the whiskers

retract and go forward again
okay so if we want to understand how

neurons and s1 might be tuned for
complicated multi whisker patterns one

logical place to start is in how s1
represents adjacent whisker contacts

within a plus or minus 50 millisecond
interval and so that's what we did the

space of all multi whisker combinations
is way too big to explore the tuning but

there's a subspace that's tractable in
this subspace that we chose here was a

subspace of local that is adjacent to
whisker combinations and sequences with

the plus or foe but minus 50 millisecond
intervals so you can imagine a bunch of

whiskers on the face this would be like
you know this whisker we goes first and

then slightly later this one or this one
first and then this one what are these

if you think about the whiskers moving
on objects these are essentially local

motion vectors on the whisker pad okay
so we want to know is there tuning in

this space of local motion vectors so
the experiment is a classic

old-fashioned one and Mouse is
anesthetized we put our nine piezas on

the face we're going to apply exactly
the same stimulus to each whisker but

now we're gonna present lots of
different stimuli and map responsiveness

within this two whisker space this is an
anesthetized Mouse okay and for these

experiments were recording most of the
time in the d1 column we Center the

array on that column so that through the
array we can now interleave these

different types of stimuli we either
present each individual with

alone okay or all - whisker combinations
that involve that central whisker for

the column or recording I'm going to
call the columnar whisker or CW - all

combinations of that whisker plus the
adjacent surround whiskers so CW SW

combinations are these adjacent
combinations that involve the columnar

whisker or all the other possible -
whisker combinations which are surround

whisker surround whisker come and we do
that in this first experiment I'll show

you at just a few different Delta T's
within that relevant ring and what do we

see okay here's an example neuron each
of these bars is the response to one of

these complicated two whisker
combinations stimuli here's whisker one

here's whisker two the black ones are
the single whisker stimuli so this

particular neuron spikes best to the
single whisker stimulus of the Gamma

whisker and less to the d1 whisker and
on and on

okay but if you look at the responses to
all these two whisker stimuli you see

there's a bunch of them that drive
stronger responses than the single

whisker alone and the favorite one is
the combination of gamma and d1 okay

here's another cell same kind of
principle except it likes a different

combination it likes the e 2d one
whisker coming and we analyzed when we

analyzed this early experiment what we
found is that virtually all cells spike

more to a two whisker combination than
to a single whisker so it's reasonable

think cells are are liking the stimuli
in some sense and when we analyze what

combinations they like 70% of the cells
within a column prefer one of these

stimuli that is combinations that are
the the columnar whisker plus a surround

whisker as opposed to something more
distant or something non adjacent okay

and not only did they prefer those but
we can make a quantitative measure of

how selective they are in their tuning
among either the set of B wsw sequences

or the set of SW SW sequences and cells
are more selective among these sequences

than they are among these sequences and
so that convinced us that there's

something in a single column that is
representing well and discriminating

different two whisker combinations that
involve the columnar whisker

okay so we did a follow-up study to
learn about this in more detail okay and

so this experiment is very very similar
they - five Mouse same stimuli but now

actually we're going to focus on those
combinations that involve the columnar

whisker with just a few combinations
that involve other whiskers and now

we're gonna present all possible Delta
T's on differences between the whiskers

from minus 52 plus 50 so that we can
present a very very large signal a set

of thousands of stimuli we record for
hours and hours and hours and then

essentially we use a reverse correlation
like method to go back and ask what was

the favorites the meals for each cell
okay so what do we find here's an

example cell this cell was recorded in
the d1 column and here I'm showing you

in the raster's on the cells response to
the d1 d2 whisker combination or the d1

c2 whisker combination etc and you can
just see here that the cell likes d1 d2

also likes d1 d1 to some degree right if
we blow up these raster's you can see

that the individual trials here are
actually trials in which were varying

delta T from 50 milliseconds d2 leads d1
to 50 milliseconds in the other

direction that is we're sampling that
whole range and with a little bit of

smoothing on this we can generate what's
essentially a spatio-temporal receptive

field or a stirrer for these cells in
which the pairwise combinations of

whiskers are on the bottom and Delta C
is on the y-axis and you can see that

this particular cell is this one prefers
d1 d2 combination at a particular time

delay of about 30 milliseconds d2
leading d1 okay so we're going to do

that for all the cell's and
statistically we're going to figure out

which cells are significantly tuned for
one particular combination over all the

others and what we find when we do that
is tremendous tuning for two whisker

combinations in a space that I hope I
can convince you looks like an intuitive

real space for tuning okay so I'm gonna
focus first on the spatial tuning that

is tuning for the different combinations
we're gonna ignore time for each cell we

found the optimal delta T and this
analysis was just so that delta T so

here's one cell that's recorded in the
d1 column

the green here is the spiked count to d1
d2 combination or D 1 C 2 or D 1 C 1 and

these are organized in the same layout
as they are on the face ok so you can

see that this cell loves d1 d2 the
asterisk shows its favorite response and

it responds to all the other
combinations with d1 much much less now

what are these other things the dashed
black line is the response to d1 alone

and the red line is the predicted linear
response to the combinations that we

applied and we're using the predicted
linear response here as kind of a boring

outcome if the cells responded with the
same number of spikes as you'd predict

from the sum of the individual whiskers
we would say there's no particular

computation that's going on there's not
you know much happening but you can see

what happens this cell responds to its
favorite stimulus with pretty much the

same number of spikes you would predict
linearly and it responds to all the

other combinations with dramatically
fewer spikes that is its sharpening a

two whisker tuning curve from the
predicted values of its inputs okay and

that's what this cell does also this
cell likes b1 and b2 that's what this

cell does it likes d1 and c2 this sub is
the same kind of thing except its

favorite response it actually manages to
generate a stronger response to the

favorite combination than was predicted
linearly and that is what it turns out

that 50% of cells in s1 do I'm just
gonna flash up a number of other

examples to convince you there's a lot
of them the only thing I would ask you

to take away from this is that all of
them obey this pattern that they sharpen

relative to the predicted linear sum and
they're all pointing in different

directions meaning that different CW SW
combinations are represented in each

column as kind of a basis set maybe but
representing these stimuli ok so I'm

gonna skip this
modification but the bottom line is that

we can do a number of different
statistical procedures to convince

ourselves and hopefully hopefully others
that this tuning is real and that the

tuning is absolutely sharper than the
predicted linear sum and actually I'll

just show you this one panel this is the
selectivity of measured responses to

these two whisker stimuli and this is
the predicted selectivity if the cell

were responding linearly the vast
majority of cells sharpen their tuning

and they sharpen it in a way that's
represented by the example fell I showed

you before
they're sculpting a weaker response to

all but one single preferred stimulus
and the preferred stimulus they respond

to pretty darn linearly okay and that
suggests maybe inhibition is suppressing

spikes to non-preferred stimuli but
somehow allowing spikes through for

preferred stimuli okay so that was about
tuning for space I'll just show you one

thing on tuning for time cells are in
fact tuned for time difference for delta

T as well that tuning can be fairly
narrow this is the average response of

neurons to different stimuli that have
different Delta T's centered on the

whatever the best delta T is for the
cell and so the width of this peak is

about 10 milliseconds and that means
that cells have on average 10

millisecond resolution in determining
which whisker went first and by how much

okay and I'm not going to show you the
data but most neurons like one

particular tuning best they like it when
a surround whisker is wiggled before the

columnar whisker and not in the opposite
direction not columnar before surin what

does that mean in terms of local motion
on the face the cells prefer local

motion that's in bound to their columnar
whisker okay and so what if cells are

doing this and it very importantly those
cells that were the pepper in the salt

and pepper map that were tuned for the
wrong whisker for the column they do

this really well what most of those
cells prefer is a combination of their

favorite single neighboring whisker with
the columnar whisker and so overall 70%

of all cells in the column
prefer a CW SW sequence as I said 50% of

those missed tuned cells actually prefer
that same class of stimuli

and 85% of the neurons that are tuned
for the columnar whisker also prefer

that so there's tremendous prevalence of
this type of T okay I'm gonna skip the

decoding but I'll just tell you that we
can what this suggests is that there are

populations of cells within a column
which are I diversely tuned for

different CW SW sequences and therefore
you can decode or the column might

represent or we can decode from the
population activity in the column which

to whisker sequence occurred and that
turns out to be quite true we can decode

that actually quite well much much
better than simple linear prediction

okay so what is mapped in whisker s1
there's clearly salt and pepper tuning

when you measure it for single whisker
stimuli but what do you find if you use

this particular simple simple subset of
multi whisker stimuli you find that many

of those cells that were tuned for the
wrong whisker and single whisker space

are in fact tuned for local motion from
some surround whisker moving towards in

an inbound direction towards the
columnar whisker okay so this map in s1

is not a simple somatic topic map this
map has salt and pepper organization but

that organization doesn't just reflect
noise it reflects something coherent and

we think one thing that's being
reflected here is that this organization

reflects cells tuned for local multi
whisker motion and a particular type we

proposed at the bottom here that what
each column is doing is its representing

both single whisker features of course
but in addition the set of all local

motion sequences that are inbound to the
whisker and this would be a new way of

thinking about what's represented at the
columnar level and I should say that

there are a few very nice papers this is
actually just a couple there are more

that have been searching for
higher-order features that are

represented in s1 and there's been lots
of suggestions for what these might be

people have focused on very high other
properties like large-scale coherent

motion across the whole pad for example
but what we're suggesting here is if you

think on local elementary levels you can
see very robust tuning across many men

finally if you think about how a neuron
might be constructing this narrow tuning

for two whisker sequences by sculpting
its response to the linear prediction

this is very very similar to a
computation that many of you may have

seen before
this is Direction selectivity in motion

in the visual motion right visual motion
like even in the retina neurons will

receive inputs from lots of local
locations in space but they might only

respond when a bar is being swept in one
direction and not in the other direction

how do they do that they do that through
a very specific and quite beautiful

inhibitory circuit that's been described
that suppresses the cell's response to

the non-preferred direction of motion
that's exactly what's going on here

there's some kind of suppression we
don't know what it is that's inhibiting

the response of the cell to the
non-preferred inputs but allowing the

preferred input to get through unscathed
and so maybe this is a hint that there's

some kind of common sensory computation
going on here in cortex there in the

retina that's pulling out some useful
features of the sensory world okay

so how have we understood these
salt-and-pepper Maps higher-order

feature selectivity is intermixed in
rodents and I and I would argue that's

why we see this apparently this or
disordered map because these cells

actually are tuned for other
higher-order features there is some of

that feature tuning in primary cortex of
larger animals as well but we think of

it as taking place in downstream
higher-order cortical areas maybe

there's just not enough cortical
territory in rodents you do that all in

some higher dedicated area maybe more of
that is done intermixed with in primary

cortex the columns that were proposed by
mount castle I would argue that they do

exist despite this but cells don't share
all exactly the same tuning which was

his primary proposal they share related
tuning features and in our case the

related feature is local motion that
involves the columnar whisker in some

way and this suggests that columns are
in fact important for computation but

exactly how we still don't know okay so
all the work I showed you today was done

by four students
Kelly Clancy super talented graduate

student who now is a postdoc with
Chalmers tick flow goal mu limits she

did the
salt-and-pepper description a mule in

Missouri I did the enrichment experiment
and she just finished her PhD on Chad

Wang is a postdoc he did the awake
experiment for measuring receptive

fields in awake animals and Kevin leboy
Juarez did all the work on two whisker

combination tuning thank you very much

yeah yeah so the question is do we also
see a map of the local motion preference

and actually the the primary way I would
predict that that map would take place

is actually within a single column I
would imagine that the neurons on one

side of the column let's say towards d2
they prefer d2 motion and bound of the

whisker on the other side they prefer
the other way that's absolutely a

prediction we are gearing up right now
to do the those experiments with calcium

imaging but we wanted to do that in the
awake case so actually we were waiting

for the awake behavior to be good enough
for that and it's ready now basically so

that's the next thing to look at yeah

far away yeah absolutely yeah we've had
so the question is when we look at

signal or noise correlations like after
enrichment they do fall off around the

central column but they often oddly pop
back up again on the edges and so we've

had some debate in the lab about this in
the way that we're normalizing the

position the normalization is quite
accurate around the column but once you

get far away from the column there's
variation across animals about exactly

which column you're in and so what we
need to do is an another analysis of

that data to figure out is that tuning
systematically related in some way to

the anatomical location of those cells
and that's our hypothesis but I can't

answer that for you yeah good question

auditory tasks correlation prankster
right yeah yeah that's a great that's a

great question so what
so did we do a task in which an animal

was changing his attention from auditory
to visual and back on Tori the whisker

and back again and then we see that
effect no those are actually separately

trained groups of animals so we do not
know yet whether that's a training

effect or an attention effect however
where now we've developed now a new task

in the lab which is a cued attention
task with the hope of getting at exactly

that as you might know those are quite
challenging to train and mice but I

think we're getting there so I want to
be able to answer that but I cannot tell

you that that would be very exciting if
that's true particularly because we can

do you know all the cell type-specific
imaging to try to figure out what's the

circuit basis for that if the
correlations are changing yeah

yeah so is the map salt and pepper just
because the brain is shrunk and there's

no room to separate things out why I
mean certainly some people have made

that argument but I think you need to be
maybe more precise in the alternative

that you're suggesting so for example in
the extreme you might say that maybe

there's not enough cortical space in
mice have separate higher-level areas

and all that processing has to be
intermixed in primary areas and we know

that's not true right because around
visual quartet v1 of mice there's an

array of beautiful higher-order areas
that seem to be doing higher-order

feature extraction they're much smaller
right than you get in primates but they

do exist whether that could be true at
the columnar scale the way I would

phrase that is there are gonna be
developmental processes and plasticity

processes that help decree and shape the
circuits that are building the columns

right and those are gonna have some
inherent spatial scale maybe the scale

is such in rodents that you can't
precisely wire down to that level and I

think that argument you know has some
merit perhaps but then there's the

counter observation that like when
merciful has looked for synaptic

connectivity between cells that have the
same tuning but are dispersed within the

map there's very precise beautiful
organization of that so at some level

the brain is able to make very specific
wiring so my gut feeling is there's

something correct in the question that
you're asking but so far we don't have a

good or the field I don't think has a
good you know mechanistic Pacific

explanation for why it exists very
high-level questions

talked about v1 and vision Derrida
hazard this conjecture about just the

Madison Station in general in the skin
it would be a total speculation is there

salt and pepper in the skin I will say
that when when people have done fairly

basic experiments characterizing
topography within the body map

basemap and s1 like with classic
tungsten electrodes that just creep

along there's certainly orderly
organization but it's not perfect

they're already very clear from those
papers that it's not perfect in fact

it's less perfect than I think was
suggested on average from the early

whisker work so I would suspect that it
occurs but it hasn't been being off or

having errors
whereas continuing I agree I think

there's an interesting ongoing
discussion about whether different

cortical different sensory areas for
example in rodents are all doing the

same thing or uniquely different from
each other and the Bill of cortex

because it has those barrels seems like
well maybe it's different but many many

of these features are the same I would
argue that these primary courtesies are

very similar to each other but all
cortex is a useful place to do this work

because we have beautiful anatomical
markers of where the column boundaries

are and that makes it very possible to
do this precisely but so far we haven't

seen a phenomenon here which seems
completely out of the ordinary for other

areas but I think this is the
organizational principle exploring all

possible Delta T's there's a lot of
overlap between them yeah yeah yeah so

the way that we made the stimulus space
tractable for these experiments was by

not probing single whisker kinematics at
all but what we did was we took a

waveform for how you move away single
whisker that had been determined by Dan

Scholz's group using reverse correlation
as like an optimal stimulus for driving

s1 spiking in rats and we used that as
the fixed stimulus that we never varied

from whisker to whisker the only thing
that we varied was which whisker was

being wiggled and the delta T so it
would be very interesting to know how

this correlates with other types of
tuning features but the dimensionality

explodes and how to fit all those
stimuli within an experiment becomes

I think pretty much impossible well
let's thank Dan