EEG,ERPs, & relation to single neuronal activity.

Report
EEG,ERPs, & relation to
single neuronal activity.
(For 312, & Prosem.)
2 parameters determine
what an electrode records:
• 1. Electrode tip diameter
– A) .1 to 3 microns (mu or µ) gives you
action potentials (spikes) or psps.
– B) > 50 mu : EEG and its derivatives,
such as event-related potentials
(erps), event-related spectral
disturbances (ersps).
…and the other parameter:
• 2. Biological amplifier filter
settings:
– A) High pass: passes frequencies >
1KHz: gives you spikes, multiple unit
activity or “hash,” multiple neurons
near one electrode (usually 5-50 mu).
Nobody liked hash for along time
until Nikos Logothetis (more later).
– B) Low pass (0-100 Hz; usually 0-30)
gives you: EEG, ERPs, ERSPs PSPs
The point……
• The Bio Amplifier subtracts one input
from the other.
•
Referential (Signal – Reference) is standard EEG/ERP set up.
It maximizes signals because signal is distant from “0” the
“quiet “ reference. (No such thing, really, but reference
<<<<<< signal is ok, usually.)
•
Differential (Signal 1 –Signal 2)maximizes difference between
electrodes. Usually used in animals where one can look
surface to depth of cortex, yielding a huge signal. In humans,
we are restricted to scalp, usually. Though for EOG (eye
artifact channel), electrodes above and below eye can be
hooked up differentially to great advantage. (Explain, John,
Meixner).
EEG and derivatives—meaning all these
are obtained from EEG hook-up:
• 1. Spontaneous EEG: Love to show you
picture but can’t draw in ppt. So take a look
at
http://en.wikipedia.org/wiki/Electroencephal
ography
• I just want you to look at voltages varying
as a f (time) –that’s EEG. It’s always there,
in (on) your brain. The 2 major variables
describing it are amplitude
(height/magnitude) and frequency or how
many times per sec the signal crosses 0
microvolts (uV) and goes from plus to
minus.
•
Classical EEG frequencies
and their meanings:
• Alpha: (8-13 Hz). Associated with unfocused relaxation,
not sleep. Sinusoidal-synchronous high amplitude.
• Beta: (14-30 Hz) seen in intense arousal—as the exam
is handed out. Or during dreaming sleep, called
paradoxical sleep. What’s the paradox? Hardest to
wake you up, but looks like you’re up and taking exam,
or anticipating torture. Low amplitude, unsynchronous
(random-ish) frequencies.
• Theta: (5-7 Hz) seen as you fall asleep, associated with
hyponogogic images. Synchronous.
• Delta: (0- 4 Hz) seen in deep, dream-free sleep, or in
pathological state—over a lesion, a cyst with pus. But,
despite what some fools believe, also seen in waking,
as in CNV (described below). High amplitude.
Examples:
Other EEG Phenomena:
• Gamma: (30-100 Hz—40-50 average).
Recently studied. Associated with
higher cognitive phenomena, like
learning, binding. When it is
necessary that many cortical areas
communicate (bind), the observed
carrier wave among them is gamma.
During a memory involving sight and
sound, for example, visual and
auditory cortices must talk to each
other. Gamma theorized to carry the
information.
Other EEG Phenomena:
• Petit Mal epileptic activity (absence
seizure—see my imitation or see movie “The
Andromeda Strain”) : is a 3 Hz repeating
cycle of “spikey” looking EEG waves
followed by domes—spike and dome, spike
and dome, spike and dome, 3 times a
second, accompanied by rhythmic blinking,
also at 3 Hz, and the apparent
inattentiveness (absence) of the sufferer.
• This isn’t Grand Mal epilepsy, characterized
by falling down, foaming at mouth, spasms
of muscles, biting the tongue, and huge,
wild EEG spikes, and occasional death.
Other EEG Phenomena:
• Sensory-motor rhythm (SMR),
which is synchronous 12-14 Hz
(like high alpha) activity seen over
somatic sensory and motor
cortices. The lack of normal
amounts of SMR is also
associated with epilepsy.
EEG Derivatives
• Event-Related Potentials(ERPs):
• If as the spontaneous EEG is coming
along, one presents a discrete stimulus
to the subject—like a light flash or tone
pip-- the EEG breaks up into a series of
much larger peaks and troughs. This
series of waves is the ERP, related to the
stimulus event, whatever it was. The
number of peaks and troughs depends
on the complexity of the stimulus. For
simple stimuli, there may be just 2-4. For
complex psychological stimuli—like
names of people– there may be 5-8.
EEG Derivatives-2
• If you go to
http://mitpress.mit.edu/catalog/item/defa
ult.asp?ttype=2&tid=10677
• ….and download that sample chapter
and go to Fig. 1.1B, you’ll see an
example of a continuous EEG wave
where there is a stimulus presented
about every second. Note the ERPs
towering over the background EEG when
“O” is presented (see caption).
• I also provided a handout where you can
see EEG breaking into an ERP.
EEG Derivatives-3
• To clearly see the O-evoked ERP
components (called P300), you
average all the O waves into 1 bin,
and all the X waves into another.
• An example of the average ERPs
is in next slide
Actually , this is from my work. It shows a big ERP
component (arrow) called P300 which follows
psychologically meaningful stimuli, like your birth date,
and the other flatter average is to other, irrelevant dates.
Note other down-going and
up-going waves in the ERP.
• Each is called a component. The earliest waves—not
easily visible here, because I heavily filter them out—
represent the sensory information reaching specific
sensory cortex from lateral pathways. Each
wave/component represents the discharge of a certain
synaptic organization. The next set of waves
represents the sensory information mediated via
medial reticular pathways. Both kinds of components
are “exogenous” ERPs, because they represent
external, sensory info. (Why do reticular components
come later?) These exogenous ERPs are also called
(stimulus-) evoked potentials . People used that term
to describe all time-locked EEG events, but then when
the endogenous (see below) and motor potentials were
encountered, Herbert Vaughn coined the more general
ERP term.
Endogenous ERPs
• These are the latest (in msec) set of
waves or components in the ERP, and
are of most interest to
PSYCHOphysiologists, because they
represent psychological reactions to
externally presented but meaningful
stimuli.
• There are not that many, maybe 5 or 6
discovered to date, though folks argue
about the number. Here are a few:
Endogenous ERPs
• P300: this positive wave comes following rarely (<
40% of time) presented, meaningful stimuli. P300
was first discovered by Sutton et al. in 1965, who
presented high and low tones in a random series,
about 2-3 sec apart, in the ratio 8:2 and told Ss to
internally count the rare tones. Whether high or
low was rare, the P300 always followed the rare,
counted tone. (Why was this a critical
demonstration?)
• The stimuli were simple tone pips so indeed the
latencies were 250-300 ms But complex stimuli
can lead to 400-800 ms latencies. I.e. P300
LATENCY REPRESENTS STIMULUS COMPLEXITY,
and processing time..
More P300…
• The amplitude of P300 is directly
proportional to rareness, and directly
proportional to meaningfulness.
• You can make a stimulus meaningful by
assigning a task to it—like counting it.
Some stimuli are intrinsically
meaningful—like self-referring
information: names, birthdays, phone
numbers—and crime details which led to
our P300-based lie detector, more on
which later.
More P300…
• The scalp distribution of P300 (or
any ERP) helps to define it and
may represent cognitive
phenomena: P300 is usually
largest over Pz and smallest over
Fz, but there are many ways that
can be so:
Many curves where
Fz<Cz<Pz
ERP-defining attributes
• 1. Polarity (P300 is positive, which
for traditional reasons, we plot
down-going).
• 2. Latency (300-800 ms for P300)
• 3. Scalp distribution (Pz > Cz >Fz for
P300)
• 4. Antecedent conditions or
experimental manipulations (for
P300, rareness (“oddball stimuli”
and meaningfulness).
Another cognitive ERP: N400
• This is the response to semantic incongruity
(antecedent condition). The following 3
stimuli, pieces of one sentence, are
presented one at a time every second
• Today
• I ate
• My breakfast
• would NOT evoke N400, because the third
stimulus is congruent and unsurprising. You
DO get N400 if the 3rd stimulus is
• My motorcycle. (Get it?!)
Another cognitive ERP: CNV (this
was actually the first discovered.)
• There are 2 stimuli presented, tones, say
1-3 sec apart. The first, S1, alerts the
subject that another (S2) is coming which
will require a response, lest an aversive
event (shock) occur. The EEG will drift
negatively after S1 until S2 is presented,
whereupon it resolves. Since the stimuli
could be 2 sec apart (frequency is .5 Hz,
figure it out….), you need to pass very low
frequencies to see this ERP, called the
C(ontingent)N(egative)V(ariation).
• BTW:There must be CNV-like (anticipatory) states
during spontaneous life, so there must be delta
activity in normal waking people.
Other cognitive ERPs:
• ERN or error-related negativity. This one occurs
after you make an error, and realize it.
• RP or Recognition Potential, discovered by my old
pal, Al Rudell just a few years ago: A series of
random stimuli with a Chinese character
interspersed elicits an RP in Chinese, but not
English speakers. A series of random stimuli with
an English character interspersed elicits an RP in
English, but not Chinese speakers. Some say, “Oh
this is just a P300” (especially P300 enthusiasts),
but they are wrong, because the RP, though
positive, occurs too early (250 ms) and varies
across the scalp with stimulus location in visual
field, unlike P300.
Other cognitive ERPs:
• MMN: Mismatch Negativity. This occurs
when an occasional oddball stimulus
occurs in a series of like stimuli. Again,
people say “isn’t this like a P300?”
• The answer is no, because you need to
be awake and attentive to make a P300,
but MMN occurs in sleep or distraction.
Also, MMN has a different scalp
distribution, latency (~100-200 ms), and,
obviously, polarity (negative).
Motor ERPs
• There are also ERPs seen over
motor areas which precede actual
movements. These represent the
summed synchronous synaptic
activity of pyramidal neurons,
issuing commands to lower motor
neurons.
Another recently discovered EEG
derivative: the Event Related
Spectral Perturbation or ERSP
• This simply means a change in the
frequency of spontaneous EEG related
to an event.
• The alpha blocking pattern associated
with arousal or reticular activation is an
example.
• Measuring these was only possible with
the advent of recent, very fast
computers, because a fast Fourier
transform is necessary every few msec.
Fourier Transform
• The usual EEG is a plot of voltage as
a f(time), as you know. A Fourier
Transform of a section of EEG ( an
epoch of say 60 seconds) results in
a plot of magnitude (or energy or
power) as a f(frequency) in that
epoch. Frequency Spectra are the
resulting plots that show where in
the EEG spectrum one finds
maximum power in different
psychological states:
Fast Fourier Transforms
• It used to take weeks to do an FT
(not FFT) even with a calculator, as
matrix inversion was involved. Even
with a computer (circa 1960s) it
took several days. Then 2
programmers came up with a
shortcut, the FFT. With modern fast
computers and tools like MATLAB,
they can be done in just a few
milliseconds, so that one can go
back to the time domain and plot
frequency as f (time):
Why bother? Can’t you just look at the
EEG and see alpha blocking in the usual
way?
• Yes, the arousal response of alpha
blocking is one of the very few EEG
effects that are obvious from the
usual display of EEG voltage as a
f(time). Likewise for development of
epileptic seizure activity. But subtle
changes in cognitive or emotional
state have very subtle EEG effects
which you just can’t eyeball.
OK, but why can’t you just average EEG
from the stimulus time, as with ERPs?
• You can average ERPs because the stimulus generates
a series of synaptic events always time-locked and
phase locked to the evoking event. But it’s impossible
to predict what the spontaneous EEG is doing (e.g., is
it high or low?) when the perturbing event (stimulus) is
presented, so lack of phase locking in spontaneous
EEG prevents averaging from the stimulus, since outof-phase signals will average to a straight line. That’s
why we need the new technology to see ERSPs…. And
many new neural signs of cognitive and emotional
events have been reported (especially in Europe) with
the advent of the new methods.
OK…..New Topic!
• So…..
Shift gears!
Are ERPs and EEG neural
activities?
• Yes, of course they are, but prior to 1965, there was
considerable doubt. The great invertebrate
neurophysiologist T.H. Bullock used to visit our lab
and joke to us: “Maybe that wiggle in the EEG
simply represents a red blood cell passing near your
electrode, first presenting positivity, then negativity
from the flip side, as it passes your electrode.”
• (Just as until 2000, there was the same doubt
about fMRI, --“Is it Neural?”-- which is why you are
assigned to read:
• Heeger,D.J. et al. (2000) Spikes versus BOLD: what does
meuroimaging tell us about neuronal activity? Nature Neuroscience,
Vol 3, pp 631-633.)
..but back to EEG and ERPs
Were there data?
• Mostly negative data: people would
record from one macro-electrode on
cortical surface, and another microelectrode just outside a nearby
neuron. The spike and EEG records
would be separately recorded on
chart paper, and people would look
(in vain) for hours to find a
correspondence; by which I mean a
pattern;-- such as you only get
spikes when the EEG peaks occur,
and nothing during the EEG troughs.
New development: The C.A.T. computer
and Fox’s pussycat experiment with it.
• The Computer of Average
Transients (CAT; “transient”
means temporary signal like an
ERP) was adapted by Mary Brazier
(a Fox mentor at UCLA) for ERP
averaging. (The CAT was initially
invented to average atomic wave
phenomena.) Fox took one with
him to Michigan, then Iowa.
WOW! The CAT had 400 bytes
(not MB or GB) of memory!
• These bytes were in 100 byte
segments, so you could have 4
average erps—say to 4 different
stimuli—separately averaging in 4
separate channels. (Before the CAT,
people would simply superimpose
repeated sweeps into a fuzzy mess
on a storage oscilloscope.)
• The CAT could make one other
thing: the post stimulus time
histogram (PSTH)
PSTH is a latency distribution, much
like a normal bell curve is a grade
distribution.
Latency distribution
• After, say, 1000 trials, the numbers in each bin in
the memory string (called a “buffer”) would
contain the number of times in 1000 sweeps that
spikes occurred at that particular time.
• This is a decent approximation to the probability
of a spike at, say, 200 msec (or whatever time
bin), post stimulus. If a spike occurs on 500 trials
at 200 msec post-stimulus, the probability of a
spike at 200 ms = around .5.
• So in the PSTH, you have a sequence of spike
probabilities for various times following the
stimulus. It is indeed a probability distribution, a
distribution of firing probabilities at various
times(latencies) after the stimulus. The post
stimulus time histogram (PSTH) is like any other
histogram = distribution.
CAT “software”
• To decide whether one of the 4 CAT
channels would accumulate average
ERPs or PSTHs, the operator pulled
the side case of the CAT away and
either pulled out a 3 by 5 inch circuit
board (for the PSTH) or left it in (for
the average ERP, or AERP).
• (Some “program!”)
The Fox-O’Brien (1965) cat (the
meow type) preparation:
• They prepared a cat, under
general anesthesia, with scalp
retracted, hole drilled in skull,
through which a micro-electrode
was passed until a single neuron
was encountered. Then the cat’s
eyes were opened and maintained
with lubricant, and a light flashed
every 2 seconds.
The Fox-O’Brien (1965) cat
(the meow type) preparation
• The electrode output was divided
into 2 channels, a high pass
channel and a low pass channel.
• The former was interfaced to the
CAT PSTH generator, the latter to
an AERP generator:
The Fox-O’Brien (1965) cat (the
meow type) preparation:
They took the new CAT out of the
box, operated on and set up the
CAT connected to the cat…
• Then, the went out to dinner, as
the lights flashed in the cat’s
face, every 2 sec, and the PSTH in
one channel, and AERP in the
other continued developing.
Returning from dinner (1000+
trials), they saw the following on
the screen:
They were shocked that the PSTH looked
exactly like the AERP. (The resemblance
was better than my drawing, check the
required Fox&O’Brien paper )
• O’ Brien (grad stud) said, “Oh we must
have done something wrong,” so he hit
re-set! Fox screamed, but then calmed
down, knowing that any real
phenomenon replicates. So they started
all over, but this time sat there and
watched the PSTH and AERP develop
from a flat line to the same pair of
correlated patterns you saw in previous
slide. This was repeated with dozens
more neurons, and in many other
laboratories.
What had they really shown?
• For the first time they showed that the
moment-to-moment amplitude of the sensory
evoked ERP was a good predictor of neuronal
excitability, and so the ERP was indeed
neural.
• How did they show this? Well they showed
that the ERP voltage at any time following the
stimulus was tightly correlated with the
probability of a spike (from the PSTH) at that
time. Of course the probability of a spike is a
direct definition of excitability which is
classically defined as closeness to firing
threshold: Clearly, the closer a neuron is to
firing, the greater its excitability (Duhh!)
Were there nay-sayers?
• Always. If you discover something
important, others will snipe.
• In this case they said, “well despite
your filtering, some eeg leaked into
the PSTH channel, and some spikes
leaked into the wave channel.”
• This is patently idiotic, but was easy
to deal with. They just tapped on the
electrode, killing the cell, and
repeated the procedure.
The result:
• Of course there was no more PSTH—there
were no spikes to leak from the dead cell,
but the AERP replicated. Where did this
signal come from? Other nearby neurons-too far for the spikes to reach (thus, no
PSTH) but the PSPs from these other cells
made it fine all the way.
• I used to teach that “the brain itself passes
low frequencies better than it does high
frequencies. (Also true of sound energy—
which is why foghorns don’t tweet, but are
bassos in need of a good sub-woofer.)” Very
cute line, but proven wrong recently by
Nikos Logothetis.
This had a major
implication:
• ….which was that ERPs are not spike
envelopes—sums of spikes.
• No, ERPs are the complex sums of PSPs
from the population of neurons in the
neighborhood of the recording
electrode. But a PSP is a perfectly
respectable neural event.
• So when I asked before: “Are ERPs and
EEG neural activities?” the answer is
clearly yes. (We didn’t yet prove it for
spontaneous EEG, but that’s next.)
The Fox-Norman 1968
experiment
• This study extended Fox-O’brien
to the situation with spontaneous
EEG, rather than evoked ERPs.
• No more CAT computer, but a real
programmable one, the DEC PDP-8
that had….4KB (!) of memory. (Not
MB, let alone GB, let alone TB).
• But with spontaneous EEG, there
is no time = 0, so what to do?
That is….. Things just keep rollin’
along
Well they re-used the Fox& O’Brien
set-up with key difference
The 2 channels: 2 distributions
were developed as follows:
• (1.) Every millisecond they
sampled the EEG channel and
added the current value of the
amplitude into gradually
accumulating distribution: The got
a nice normal frequency
(“oftenness”) distribution of
amplitudes.
They were SIMULTANEOUSLY
collecting a second distribution
(2.) …by looking at both the EEG
channel AND the spike channel to
see if there was a spike occuring.
If they saw no spike, they did
nothing. If they DID see a spike,
then they incremented the counts
of those amplitudes with
simultaneous spikes. Is this
distribution larger or smaller than
the first?
Suppose we talked about
coin flips:
• How does the frequency of head
tosses compare to the frequency
of all tosses, heads & tails?
(Duhhhhhh!)

similar documents