The physiology of the BOLD signal
Klaas Enno Stephan
Laboratory for Social and Neural Systems Research
Institute for Empirical Research in Economics
University of Zurich
With many thanks for slides & images to:
Functional Imaging Laboratory (FIL)
Wellcome Trust Centre for Neuroimaging
University College London
Ralf Deichmann
Meike Grol
Tobias Sommer
Methods & models for fMRI data analysis in neuroeconomics
April 2010
Ultrashort introduction to MRI physics
• Step 1: Place an object/subject in a big magnet
• Step 2: Apply radio waves
• Step 3: Measure emitted radio waves
Step 1: Place subject in a big magnet
Spin = rotation of a proton
around some axis
Movement of a positive
charge → magnetic moment
Images: www.fmri4newbies.com
When you put any material in an MRI
scanner, the protons align with the
direction of the magnetic field.
Image: Ralf Deichmann
Step 2: Apply radio waves
T1
T2
When you apply radio waves (RF
pulse) at the appropriate frequency
(Larmor frequency), you can change
the orientation of the spins as the
protons absorb energy.
Images: Ralf Deichmann
After you turn off the RF pulse, as the
protons return to their original
orientations, they emit energy in the
form of radio waves.
Step 3: Measure emitted radio waves
T1 = time constant of how quickly the
protons realign with the magnetic field
fat has high
signal  bright
CSF has low
signal  dark
Images:
fmri4newbies.com
T1-WEIGHTED ANATOMICAL IMAGE
T2 = time constant of how quickly the
protons emit energy when recovering to
equilibrium
fat has low
signal  dark
CSF has high
signal  bright
T2-WEIGHTED ANATOMICAL IMAGE
T2* weighted images
• Two factors contribute to the decay of transverse magnetization:
1) molecular interactions → dephasing of spins
2) local inhomogeneities of the magnetic field
• The combined time constant is called T2*.
• fMRI uses acquisition techniques (e.g. EPI) that are sensitive to
changes in T2*.
The general principle of MRI:
– excite spins in static field by RF pulses & detect the emitted RF
– use an acquisition technique that is sensitive to local differences in
T1, T2 or T2*
– construct a spatial image
Functional MRI (fMRI)
• Uses echo planar imaging (EPI) for fast
acquisition of T2*-weighted images.
• Spatial resolution:
– 3 mm
(standard 1.5 T scanner)
– < 200 μm (high-field systems)
EPI
(T2*)
• Sampling speed:
– 1 slice: 50-100 ms
• Problems:
dropout
– distortion and signal dropouts in certain regions
– sensitive to head motion of subjects during
scanning
• Requires spatial pre-processing and statistical
analysis.
But what is it that makes T2*
weighted images “functional”?
T1
The BOLD contrast
BOLD (Blood Oxygenation Level Dependent) contrast =
measures inhomogeneities in the magnetic field due to
changes in the level of O2 in the blood
B0
Oxygenated hemoglobine:
Diamagnetic (non-magnetic)
 No signal loss…
Deoxygenated hemoglobine:
Paramagnetic (magnetic)
 signal loss !
Images: Huettel, Song & McCarthy 2004, Functional Magnetic Resonance Imaging
The BOLD contrast
deoxy-Hb/oxy-Hb 
rCBF 
?
neurovascular
coupling
neuronal metabolism 
?
synaptic activity 
D’Esposito et al. 2003
Due to an over-compensatory increase of rCBF, increased neural
activity decreases the relative amount of deoxy-Hb
 higher T2* signal intensity
The BOLD contrast
REST
neural activity   blood flow   oxyhemoglobin   T2*   MR signal
ACTIVITY
Source: Jorge Jovicich, fMRIB Brief Introduction to fMRI
The temporal properties of the BOLD signal
• sometimes shows
initial undershoot
Peak
• peaks after 4-6 secs
• back to baseline
after approx. 30
secs
• can vary between
regions and
subjects
Brief
Stimulus
Undershoot
Initial
Undershoot
u
BOLD signal is a nonlinear
function of rCBF
stimulus functions
t
m

dx 
  A   u j B ( j )  x  Cu
dt 
j 1

neural state
equation
0.4
0.2
vasodilato
ry signal
0
s  x  s  γ( f  1)
f
0
2
4
6
8
10
12
s
s
N
RBM N,  = 1
CBM ,  = 1
N
RBM ,  = 2
1
flow induction(rCBF)
0.5
f  s
hemodynamic
state
equations
N
CBM N,  = 2
0
f
Balloon model
changes involume
τv  f  v1/α
v
 ( q, v ) 
14
RBM N,  = 0.5
CBM ,  = 0.5
v
0
2
4
6
8
10
12
14
0
2
4
6
8
10
12
14
0.2
changes indHb
τq  f E ( f,E0 ) qE0  v1/α q/v
q
0
-0.2
-0.4
-0.6
S


 q
 V0 k1 1  q   k2 1    k3 1  v 
S0
 v


k1  4.30 E0TE
k2  r0 E0TE
k3  1  
BOLD signal
change equation
Stephan et al. 2007, NeuroImage
Three important questions
1. Is the BOLD signal more strongly related to
neuronal action potentials or to local field
potentials (LFP)?
2. Does the BOLD signal reflect energy
demands or synaptic activity?
3. What does a negative BOLD signal mean?
Neurophysiological basis of the BOLD
signal: soma or synapse?
BOLD & action potentials
Red curve: “average firing
rate in monkey V1, as a
function of contrast,
estimated from a large
database of microelectrode
recordings (333 neurons).”
Heeger et al. 2000, Nat. Neurosci.
Rees et al. 2000, Nat. Neurosci.
In early experiments comparing human BOLD signals and monkey
electrophysiological data, BOLD signals were found to be
correlated with action potentials.
Action potentials vs. postsynaptic activity
Local Field Potentials (LFP)
• reflect summation of post-synaptic
potentials
Multi-Unit Activity (MUA)
• reflects action potentials/spiking
Logothetis et al., 2001, Nature
Logothetis et al. (2001)
• combined BOLD fMRI and
electrophysiological recordings
• found that BOLD activity is more
closely related to LFPs than MUA
BOLD & LFPs
blue: LFP
red: BOLD
grey: predicted BOLD
Logothetis & Wandell 2004, Ann. Rev. Physiol.
Dissociation between action potentials and rCBF
• GABAA antagonist picrotoxine
increased spiking activity without
increase in rCBF...
• ... and without disturbing
neurovascular coupling per se
Thomsen et al. 2004, J. Physiol.
 rCBF-increase can be
independent from spiking
activity, but seems to be
always correlated to LFPs
Lauritzen et al. 2003
Current conclusion: BOLD signal seems to be more
strongly correlated to postsynaptic activity
• BOLD can be correlated both to action potentials and to
postsynaptic actitivity (as indexed by LFPs)
• Indeed, in many cases action potentials and LFPs are
themselves highly correlated.
• rCBF-increase can be independent from spiking activity,
but so far no case has been found where it was
independent of LFPs.
• This justifies the (present) conclusion that BOLD more
strongly reflects the input to a neuronal population as well
as its intrinsic processing, rather than its spiking output.
Lauritzen 2005, Nat. Neurosci. Rev.
Three important questions
1. Is the BOLD signal more strongly related to
neuronal action potentials or to local field
potentials (LFP)?
2. Does the BOLD signal reflect energy
demands or synaptic activity?
3. What does a negative BOLD signal mean?
Is the BOLD signal driven by energy demands
or synaptic processes?
deoxy-Hb/oxy-Hb 
rCBF 
?
neurovascular
coupling
neuronal metabolism 
?
synaptic activity 
D’Esposito et al. 2003
Localisation of neuronal energy consumption
Salt loading in rats and 2-deoxyglucose
mapping
→
glucose utilization in the posterior
pituitary but not in paraventricular
and supraoptic nuclei (which
release ADH & oxytocin at their
axonal endings in the post. pituitary)
→
neuronal energy consumption takes
place at the synapses, not at the
cell body
Compatible with findings on BOLD relation to
LFPs!
Schwartz et al. 1979, Science
But does not tell us whether BOLD induction
is due to energy demands (feedback) or
synaptic processes (feedforward)...
Energetic consequences of postsynaptic activity
Glutamate reuptake and recycling by astrocytes
requires glucose metabolism
Courtesy: Tobias Sommer
Also, ATP needed for restoring ionic
gradients, transmitter reuptake etc.
Attwell & Iadecola 2002, TINS.
Increased rCBF due to lack of energy?
1. Initial dip:
possible to get more O2
from the blood without
increasing rCBF (which
happens later in time).
Friston et al. 2000, NeuroImage
2. Controversial reports on
whether or not there is
compensatory increase in
blood flow during hypoxia
(Mintun et al. 2001;
Gjedde 2002).
rCBF map during
visual stimulation
under normal
conditions
rCBF map during
visual stimulation
under hypoxia
Mintun et al. 2001, PNAS
Blood flow might be directly driven by
excitatory postsynaptic processes
Glutamatergic synapses:
A feedforward system for eliciting the BOLD signal?
Lauritzen 2005, Nat. Neurosci. Rev.
Forward control of blood flow
Courtesy: Marieke Scholvinck
O2 levels determine
whether synaptic activity
leads to arteriolar
vasodilation or
vasoconstriction (via
prostaglandines)
Gordon et al. 2008, Nature
Three important questions
1. Is the BOLD signal more strongly related to
neuronal action potentials or to local field
potentials (LFP)?
2. Does the BOLD signal reflect energy
demands or synaptic activity?
3. What does a negative BOLD signal mean?
Negative BOLD is correlated with decreases
in LFPs
positive BOLD
positive BOLD
Shmuel et al. 2006, Nat. Neurosci.
Impact of inhibitory postsynaptic potentials
(IPSPs) on blood flow
Lauritzen 2005, Nat. Neurosci. Rev.
Negative BOLD signals due to IPSPs?
Lauritzen 2005, Nat. Neurosci. Rev.
Potential physiological influences on BOLD
cerebrovascular
disease
medications
structural lesions
(compression)
blood
flow
blood
volume
hypoxia
autoregulation
(vasodilation)
volume status
hypercapnia
anesthesia/sleep
BOLD
contrast
biophysical effects
anemia
smoking
oxygen
utilization
degenerative
disease
Drug effects
Nitrates (against coronary
heart disease)
Analgetics (NSAIDs)
Summary
•
The BOLD signal seems to be more strongly
related to LFPs than to spiking activity.
•
The BOLD signal seems to reflect the input to a
neuronal population as well as its intrinsic
processing, not the outputs from that population.
•
Blood flow seems to be controlled in a forward
fashion by postsynaptic processes leading to the
release of vasodilators.
•
Negative BOLD signals may result from IPSPs.
•
Various drugs can interfere with the BOLD
response.
Thank you