bold imaging

Brain energy use, control of blood flow, and the basis of BOLD signals

David AttwellUniversity College London

BOLD imaging

Hariri et al. (2002) Science 297, 400

Overview• Brief review of BOLD imaging

• Coupling of neural activity to CBF, by (i) energy use or (ii) other signalling pathways

• Energy budget for cerebral cortex

• Energy use in neuronal microcircuits: cerebellum

• Local regulation of CBF by glutamate

• Global regulation of CBF by amines

• Regulation of CBF by arterioles and capillaries

• What does BOLD measure

blood vessels

HbO2Hb

O2

FLOW

basket

stellate

granule Golgi

Purkinje

input mossy fibresoutput

input climbing fibre

blood vessels

HbO2Hb

O2

FLOW

?

VOL

basket

stellate

granule Golgi

Purkinje

input mossy fibresoutput

input climbing fibre

Signalling from neurons to blood vessels

• The neuron to CBF signal is often assumed to be energy usage or energy lack (assumes CBF increases to maintain glucose/O2 delivery to neurons)

• So where does the brain use energy?

GLUTAMATE

GlialCell

3Na+

H+

K+

Na +

Post-Synaptic

Neuron

Pre-Synaptic

Neuron

Na +

Ca2+

GLU

GLN

ATP

2K3Na

ATP

2K3Na

ATP

2K3Na

ATP

distribution of ATP consumption in rat grey matterfor a mean action potential rate of 4Hz

action potentials 47%

postsynaptic receptors 34%

resting potentials 13%

3%3%

glu recycling

presynaptic Ca2+

Primates vs rodents• Primates: 3-10 times less cell density with

same synapse density (so 3-10 times more synapses/cell)

• Predicts a lower overall energy usage (54% for 10-fold - experimental value is 54%)

• Increases fraction on glutamatergic signalling (from 34% to 74%)

distribution of ATP consumption in primate grey matterfor a mean action potential rate of 4Hz

action potentials 10%

postsynaptic receptors 74%

resting potentials 3%

glu recycling 5%

presynaptic Ca2+ 7%

Energy use by neuronal microcircuits: the cerebellum as an example

basket

stellate

granule Golgi

Purkinje

input mossy fibresoutput

input climbing fibre

basket

stellate

granule Golgi

Purkinje

input mossy fibresoutput

input climbing fibre

distribution of ATP consumption in rat grey matterfor a mean action potential rate of 4Hz

action potentials 47%

postsynaptic receptors 34%

resting potentials 13%

3%3%

glu recycling

presynaptic Ca2+

resting potentials 28%

action potentials 50%

postsynaptic receptors 17%

presynaptic3%

2%glu/GABA recycling

cerebral cortex cerebellar cortex

basket

stellate

granule Golgi

Purkinje

input mossy fibresoutput

input climbing fibre

Predicted total ATP usage: 26.6 moles/g/min

Measured: 20 moles/g/min (Sokoloff & Clarke in anaesthetized albino rats)

0

5x109

10x109

15x109

20x109

basket

stellate

granule Golgi

Purkinje

input mossy fibresoutput

input climbing fibre

Purkin

je

bask

et/st

ellat

e

Golgi

gran

ule ce

ll

mos

sy fib

re

clim

bing

fibre

Bergm

ann

ATP/sec/cell

astro

cyte

0

5x109

10x109

15x109

20x109

basket

stellate

granule Golgi

Purkinje

input mossy fibresoutput

input climbing fibre

Purkin

je

bask

et/st

ellat

e

Golgi

gran

ule ce

ll

mos

sy fib

re

clim

bing

fibre

Bergm

ann

ATP/sec/cell

0

20x1018

40x1018

60x1018

80x1018

100x1018

Purkin

je

Golgibc

/sc

gran

ule ce

ll

mos

sy fib

re

clim

bing

fibre

Bergm

ann

astro

astro

cyte

ATP/sec/m2

ATP/sec/cell

resting potential

action potentials

post-synaptic

pre-syn

action potentials

post-synaptic

Granule Cell Purkinje Cell

action potentials

rp

rp

post-synaptic

glu

pre-synglu

Stellate/Basket Cell Golgi Cell

post-synaptic

rp

action potentials

ATP Usage by Subcellular Task

Firing Rate (Hz)

0 20 40 60 80 100

AT

P/m

2/s

ec

0

20

40

60

80

100

120

140

160

180

granule cells

mossy fibres

Purkinje cell (simple spikes)

Effect of altering firing rate in a single cell type

Energy use by neuronal microcircuits: the cerebellum as an example

(1) Most energy goes on granule cells re-mapping the sensory and motor command input arriving on the mossy fibres into a sparse coded representation used by the Purkinje cells to retrieve motor output patterns

basket

stellate

granule Golgi

Purkinje

input mossy fibresoutput

input climbing fibre

Energy use by neuronal microcircuits: the cerebellum as an example

(1) Most energy goes on granule cells re-mapping the sensory and motor command input arriving on the mossy fibres into a sparse coded representation used by the Purkinje cells to retrieve motor output patterns

(2) 1011 ATP molecules are used per second to be able to retrieve 5kB of information from each Purkinje cell (which can store 40,000 input-output associations), or 2x1016 ATP/GB/s = (3.3x10-8moles/sec)x31kJ = 1mW/GB. Computer hard disks now use ~5mW/GB

basket

stellate

granule Golgi

Purkinje

input mossy fibresoutput

input climbing fibre

How is blood flow controlled?

ML

GLPC

Does an energy-lack signal increase blood flow?

• When [ATP] (or [O2] or [glucose]) falls, or [CO2] or [H+] or [lactate] rises, does that make blood flow increase?

• In other words, do BOLD signals reflect the presence of a feedback system to conserve energy supply?

blood vessels

HbO2Hb

O2

FLOW

energy lack?

VOL

basket

stellate

granule Golgi

Purkinje

input mossy fibresoutput

input climbing fibre

What controls cerebral blood flow during brain activation?

• Not glucose lack (Powers et al., 1996)

• Not oxygen lack (Mintun et al., 2001)

• Not CO2 evoked pH change (pHo goes alkaline due to CBF increase removing CO2: Astrup et al., 1978; Pinard et al., 1984)

• So CBF is not driven directly by energy lack maintaining O2/glucose delivery to neurons and keeping [ATP] high Powers et al., 1996

What controls cerebral blood flow during brain activation?

• CBF is not driven by energy lack• Not the spike rate of principal neurons (Mathiesen et

al., 1998; Lauritzen 2001)• BOLD correlates (slightly!) better with synaptic

field potentials than spike output (Logothetis et al., 2001)

• So does synaptic signalling control CBF (i.e. is it a feedforward, rather than a feedback, system)?

Feedforward vs feedback control of CBF

Neuronal activity

Neuronal activity

Energy falls Increase CBF

Increase CBF Energy supplied

-

Negative feedback

Feedforward

GLUTAMATE

GlialCell

3Na+

H+

K+

Na +

Post-Synaptic

Neuron

Pre-Synaptic

Neuron

+NaCa 2+

GLU

GLN

ATP

2K3Na

ATP

2K3Na

ATP

2K3Na

ATP

PLA2

NOS

AA,PG

NO

Ca2+

PLA2

Glutamate is responsible for cerebellar CBF increase

Purkinje cell spikes

CBF

Parallel fibrestimulation

Climbing fibrestimulation

Matthiesen et al., 1998

CBF

blood vessels

HbO2Hb

O2

FLOW

Glutamate (vianeurons and glia)

VOL

basket

stellate

granule Golgi

Purkinje

input mossy fibresoutput

input climbing fibre

Glutamate controls CBF and BOLD signals

• Energy calculations implicate postsynaptic currents as the main energy consumer - so if energy use drove BOLD signals, BOLD would reflect the release of glutamate

• In fact energy use does not drive CBF, but glutamate does - so BOLD is still likely to reflect glutamate release (via its postsynaptic actions)

What does BOLD measure?

• If BOLD signals largely reflect glutamate release:• (a) BOLD does not tell us about the spike output of an

area, and will only reflect principal cell activity if most glutamate is released onto principal cells

• (b) altered processing with no net change of glu release might produce no BOLD signal

• (c) altered glu release with no change of the spike output of an area could produce a BOLD signal

blood vessels

HbO2Hb

O2

FLOW

Glu

VOL

AMINESNA, DA, 5-HT

basket

stellate

granule Golgi

Purkinje

input mossy fibresoutput

input climbing fibre

Control of cerebral blood flow by distributed systems using amines and ACh

• Dopaminergic neurons (from VTA) innervate microvessels - DA constricts (Krimer et al., 1998): D1,2,4,5

• Noradrenergic neurons (from locus coeruleus) also constrict microvessels (Raichle et al., 1975): 2

• Serotoninergic neurons (from raphe) constrict cerebral arteries and microvessels (Cohen et al., 1996): 5-HT1,2

• All are wide ranging systems - control CBF globally

Smooth Muscle vs Pericytes

blo

od

flo

w

capillary

sm

oo

th m

us

cle

end

oth

elia

l cel

ls

10 µm

SM

10 µm

5 µm

s

p 5 µm

s p

pericytes

Smooth Muscle vs Pericytes

blo

od

flo

w

capillary

sm

oo

th m

us

cle

end

oth

elia

l cel

ls

10 µm

SM

10 µm

5 µm

s

p 5 µm

s p

pericytes

65% of noradrenergic innervation is of capillaries,not arterioles

390s185s

b

295s

c d

10

8

6

4

2

0

diam

eter

(m

icro

ns)

4003002001000

time (s)

1mM Glu1M NA

70s

a

o

•

Peppiatt, Howarth, Auger & Attwell, unpublished

Noradrenaline constricts and glutamate dilates cerebellar capillaries

Pericytes communicate with each other and could communicate from neurons near capillariesto precapillary arterioles

Implications of control of CBF by aminesfor neuropsychiatric imaging

• Clinical disorders often involve disruption of amine function (schizophrenia, Parkinson’s, ADHD)

• In imaging we would like a change in BOLD signals to imply an effect of the amine disorder on cortical processing

• If amines control CBF, altered amine function may alter the relation between neural activity and BOLD signals (extreme example: amine depletion maximally dilates vessels, so no further dilation or BOLD signal possible)

• Consequently altered BOLD signals may just reflect altered control of CBF, and provide no information on neural processing

blood vessels

HbO2Hb

O2

FLOW

VOL

AMINESNA, DA, 5-HT

basket

stellate

granule Golgi

Purkinje

input mossy fibresoutput

input climbing fibre

Glu

BOLD imaging

Hariri et al. (2002) Science 297, 400

Conclusions• In primates, most of the brain’s energy goes on postsynaptic currents (and

action potentials)

• CBF changes and BOLD aren’t driven by O2/glucose lack nor by CO2 production, so are not driven by energy lack

• CBF changes and BOLD don’t correlate well with spiking• Glutamate controls local CBF so BOLD signals will reflect glutamatergic

signalling• Amines control CBF more globally - could confound studies on amine-

related diseases• CONCLUSION: to interpret BOLD signals you need to consider the

neural wiring of the area being studied

Collaborators

Clare Howarth

Claire Peppiatt

Céline Auger

Simon Laughlin