30. A. D. Craig, K. Chen, D. Bandy, E. M. Reiman, Nature
Neurosci. 3, 184 (2000).
31. A. Ploghaus et al., Science 284, 1979 (1999).
32. P. Petrovic, E. Kalso, K. M. Petersson, M. Ingvar, Science
295, 1737 (2002).
33. A. D. Craig, Curr. Opin. Neurobiol. 13, 500 (2003).
34. J. S. Winston, B. A. Strange, J. O’Doherty, R. J. Dolan,
Nature Neurosci. 5, 277 (2002).
35. A. J. Blood, R. J. Zatorre, Proc. Natl. Acad. Sci. U.S.A.
98, 11818 (2001).
36. H. Olausson et al., Nature Neurosci. 5, 900 (2002).
37. N. I. Eisenberger, M. D. Lieberman, K. D. Williams,
Science 302, 290 (2003).
38. P. S. Churchland, Science 296, 308 (2002).
39. A. R. Damasio, Descartes’ Error (Putnam, New York,
1994).
40. This work was supported by a personal grant for T.S.
from the German Academy of Natural Sciences
Leopoldina, Halle, with the grant BMBF-LPD 9901/8-
73 from the Ministry of Education and Science, by the
Wellcome Department of Imaging Neuroscience, and
by the University College of London. R.J.D. and C.D.F.
are in receipt of Wellcome Trust program grants. We
thank S. Kiebel, J. Schultz, K. Wiech, R. Kalisch, P.
Aston, E. Featherstone, and P. Allen for their help.
Supporting Online Material
www.sciencemag.org/cgi/content/full/303/5661/1157/
DC1
Materials and Methods
Fig. S1
Tables S1 to S5
References
11 November 2003; accepted 23 December 2003
Placebo-Induced Changes in fMRI
in the Anticipation and
Experience of Pain
Tor D. Wager,
1
*† James K. Rilling,
2
Edward E. Smith,
1
Alex Sokolik,
3
Kenneth L. Casey,
3
Richard J. Davidson,
4
Stephen M. Kosslyn,
5
Robert M. Rose,
6
Jonathan D. Cohen
2,7
The experience of pain arises from both physiological and psychological factors,
including one’s beliefs and expectations. Thus, placebo treatments that have no
intrinsic pharmacological effects may produce analgesia by altering expecta-
tions. However, controversy exists regarding whether placebos alter sensory
pain transmission, pain affect, or simply produce compliance with the sugges-
tions of investigators. In two functional magnetic resonance imaging (fMRI)
experiments, we found that placebo analgesia was related to decreased brain
activity in pain-sensitive brain regions, including the thalamus, insula, and
anterior cingulate cortex, and was associated with increased activity during
anticipation of pain in the prefrontal cortex, providing evidence that placebos
alter the experience of pain.
The idea that sensory experience is shaped
by one’s attitudes and beliefs has gained
currency among psychologists, physicians,
and the general public. Perhaps nowhere is
this more apparent than in our ability to
modulate pain perception. A special case of
this phenomenon is placebo analgesia, in
which the mere belief that one is receiving
an effective analgesic treatment can reduce
pain (1–5 ). Recently, some researchers
have attributed placebo effects to response
bias and/or to publication biases (6), which
raises the issue of whether placebo treat-
ments actually influence the sensory, affec-
tive, and cognitive processes that mediate
the experience of pain.
One important piece of evidence that
placebo effects are not simply due to re-
sponse or publication bias is that such ef-
fects can be reversed by the mu-opioid
antagonist naloxone (2, 3, 7 ), suggesting
that some kinds of placebo effects may be
mediated by the opioid system. However,
naloxone has also been shown to produce
hyperalgesia independent of placebo, in
some cases offsetting rather than blocking
the effects of placebo analgesia (8). Al-
though pharmacological blockade provides
suggestive evidence regarding the neuro-
chemical mechanisms mediating placebo
effects, such data do not illuminate the
nature of the information-processing sys-
tem that gives rise to such effects. Neuro-
imaging data can provide complementary
evidence of how pain processing in the
brain is affected by placebos and about the
time course of pain processing. Identifying
placebo-induced changes in brain activity
in regions associated with sensory, affec-
tive, and cognitive pain processing (9) may
provide insight into which components of
pain processing are affected by placebo. In
addition, identifying changes that occur at par-
ticular times—in anticipation of pain, early or
late during pain processing—may shed light on
how cognitive systems mediating expectancy
interact with pain and opioid systems.
In two functional magnetic resonance
imaging (fMRI) experiments (n ⫽ 24 and
n ⫽ 23), we examined two hypotheses re-
garding the psychological and neural mech-
anisms that underlie placebo analgesia. Our
first hypothesis was that if placebo manip-
ulations reduce the experience of pain,
pain-responsive regions of the brain should
show a reduced fMRI blood oxygen level–
dependent (BOLD) signal (a measure relat-
ed to neural activity) during pain. [Pain-
responsive regions, or the “pain matrix,”
include thalamus, somatosensory cortex,
insula, and anterior cingulate cortex (10–
14).] Our second hypothesis was that pla-
cebo modulates activity of the pain matrix
by creating expectations for pain relief,
which in turn inhibit activity in pain-
processing regions. Converging evidence
suggests that the prefrontal cortex (PFC),
the dorsolateral aspect (DLPFC) in partic-
ular, acts to maintain and appropriately up-
date internal representations of goals and
expectations, which modulate processing in
other brain areas (15, 16). Thus, stronger
PFC activation during the anticipation of
pain should correlate with greater placebo-
induced pain relief as reported by participants
and greater placebo-induced reductions in neu-
ral activity within pain regions (17 ).
Placebo reduces reported pain and
brain activity in Study 1 (shock pain). The
design of Study 1 is illustrated in Fig. 1A
(see the figure legend for a description)
(18). First, to confirm that application of
shock elicited a neural response in pain-
related areas, we compared brain activity in
the intense shock versus no shock condi-
tions. This revealed activation of the classic
pain matrix (11, 14, 19, 20), including thal-
amus, primary somatosensory cortex/
primary motor cortex (S1/M1), secondary
somatosensory cortex (SII), midbrain, an-
terior insula, anterior cingulate cortex
(ACC), ventrolateral prefrontal cortex, and
cerebellum (fig. S1). As expected, activa-
tions in thalamus, S1, SII, and M1 were
larger in the left hemisphere, contralateral
to the wrist where shocks were applied,
whereas cerebellar activation was ipsilater-
al, although some bilateral activation was
observed in each of these areas. We also
1
Department of Psychology, University of Michi-
gan, 525 East University, Ann Arbor, MI 48109–
1109, USA.
2
Center for the Study of Brain, Mind and
Behavior, Princeton University, Princeton, NJ
08544, USA.
3
Department of Neurology, Veterans
Affairs Medical Center, University of Michigan, Ann
Arbor, MI 48109, USA.
4
Department of Psychology,
University of Wisconsin, Madison, WI 53706, USA.
5
Department of Psychology, Harvard University,
Cambridge, MA 02138, USA.
6
Mind Brain Body and
Health Initiative, University of Texas Medical
Branch, Galveston, TX 77555, USA.
7
Department of
Psychology, Princeton University, Princeton, NJ
08544, USA.
*To whom correspondence should be addressed. E-mail:
torw@umich.edu, tor@paradox.psych.columbia.edu
†Present address: Department of Psychology, Colum-
bia University, 1190 Amsterdam Avenue, New York,
NY 10027, USA.
R ESEARCH A RTICLES
20 FEBRUARY 2004 VOL 303 SCIENCE www.sciencemag.org1162
compared intense shock with mild shock to
determine which pain regions responded
more specifically to the painful aspects of
the stimulus, which produced a similar net-
work of activated regions (Fig. 1C and
table S1). These regions, which we refer to
as “pain-responsive” regions because they
track the magnitude of painful stimulation
(10), constituted the pain-sensitive regions
of interest (ROIs) in which we expected to
find placebo effects (21). Anticipation of
shock activated contralateral S1, SII, M1,
and dorsal amygdala (fig. S2).
Turning to placebo effects, we first as-
sessed the placebo effect based on partici-
pants’ reports, calculated as the difference
between the average rating of intense
shocks in the placebo and control condi-
tions. Reported pain was greater for control
than for placebo conditions across partici-
pants (x¯ ⫽ 0.21, ⫽0.47, t(23) ⫽ 2.20,
P ⬍ 0.05), indicating a significant analge-
sic effect of the placebo. However, the
relatively high variability in the placebo
response across participants (only 8 of the
24 participants both showed a placebo ef-
fect in our measure and reported some pain
relief in a postsession debriefing) allowed
us to examine correlations between mea-
sures of reported pain relief and corre-
sponding neural responses, as discussed be-
low. In contrast to intense shocks, we found
no placebo effect for the ratings of mild
shocks for the group as a whole [x¯ ⫽ 0.04,
⫽0.54, t(23) ⫽ 0.36], and thus will not
further discuss findings from this condition.
Our first prediction was that the placebo
treatment would attenuate activation within
pain ROIs. We found that the magnitude of
the reduction between control and placebo
trials in reported pain (hereafter referred to
as control ⬎ placebo, a measure of experi-
enced placebo analgesia) correlated with
the magnitude of reduction in neural activ-
ity during the shock period (control ⬎ pla-
cebo, a measure of placebo analgesia in the
brain) in pain-responsive portions of sever-
al brain structures. These structures includ-
ed the rostral anterior cingulate cortex
(rACC) at the junction between rostral and
caudal ACC (r ⫽ 0.66), contralateral insula
(r ⫽ 0.59), and the contralateral thalamus
(r ⫽ 0.53) (22). These findings were all
significant at P ⬍ 0.005, and are shown in
Fig. 2A, C, and E, respectively. [All brain-
behavior correlations we report compare
the magnitudes of placebo effects (control–
placebo) on reported pain with magnitudes
of placebo effects in neural activity (control
– placebo).] Because the thalamus is the
major cortical relay for afferent pain fibers,
this correlation is predicted by theories of
placebo that hypothesize inhibition of af-
ferent sensory pain transmission (23). The
insula has been associated with both the
sensory-discriminative and affective com-
ponents of pain (10, 24 ), and the rACC has
been shown to track changes in reported
pain induced by hypnosis (25 ), at coordi-
nates [7 20 29], 5 mm from the center of
our activation.
Placebo increases prefrontal activity in
anticipation of painful shock. To evaluate
our second hypothesis—that expectation of
pain relief is represented in PFC and medi-
ates placebo analgesia—we examined cor-
relations between reported placebo effects
in ratings (control ⬎ placebo) and fMRI
activity in the anticipation period (place-
bo ⬎ control). We restricted our analysis to
DLPFC and orbitofrontal cortex (OFC),
Fig. 1. (A) Time course of a single trial in Study 1. Twenty-four
participants were scanned by fMRI as they received painful and
nonpainful electric shocks to their right wrist. We modeled our
design after a study by Ploghaus et al.(58), which allowed us to
distinguish the brain’s response to pain from its anticipation of
pain. The experiment consisted of five blocks of 15 trials. Each
trial lasted 30 s and began with a 3-s warning cue—a red or blue
spiral icon—that indicated whether the upcoming shock would
be intense or mild, respectively (18). An ensuing anticipation
epoch varied between 3 and 12 s, and was followed by a 6-s
epoch of either intense or mild shock. After the shock, partici-
pants rated the intensity of the shock on a 10-point scale,
followed by a variable rest period until the end of the trial.
Shocks were randomly omitted on one-third of all trials, in order
to increase the number of total test trials without compromis-
ing expectations regarding pain. Participants were told that they
were taking part in a study of brain responses to a new analgesic
cream. In the first block of trials, participants received shocks
without any treatment. After Block 1, an investigator applied a
skin cream to the participant’s right wrist with the participant
still in the scanner. Half the participants were told that this was
an analgesic cream that would reduce but not eliminate the pain
of the shocks. After Blocks 2 and 3 were completed in this
placebo condition, the cream was removed and the same cream
was reapplied. Then participants were told that the cream was
actually a different, ineffective cream needed as a control. Participants
then completed Blocks 4 and 5. For the other half of the participants, we
reversed the order of placebo and control conditions. Our measures of the
placebo effect were the differences in reported ratings of pain and
regional brain activity in the control versus placebo conditions (control –
placebo in both behavior and brain). During pain and rest periods,
participants saw a fixation cross. (B) Time course of a trial in Study 2. The
design was similar to that of Study 1, with the following differences. The
cue was the words “Get ready!” in red letters (1 s duration). A painful
thermal stimulus was applied for 20 s (17 s peak, 1.5 s ramp up/down),
allowing us to analyze pain responses in three separate segments (early,
peak, and late). Different patches of skin on the left forearm (38) were
treated with placebo and control topical creams (which were identical).
Thermal stimuli were applied to these patches of skin in three phases.
During the calibration phase, the stimulus was varied to identify temper-
atures corresponding to reported pain levels of 2, 5, and 8 on a 10-point
scale (1 ⫽ just painful; 10 ⫽ unbearable pain) for each participant (59).
This was followed by the manipulation phase, included to enhance
participants’ expectations of pain relief and thereby increase placebo
responding. In this phase, pain was surreptitiously reduced in the placebo
condition (5, 60). During one block of trials the stimuli were applied to
the placebo-treated patch of skin, and during another block the stimuli
were applied to the control-treated patch (order counterbalanced across
participants). Participants were told that all stimuli were at level 8.
However, they were administered at level 2 in the placebo-treated patch
and at level 8 in the control-treated patch. Finally, during the test phase,
two additional blocks of stimuli were administered to placebo- and
control-treated patches of skin. Again, participants were told these
were at level 8, but both were delivered at level 5, in keeping with the
paradigm used in (5). Because the stimuli were identical, any differ-
ences in reported pain (control – placebo) during this phase are
attributable to placebo effects. (C) Pain-responsive regions, identified
by their significance in (intense – mild stimulation) contrasts in Study
1 or Study 2. These regions were ROIs in which we looked for placebo
effects. ACC: anterior cingulate; rACC: rostral anterior cingulate; SII:
secondary somatosensory cortex; INS: insula; TH: thalamus.
R ESEARCH A RTICLES
www.sciencemag.org SCIENCE VOL 303 20 FEBRUARY 2004 1163
based on our hypothesis. OFC is thought to
play an important role in configuring con-
trol mechanisms and learning based on re-
ward information (26–31). Regions within
bilateral DLPFC showed significant corre-
lations [r ⫽ 0.62 within both left (L) and
right (R) hemispheres] (32). Regions within
bilateral OFC showed similar correlations
(OFC; r ⫽ 0.65/0.76 in L/R hemispheres,
respectively) (Fig. 3B) (32). Previous re-
search (33) suggests that rACC may also
serve as a control region, because it was
activated in placebo relative to control con-
ditions. Our data support this notion, as we
also found correlations of the form de-
scribed above for rACC (32). Correlations
between reported placebo effects and pre-
frontal activation are consistent with the
hypothesis that regions involved in gener-
ating and maintaining expectations contrib-
ute to placebo-related analgesia.
We also tested for correlations between
anticipation activity in expectancy areas
and pain activity in pain regions. Negative
correlations would support the view that
prefrontal activity is an antecedent to re-
duction in pain. Placebo-induced increases
in DLPFC were correlated with placebo-
induced reductions during pain in several
regions: (i) contralateral thalamus, r ⫽
⫺0.56/⫺0.38 for L and R DLPFC; correla-
tions whose absolute value is greater than
0.4 are significant at P ⬍ 0.05; (ii) insula,
r ⫽⫺0.59/⫺0.26 for L and R DLPFC; and
(iii) rACC, r ⫽⫺0.44/⫺0.45 for L and R
DLPFC. Similar correlations were observed
between placebo increases in OFC and pla-
cebo reductions in pain activity: (i) thala-
mus, r ⫽⫺0.52/⫺0.63 for L and R OFC;
(ii) insula, r ⫽⫺0.61/⫺0.56 for L and R
OFC; (iii) rACC: r ⫽⫺0.65/⫺0.70 for L
and R OFC.
We also found increased activity (place-
bo ⬎ control) during the anticipation peri-
od in the midbrain, in the vicinity of the
periaqueductal grey (PAG), which contains
a high concentration of opiate neurons with
descending spinal efferents (23, 34). Mid-
brain placebo increases (placebo ⬎ control,
at coordinates [10 –26 ⫺14]) (35), were
positively correlated with both reported
placebo effects (control ⬎ placebo) and
brain placebo effects (control ⬎ placebo) in
some pain areas (r ⫽ 0.47 for thalamus and
r ⫽ 0.48 for rACC) (36 ). Furthermore,
midbrain placebo ⬎ control activity was
correlated with anticipation-period activa-
tion (placebo ⬎ control) of the right PFC
(r ⫽ 0.51) and OFC (r ⫽ 0.48/0.39 for L
and R hemispheres) (37).
Placebo reduces reported pain and brain
activity in Study 2 (thermal pain). In Study 2
we used a stronger placebo induction, a
different pain modality, and an experimen-
tal design that allowed us to analyze the
time course of placebo-related effects dur-
ing the pain epoch. These manipulations
provided greater power to test the influence
of the placebo manipulation on activation
of the pain matrix, and to test further hy-
potheses regarding the mechanisms of pla-
cebo action. For example, if placebo can
affect the pain matrix through expectation
alone, we expect such effects to occur early
during pain, whereas if placebo effects also
involve direct (e.g., opioid release) or indi-
rect (cognitive reappraisal) processes that
evolve over time (e.g., in response to the
sensory stimulus), we expect them to occur
later during pain stimulation. The sequence
of events on each trial is shown in Fig. 1B,
and other aspects of the design are dis-
cussed in the figure legend (18, 38, 39).
Fifty participants were studied using the
procedures described above before fMRI
scanning, including a manipulation phase
designed to enhance placebo-related expec-
tations. On average, placebo resulted in a
22% decrease in reported pain during the
test phase, with 72% of participants show-
ing effects in the expected direction
[t(49) ⫽ 5.87, P ⬍ 0.0001] (fig. S3A). This
high rate of response confirmed that we had
effectively enhanced participants’ belief in
the placebo. Placebo responders were invit-
ed to return for fMRI scanning (40, 41).
As in Study 1, we found significant pain
activation in expected regions (averaging
over control and placebo), shown in red in
fig. S1. These included bilateral insula,
S1/M1, SII, thalamus, and anterior and dor-
solateral PFC, as well as ACC, medial PFC,
and cerebellar vermis. Comparing intense
(level 8) pain with mild (level 2) pain
during the manipulation phase produced
activations within all of these regions (table
S1). We used these regions to test for pla-
cebo effects.
The results provided further support for
our first hypothesis, that placebo would
reduce activity in pain-responsive areas.
We expected main effects of placebo (con-
trol ⬎ placebo) in Study 2, because only
placebo responders were selected as partic-
ipants; thus, the range of the placebo re-
sponse was restricted in Study 2, although
this selection procedure does not preclude
finding correlations as well. As in Study 1,
contralateral thalamus, anterior insula, and
rACC all showed significant placebo ef-
fects. In the rACC pain region (Fig. 2B),
reported placebo effects (control ⬎ place-
bo) were correlated with neural placebo
effects (control ⬎ placebo) in the early heat
period (r ⫽ 0.58) (42). In contralateral
Fig. 2. Pain regions showing corre-
lations between placebo effects in
reported pain (control ⫺ placebo)
and placebo effects in neural pain
(control ⫺ placebo). (A) Rostral
anterior cingulate (rACC) effects in
Study 1. (B) rACC effects in early
heat in Study 2. (C) Contralateral
(left) insula (INS) in Study 1, z ⫽
⫺4 mm. (D) Contralateral (right)
INS effects in Study 2, z ⫽⫺4
mm. The parahippocampal cortex
(PHCP) activations extended into
the basal forebrain and are contig-
uous with thalamic activations;
however, only thalamic activations
are in pain-sensitive regions. (E)
Contralateral INS and thalamus
(TH) in Study 1, z ⫽ 6. (F) Con-
tralateral INS and TH in Study 2,
z ⫽ 6 mm.
R ESEARCH A RTICLES
20 FEBRUARY 2004 VOL 303 SCIENCE www.sciencemag.org1164
insula and thalamus (Fig. 2, D and F), main
effects of placebo (control ⬎ placebo) were
found in the late heat period (43). Thalamic
activations extended into the basal fore-
brain and medial temporal cortex (Fig. 2E),
which has been implicated in enhanced
pain due to anxiety (44). All these placebo
activations fell within pain-sensitive regions,
consistent with the hypothesis that they reflect
modulation of the pain experience.
Time courses of neural placebo effects
(Fig. 4) show the predominant decrease late
in the pain response, after stimulus offset
(although there is a trend toward control ⬎
placebo effects earlier in stimulation as
well) (45). The late decreases suggest that
placebo effects may require a period of pain
to develop, and may modulate pain signals
most strongly after stimulation is removed.
This may be especially true of protracted
painful stimuli, such as the thermal stimu-
lus used in Study 2. The late decreases may
reflect cognitive reappraisal of the signifi-
cance of pain, resulting in decreases in pain
affect and pain experience (5, 8). Alterna-
tively, the late decreases may reflect en-
gagement of opioid mechanisms triggered
by prolonged pain.
Placebo increases prefrontal cortex
and midbrain activity in anticipation of
thermal pain. Study 2 also provided fur-
ther support for our second hypothesis, that
the expectation of pain relief is mediated by
PFC. Regions within both right and left
DLPFC, similar to those observed in Study
1 and shown in Fig. 3C, were significantly
more active during anticipation in the pla-
cebo versus control conditions (placebo ⬎
control) (46 ). Study 2 also confirmed pla-
cebo-increased activation during the antic-
ipation period of a midbrain region contain-
ing the PAG (46) (Fig. 3D), which again
correlated significantly with DLPFC activ-
ity (r ⫽ 0.60 for both L and R DLPFC)
(Fig. 3E). Finally, Study 2 showed the ex-
pected placebo-induced activation of rACC
(47 ). Interestingly, this is the same area in
which we found placebo-induced decreases
during early heat, suggesting that this pain-
responsive region may also serve as part of
the network for cognitive control.
Overall impact of the studies. These
two studies provide important insights into
the neural mechanisms underlying placebo
analgesia. First, they support the hypothesis
that placebo manipulations decrease neural
responses in brain regions that are pain
sensitive. In addition, the magnitude of
these neural decreases correlates with re-
duction in reported pain. These findings
provide strong refutation of the conjecture
that placebo responses reflect nothing more
than report bias (6 ).
Our findings also provide support for a
specific hypothesis regarding one potential
mechanism of placebo action, the repre-
sentation of expectations within regions
of PFC that modulate activity in pain-
responsive areas. We found significant cor-
relations of DLPFC and OFC activity with
placebo response, measured both behavior-
ally (as the reported experience of pain)
and neurally (as activity in pain-responsive
areas). The DLPFC is an area that has
consistently been associated with the rep-
resentation and maintenance of information
needed for cognitive control (16, 48 ),
whereas the OFC is more frequently asso-
ciated with representing evaluative and re-
Fig. 3. Prefrontal re-
gions activated with
placebo during antici-
pation. Regions in
Study 1 showed posi-
tive correlations be-
tween reported place-
bo effects (placebo ⬎
control) and brain pla-
cebo effects (place-
bo ⬎ control). Regions
in Study 2, in which
placebo responders
were preselected for
fMRI, showed main ef-
fects of placebo (pla-
cebo ⬎ control). (A)
Right dorsolateral pre-
frontal cortex (DLPFC)
in Study 1, z ⫽ 28
mm. Left DLPFC acti-
vation (found superior
to this slice) is not
shown. (B) Regions of
orbitofrontal cortex
(OFC) showing corre-
lations in Study 1, z ⫽
⫺12. (C) Right and
left DLPFC showing
main effects (placebo ⬎ control) in Study 2, z ⫽ 30 mm. (D) Midbrain placebo-induced activations
(placebo ⬎ control) in anticipation, z ⫽⫺10 mm, Study 2. (E) Scatterplot showing the correlation
between midbrain placebo effects and right DLPFC placebo effects in Study 1 (green triangles, solid
line) and Study 2 (blue x’s, dashed line).
Fig. 4. Time courses of pain responses for regions showing main effects of placebo (control ⬎
placebo) in late heat in Study 2. For display, time courses were extracted from regions showing a
main effect of pain at Z ⬎ 4.1. (A) Group-averaged, finite impulse-response deconvolved responses
to pain for placebo (blue) and control (red) in the contralateral insula (the exact region is shown
in red in the slice at left), partialing out signal contributions from the anticipation and response
periods. Black bars show average timing of trial events, although timing varied from trial to trial.
(B) Time courses in contralateral thalamus, as in (A).
R ESEARCH A RTICLES
www.sciencemag.org SCIENCE VOL 303 20 FEBRUARY 2004 1165
ward information relevant to the allocation
of control (26, 27, 29).
Previously, Petrovic et al.(33) found in-
creases in OFC in placebo during pain,
whereas the current studies found it during
anticipation (49). However, the Petrovic
study used positron emission tomography and
did not include an anticipation period, and so
could not discriminate neural responses dur-
ing anticipation from those associated with
the painful stimulus itself. Nevertheless, it
may be that the OFC is involved in processes
that occur in advance of pain only if warning
stimuli signal that pain is imminent, and oth-
erwise occur during pain itself. Affective and
motivational responses to pain are examples
of such processes.
Both DLPFC and OFC activation corre-
lated with midbrain activation during antic-
ipation, consistent with the idea that pre-
frontal mechanisms trigger opioid release
in the midbrain. An alternative interpreta-
tion is that DLPFC redirects attention away
from pain, as it has also been implicated in
general attentional processes (10, 50).
However, OFC and midbrain regions are
not typically associated with directed atten-
tion; rather, activation of these regions
seems more consistent with the view that
anticipation during placebo involves a spe-
cific expectancy process that may be relat-
ed to opioid system activation. Although
the results do not provide definitive evi-
dence for a causal role of PFC in placebo,
they were predicted by and are consistent
with the hypothesis that PFC activation
reflects a form of externally elicited top-
down control that modulates the experience
of pain.
The studies also provide additional in-
formation about which aspects of pain—
sensory, affective, or cognitive evaluation—
are affected by placebo. Previous studies
showing reversal of placebo effects by opi-
oid antagonists (2, 3), coupled with theories
implicating opioids in the inhibition of
spinal pain afferents (23), suggest that
placebo affects sensory pain transmission
at the earliest stages. Inhibition of spinal
afferents might be expected to produce pla-
cebo decreases throughout the pain matrix;
however, we found such reductions only in
a few regions (table S1). Our findings
provide evidence for multiple components
of expectation-induced placebo effects,
with (potentially) opioid-containing regions
in the midbrain active during anticipation,
anterior cingulate showing decreased re-
sponses early in pain, and contralateral
thalamus and insula showing decreases
only after more prolonged pain (Study 2).
Although our results are consistent with the
hypotheses that at least a part of the place-
bo effect is mediated by afferent pain fiber
inhibition, a major portion of the placebo
effect may be mediated centrally by chang-
es in specific pain regions. This account
acknowledges that pain is a psychological-
ly constructed experience that includes
cognitive evaluation of the potential for
harm and affect as well as sensory compo-
nents (24, 51).
References and Notes
1. P. Wall, in Textbook of Pain, P. Wall, R. Melzack, Eds.
(Churchill Livingstone, Edinburgh, 1999), pp. 1419 –1430.
2. M. Amanzio, F. Benedetti, J. Neurosci. 19, 484 (1999).
3. F. Benedetti, C. Arduino, M. Amanzio, J. Neurosci. 19,
3639 (1999).
4. D. D. Price, J. J. Barrell, Prog. Brain Res. 122, 255 (2000).
5. D. D. Price et al., Pain 83, 147 (1999).
6. A. Hrobjartsson, P. C. Gotzsche, N. Engl. J. Med. 344,
1594 (2001).
7. J. D. Levine, N. C. Gordon, H. L. Fields, Lancet 2, 654 (1978).
8. R. H. Gracely, R. Dubner, P. J. Wolskee, W. R. Deeter,
Nature 306, 264 (1983).
9. R. Melzack, K. L. Casey, in The Skin Senses D. R. Kenshalo, Ed.
( Thomas, Springfield, IL, 1968), pp. 423– 439.
10. R. Peyron et al., Brain 122, 1765 (1999).
11. R. Peyron, B. Laurent, L. Garcia-Larrea, Neurophysiol.
Clin. 30, 263 (2000).
12. F. Schneider et al., Neuropsychobiology 43, 175 (2001).
13. K. D. Davis, S. J. Taylor, A. P. Crawley, M. L. Wood, D. J.
Mikulis, J. Neurophysiol. 77, 3370 (1997).
14. K. L. Casey, Proc. Natl. Acad. Sci. U.S.A. 96, 7668 (1999).
15. J. D. Cohen, D. Servan-Schreiber, Psychol. Rev. 99,45
(1992).
16. E. K. Miller, J. D. Cohen, Annu. Rev. Neurosci. 24, 167
(2001).
17. Expectation generation is conceptually distinct from sim-
ple direction of attention away from painful stimuli, which
has also been shown to modulate pain (52–55). The critical
distinction is that expectation-induced analgesia should (i)
engage prefrontal regions primarily during anticipation of
pain; (ii) potentially activate opioid systems in the mid-
brain PAG; and (iii) activate affective regulation mecha-
nisms in OFC and anterior medial PFC (56, 57). General
attention effects, on the other hand, should be mediated
by a distributed attentional network that remains active
throughout pain and is not linked to affective regulation
and/or opioid activity.
18. Materials and methods are available as supporting
material on Science Online.
19. M. Ingvar, Philos. Trans. R. Soc. London B Biol. Sci.
354, 1347 (1999).
20. K. D. Davis, Neurol. Res. 22, 313 (2000).
21. To avoid missing pain regions due to power issues, we
defined stimulation-responsive (intense-none) and
pain-responsive regions (intense-mild) as those that
responded in these comparisons in either Study 1 or
Study 2. This procedure also makes the pain masks
comparable across the two studies.
22. Placebo effects in brain (control ⬎ placebo) that were
positively correlated with experienced pain (control ⬎
placebo) in stimulation-responsive ROIs (defined by
pain ⫺ baseline) included rACC ([4 23 27], 27 contigu-
ous voxels, Z ⫽ 3.56; and [⫺2 32 19], 16 voxels, Z ⫽
3.12) and left (contralateral) insula ([⫺44 14 ⫺3], 19
voxels, Z ⫽ 3.04). These activations overlapped with
pain-responsive ROIs (defined by intense-mild pain) in
3, 5, and 8 voxels, respectively. Thalamic placebo effects
were just below the 10-voxel extent threshold ([11 ⫺5
14], 8 voxels, Z ⫽ 2.65, 3 voxels in pain-sensitive
regions), but stronger support for thalamic involvement
was found in Study 2.
23. R. Melzack, P. D. Wall, Science 150, 971 (1965).
24. A. D. Craig, K. Chen, D. Bandy, E. M. Reiman, Nature
Neurosci. 3, 184 (2000).
25. P. Rainville, G. H. Duncan, D. D. Price, B. Carrier, M. C.
Bushnell, Science 277, 968 (1997).
26. W. C. Drevets, Biol. Psychiatry 48, 813 (2000).
27. E. T. Rolls, in The Cognitive Neurosciences M. S. Gazzanaga,
Ed. (MIT Press, Cambridge, MA, 1995), pp. 1091–1106.
28. R. C. O’Reilly, D. C. Noelle, T. S. Braver, J. D. Cohen,
Cereb. Cortex 12, 246 (2002).
29. R. Dias, T. W. Robbins, A. C. Roberts, J. Neurosci. 17,
9285 (1997).
30. J. O’Doherty, M. L. Kringelbach, E. T. Rolls, J. Hornak,
C. Andrews, Nature Neurosci. 4, 95 (2001).
31. G. Schoenbaum, B. Setlow, Learn. Mem. 8, 134 (2001).
32. DLPFC correlations were [52 18 28], 42 voxels, Z ⫽ 3.3; [42
6 30], 14 voxels, Z ⫽ 3.3; [36 20 38], 13 voxels, Z ⫽ 3.00;
[⫺30 4 42], 22 voxels, Z ⫽ 3.30. OFC correlations were
[⫺26 38 ⫺12], 32 voxels, r ⫽ 0.64; [24 30 ⫺12], 62 voxels;
r ⫽ 0.79. Correlations were also found in other areas,
including rostral (rACC) and caudal (cACC) anterior cingu-
late, which may be part of an “executive” circuit mediating
cognitive control functions. rACC correlation loci were [6
14 26], 40 voxels, Z ⫽ 4.23; [⫺4 26 26], 21 voxels, Z ⫽
3.68; [10 32 32], 20 voxels, Z ⫽ 3.36. The cACC correlation
locus was [⫺2 ⫺8 24], 68 voxels, Z ⫽ 3.5.
33. P. Petrovic, E. Kalso, K. M. Petersson, M. Ingvar, Science
295, 1737 (2002).
34. I. Tracey et al., J. Neurosci. 22, 2748 (2002).
35. Only one voxel was significant at P ⬍ 0.005 in Study
1. However, Study 2 replicated this finding with a
substantially larger activation.
36. Pearson’s r ⬎ 0.4 are significant at P ⬍ 0.05.
37. If midbrain activation were related to opioid system
activity, we might expect placebo increases in activa-
tion throughout the period when participants experi-
enced pain, rather than during the period when they
anticipated it. However, pain-induced opioid release—
an endogenous response to pain expected to be greater
in the more painful control condition—may have offset
placebo increases during pain, resulting in no overall
placebo ⫺ control differences.
38. Stimulation of the left arm in Study 2 (as opposed to
the right arm in Study 1) allowed us to test whether
placebo effects occurred contralateral to stimulation,
or always occurred in the same hemisphere.
39. At the conclusion of the manipulation phase, partic-
ipants were asked to rate how effective they expect-
ed the analgesic to be during subsequent testing. We
administered stimulation on separate, nonoverlap-
ping patches of skin in the calibration, manipulation,
and test phases to avoid physiological sensitization
and habituation effects due to repeated stimulation.
40. Of the 24 participants who returned, 22 reported a
reduction of pain in the placebo condition during the
fMRI scanning session, revealing a highly significant
test-retest reliability of the placebo effect (r ⫽ 0.62, P).
41. Data were analyzed in the general linear model framework,
with five regressors to model BOLD responses during the
trials. Regressors were unconvolved epochs (to avoid as-
suming a particular response shape) shifted by4stoallow
for the hemodynamic lag. The five time periods modeled
were (i) early anticipation, 4 to 8 s after the cue; (ii) late
anticipation, 8 to 13 s after cue offset; (iii) early pain, 4 to
14 s after stimulation onset; (iv) peak pain, 14 to 24 s after
stimulation onset; and (v) late pain, 24 to 34 s after heat
onset (stimulation offset was at 20 s). During fMRI scan-
ning, each block of six C or P test trials constituted a
separate scanner run, and BOLD responses to anticipation
and pain were compared to the baseline interval immedi-
ately after each trial. An additional regressor for the be-
havioral response (4 to 8 s after cue to respond) was
included but not analyzed further.
42. rACC: [3 18 34], 37 contiguous voxels within pain-
responsive regions, Z ⫽ 2.92.
43. The following pain-responsive regions showing placebo
effects in late heat: Right (contralateral) insula ([41 7 1],
207 voxels, Z ⫽ 3.24); right medial thalamus ([2 ⫺15 9],
10 voxels, Z ⫽ 2.63). Additional effects in left SII ([⫺58
⫺6 10], 145 voxels, Z ⫽ 3.37) were found in Study 2
but not in Study 1. See table S2 for additional regions.
44. A. Ploghaus et al., J. Neurosci. 21, 9896 (2001).
45. Placebo-induced decreases in right insula and medial
thalamus pain-responsive regions were significantly
greater during the late heat period than during the
early heat period. Results from ROIs, averaging over
voxels, for the insula: 41% larger placebo decreases in
late heat; t (22) ⫽ 1.75, P ⫽ 0.09; for early heat,
t (22) ⫽ 3.73, P ⬍ 0.001 for late heat, and t(22) ⫽
2.33, P ⬍ 0.05 in a paired t test for the difference. For
the thalamus: 20-fold larger placebo decreases in late
heat; t(22) ⫽ 0.18. P ⫽ 0.86 for early heat, t(22) ⫽
2.95, P ⫽ 0.008 for late heat, and t(22) ⫽ 2.94, P ⫽
0.008 in a paired t test for the difference.
46. Right DLPFC: [42 4 30], 55 voxels, Z ⫽ 2.79; left
DLPFC: [⫺42 14 30], 100 voxels, z ⫽ 3.34. Midbrain:
R ESEARCH A RTICLES
20 FEBRUARY 2004 VOL 303 SCIENCE www.sciencemag.org1166