results, see figure captions). Consequently, we
used the two boxes in the main experiment to
reinforce our value manipulation.
For the main experiment, an independent
sample of 49 healthy participants was randomly
assigned to either the cheap or the expensive
treatment group (table S1). The treatment was
introduced as a cream clinically used to treat
atopic dermatitis. We induced a nocebo expectation in both groups [( 21) and table S2]. To
implement the value manipulation, we provided
different price information about the tested
creams. The “cheap group” was told to test a
cheap cream and received the nocebo cream
from the orange box, whereas the “expensive
group” was told to test an expensive cream and
received the nocebo cream from the blue box.
To compare nocebo responses to baseline pain,
an additional control cream was introduced in
both groups. In reality, all creams were identical and did not contain any active ingredient. After the nocebo and value expectation
induction, participants underwent a heat-pain
paradigm on skin patches on the left forearm
that were pretreated with either nocebo or control cream (Fig. 1A). During an experience induction phase, temperatures were covertly increased
for nocebo and decreased for control conditions to let participants experience the supposed pain-augmenting effect of the treatment.
During the test phase, temperatures in nocebo
and control conditions were identical while
blood oxygen level–dependent (BOLD) responses
were recorded.
First, we were interested to see if pain ratings
in the nocebo condition differed between cheap
and expensive nocebo treatment. The behavioral
nocebo effect was significantly greater in the ex-
pensive group than in the cheap group (Fig. 1C
and fig. S1). Notably, pain ratings did not differ
between groups during the experience induc-
tion phase, indicating that this significant dif-
ference in nocebo effects is unrelated to the
experience induction (fig. S1). Expensive treat-
ment thus enhances behavioral expectation effects
irrespective of the directionality of the expecta-
tion. In analogy to placebo effects ( 10, 11), the
most likely explanation is that participants infer
that expensive medication contains a more potent
and effective agent and, consequently, produces
more side effects.
Furthermore, we analyzed the time course of
pain ratings using single-subject linear regression models. Slopes revealed that the nocebo
effect increased significantly over time in the expensive group compared to the cheap group (Fig.
1D). This temporal strengthening of the nocebo
effect was not influenced by the control condition (fig. S1).
In a first fMRI analysis, we identified a large
number of pain-sensitive areas along the central
nervous system that were activated during painful stimulation, irrespective of expectation and
value (fig. S2). The location of the peak voxel in
the spinal cord was within a 1-mm radius of a
pain cluster reported in a previous combined
imaging study ( 20).
Next, we identified regions that displayed neu-
ral representations of nocebo effects irrespective
of value. This pooled nocebo effect was repre-
sented in the spinal cord at the height of spinal
segment C6 (Fig. 2A and fig. S4), slightly more
caudal and medial than the pain cluster. Compar-
ing this cluster with results from a previous study
that observed nocebo activation within the spi-
nal cord in an independent sample of subjects
indicated that both clusters were located at
almost identical locations within the spinal cord
(fig. S5) ( 14).
Because the behavioral nocebo effect was
stronger in the expensive group, we investigated
how value-related nocebo effects are reflected
at the neural level. The periaqueductal gray
(PAG) showed greater activation differences
between nocebo and control conditions in the
expensive group compared to the cheap group
(Fig. 2B). Similar results were observed in prefrontal areas and the right extended amygdala
(fig. S5). Nocebo effects have been conceptualized as the opposite of placebo-related effects
( 22). However, we observed similar activations
for nocebo as compared to placebo ( 4, 10, 23),
indicating that the PAG is engaged during cognitive modulation of pain processing irrespective
of the direction of expectation, possibly through
activation of on-or-off cells, which either inhibit
or facilitate nociceptive transmission in the
spinal cord ( 24).
We further examined whether the degree of
individual nocebo effects is correlated with BOLD
signal changes along the descending pain pathway. Neural activation in the rostral anterior cingulate cortex (rACC) was negatively correlated
with behavioral nocebo effects across participants, irrespective of treatment group (Fig. 2C
and fig. S6). Thus, the level of rACC deactivation
predicted the strength of reported pain increase
during nocebo treatment, which suggests that
2 12
- 5
0
0
5
5
10
10
15
15
20
25
b = 0.65
b = -0.08
- 5
0
5
10
15
20
70 30 30
30 30 70
50
70
70
50
50
50
Experience induction
Day 1
Behavioral
Day 2
Scanner off
Day 2
Scanner on
No pain
Max pain
Max pain
No pain
Control Control Control Nocebo Nocebo Nocebo
Nocebo
Cheap nocebo
Expensive nocebo
Pain intensity (%):
***
Expensive group
N = 25
Cheap group
N = 24
No
cebo
effe
ct
(
ΔVAS
)
P
ri
cei
nE
uro
(
)
*
Trial number
Cheap Expensive Cheap Expensive
N
oce
boeff
ect
(
Δ
VA
S)
Cheap
Expensive
*
Test phase
Control Nocebo Control Nocebo Control
16 14 10 8 6 4
Fig. 1. Study design and behavioral results. (A) Experimental design of the
nocebo and value manipulation with photos of the designed medical-cream
boxes. During the experience induction on day 2, participants were lying in the
scanner, but no images were acquired (scanner off). During the test phase,
BOLD responses were recorded (scanner on). (B) The blue cream box was
estimated as being significantly more expensive than the orange box (t65 =
5. 58, P < 0.001, Cohen’s d = 0.69). (C) The behavioral nocebo effect was
significantly larger in the expensive group than in the cheap group (t47 = 2.54,
P = 0.014, Cohen’s d = 0.74). (D) Time courses of the nocebo effect expressed
as slope in a linear regression model (b) differed significantly between groups
(t47 = 2.03, P = 0.048, Cohen’s d = 0.58). *P < 0.05; ***P < 0.005; VAS, visual
analog scale; bars represent means and error bars represent SEM.