Chronic back pain is associated with decreased prefrontal and thalamic gray matter density
The role of the brain in chronic pain conditions remains speculative. We compared brain morphology of 26 chronic back pain (CBP) patients to matched control subjects, using magnetic resonance imaging brain scan data and automated analysis techniques. CBP patients were divided into neuropathic, exhibiting pain because of sciatic nerve damage, and non-neuropathic groups. Pain-related characteristics were correlated to morphometric measures. Neocortical gray matter volume was compared after skull normalization. Patients with CBP showed 5-11% less neocortical gray matter volume than control subjects. The magnitude of this decrease is equivalent to the gray matter volume lost in 10-20 years of normal aging. The decreased volume was related to pain duration, indicating a 1.3 cm3 loss of gray matter for every year of chronic pain. Regional gray matter density in 17 CBP patients was compared with matched controls using voxel-based morphometry and nonparametric statistics. Gray matter density was reduced in bilateral dorsolateral prefrontal cortex and right thalamus and was strongly related to pain characteristics in a pattern distinct for neuropathic and non-neuropathic CBP. Our results imply that CBP is accompanied by brain atrophy and suggest that the pathophysiology of chronic pain includes thalamocortical processes.
Chronic Back Pain Is Associated with Decreased Prefrontal
and Thalamic Gray Matter Density
A. Vania Apkarian,
Robert M. Levy,
R. Norman Harden,
Todd B. Parrish,
Darren R. Gitelman
Department of Physiology and Institute of Neuroscience, and Departments of
Rehabilitation Institute of
Chicago, Northwestern University Feinberg School of Medicine, Chicago, Illinois 60611
The role of the brain in chronic pain conditions remains speculative. We compared brain morphology of 26 chronic back pain (CBP)
patients to matched control subjects, using magnetic resonance imaging brain scan data and automated analysis techniques. CBP
patients were divided into neuropathic, exhibiting pain because of sciatic nerve damage, and non-neuropathic groups. Pain-related
characteristics were correlated to morphometric measures. Neocortical gray matter volume was compared after skull normalization.
Patients with CBP showed 5–11% less neocortical gray matter volume than control subjects. The magnitude of this decrease is equivalent
to the gray matter volume lost in 10 –20 years of normal aging. The decreased volume was related to pain duration, indicating a 1.3 cm
loss of gray matter for every year of chronic pain. Regional gray matter density in 17 CBP patients was compared with matched controls
using voxel-based morphometry and nonparametric statistics. Gray matter density was reduced in bilateral dorsolateral prefrontal
cortex and right thalamus and was strongly related to pain characteristics in a pattern distinct for neuropathic and non-neuropathic CBP.
Our results imply that CBP is accompanied by brain atrophy and suggest that the pathophysiology of chronic pain includes thalamocor-
Key words: chronic pain; morphometry; frontal cortex; thalamus; neuropathic back pain; aging
Ten percent of adults suffer from severe chronic pain (Harstall
and Ospina, 2003). Back problems constitute 25% of all disabling
occupational injuries and are the fifth most common reason for
visits to the clinic; in 85% of such conditions, no definitive diag-
nosis can be made (Cavanaugh and Weinstein, 1994; Deyo,
1998). The impact of chronic pain on the nervous system has
been studied primarily in animal models (Woolf and Salter, 2000;
Hunt and Mantyh, 2001; Julius and Basbaum, 2001). Such stud-
ies highlight reorganization of nociceptive coding by peripheral
afferents and spinal cord neurons and provide evidence for apo-
ptosis of spinal cord cells (Whiteside and Munglani, 2001; Moore
et al., 2002; de Novellis et al., 2004). Many of these changes are
commonly observed in both inflammatory (caused by tissue in-
jury) and neuropathic (caused by neuronal injury) pain, whereas
others are specific to one type of pain or the other. Moreover,
these subtypes of chronic pain exhibit distinct clinical character-
istics (Dworkin, 2002). Although chronic pain greatly diminishes
quality of life and increases anxiety and depression (Riley et al.,
2001; Dworkin, 2002), it is assumed that the cerebral cortex pas-
sively reflects spinal changes and reverts to its normal state after
cessation of chronic pain (Price, 2000; Mendell and Sahenk,
2003). Our studies show that chronic back pain (CBP) (sustained
for ⬎6 months) is accompanied by abnormal brain chemistry
(Grachev et al., 2000), mainly a reduction in the N-acetyl-aspar-
tate– creatine ratio in the prefrontal cortex, implying neuronal
loss or dysfunction in this region and reduced cognitive abilities
on a task that implies abnormal prefrontal processing (Apkarian
et al., 2004a). To our knowledge, no studies have yet examined
brain morphometry in chronic pain conditions.
We used structural magnetic resonance imaging (MRI) brain
scan data and two independent automated morphometry ap-
proaches to contrast brain morphology of CBP patients to
matched normal subjects. We hypothesized that neocortical gray
matter should undergo atrophy in CBP above and beyond the
atrophy associated with normal aging. This hypothesis was con-
firmed. Moreover, given our evidence for prefrontal cortical bio-
chemical and cognitive abnormalities, we also hypothesized that the
regional analysis should point to prefrontal atrophy. Previous stud-
ies have shown decreased thalamic baseline activity or decreased
thalamic responses in chronic pain conditions. Therefore, we hy-
pothesized that the thalamus should also undergo atrophy. Because
both clinical and experimental findings imply that neuropathic and
non-neuropathic chronic pain conditions may have distinct under-
lying processes, we tested for differences between these groups. The
hypothesis in this case was that neuropathic pain should have a larger
impact on the brain, which was verified.
Received June 25, 2004; revised Oct. 11, 2004; accepted Oct. 12, 2004.
Thisworkwassupportedby National Instituteof Neurological Disordersand Stroke GrantNS-35115 to A.V.A.and
National Institute on Aging Grant K23 AG-00940 to D.R.G. from the National Institutes of Health. We thank all
patients and volunteers for participating in this study. We thank the Cognitive Brain Mapping Group, especially P.
Reber,for providing MRIdata of volunteers,and M. M.Mesulam, D. Small,D. Pelli, andS. A. Khanfor commenting on
previous versions of this manuscript.
Correspondenceshould beaddressed toA. VaniaApkarian, Departmentof Physiology,Northwestern UniversityMedical
School, 5-120 Ward Building, 303 East Chicago Avenue, Chicago, IL 60611. E-mail: email@example.com.
Copyright © 2004 Society for Neuroscience 0270-6474/04/2410410-06$15.00/0
10410 • The Journal of Neuroscience, November 17, 2004 • 24(46):10410 –10415
Our results demonstrate regionally specific reduced gray mat-
ter in patients with CBP. At the whole-brain level, this reduction
is related to pain duration, regionally depends on multiple pain-
related characteristics, and is more severe in the neuropathic sub-
type. Therefore, these data present strong evidence that the
pathophysiology of chronic pain includes cortical processes, and
the observed changes likely constitute the physical substrate of
the cognitive and behavioral properties of chronic pain.
Materials and Methods
Subjects. We compared 26 CBP patients with 26 matched normal volun-
teers, after obtaining informed consent. Patients fulfilled the Interna-
tional Association for the Study of Pain (IASP) criteria for CBP (Merskey
and Bogduk, 1994) and were diagnosed in accordance with recent guide-
lines (Deyo and Weinstein, 2001). Diagnosis was performed by experi-
enced clinicians (R.M.L., R.N.H.) based on history, general physical
exam, and detailed neurological exam, especially sensory, motor, reflex,
and gait examinations. Briefly, all CBP patients had unrelenting pain for
⬎1 year, primarily localized to the lumbosacral region, including but-
tocks and thighs, with or without pain radiating to the leg. Some CBP
patients also indicated presence of pain outside this region (for example,
in the upper back); they were considered to have CBP only if the main
source of pain was lumbosacral (see Fig. 3B). We did not distinguish the
source of CBP, which may be caused by various etiologies, such as frac-
ture, inflammatory joint disease, postsurgical factors, combinations of
these, or idiopathic factors (Deyo and Weinstein, 2001). The clinical data
indicated that 15 of 26 subjects (55%) had musculoskeletal diagnoses,
five (20%) had pure radiculopathy, and six (26%) had a mixture of
musculoskeletal and radiculopathic pain. Pairwise matching was done
for gender, age (⫾2 years, except in two subjects in which match was
within ⫾5 years), and scan sequence. CBP patients were divided into
neuropathic and non-neuropathic subtypes, based on symptoms of
damage to the sciatic nerve, using IASP criteria (Merskey and Bogduk,
1994). Neuropathic CBP patients were those with significant radiculop-
athy, with or without the presence of musculoskeletal pain (i.e., a large
component of the back pain was from unilateral leg pain (⬎40%)]; in
some this radiated to the foot or toes and was accompanied by numbness
or paresthesias or by reduced straight leg raising and motor sensory or
reflex changes. In non-neuropathic CBP, the leg pain component of
CBP was deemed minimal (supplemental Table 1, available at www.
jneurosci.org as supplemental material).
Brain scans. We performed anatomic T1-weighted MRI brain scans
using two slightly different three-dimensional fast low-angle shot se-
quences, “no-flow” and “fast” paradigms, on a 1.5 T scanner. The fast
protocol uses interpolation in the slice direction. The result is a 2-mm-
thick slice interpolated toa1mmthickness, imaging parameters were:
repetition time (TR), 15 msec; echo time, 5.6 msec; flip angle, 20°; matrix,
256 ⫻ 256; and a field of view of 240 mm, with 160 mm coverage in the
slice direction. In no-flow sequence, slices are acquired at 1 mm slice
thickness using a presaturation pulse to decrease ghosting artifacts in the
temporal lobes, and imaging parameters use a TR of 22 msec.
Morphometry. The brain images were used to calculate normalized
cortical gray matter volume, skull-normalized to a standard brain, ex-
cluding the cerebellum, deep gray matter, and brainstem, and used to
calculate normalized lateral ventricular volumes using a mask designed
for this purpose. These measures were derived by the cross-sectional
version of the SIENA software, SIENAX, which uses automated brain
extraction and tissue segmentation software (www.fmrib.ox.ac.uk/fsl),
yielding an estimate of total brain tissue volume (Smith et al., 2002).
Extracted magnetic resonance (MR) images are registered to a canonical
image in standardized space to provide a spatial normalization factor.
The final skull-normalized neocortical gray matter volume compensates
for body-mass variations between subjects. Because SIENAX output is
insensitive to scan parameters (Smith et al., 2002) and our results with
no-flow scans and fast scans showed no significant differences, we com-
bined results from both scan types.
Regional gray matter density was assessed with voxel-based mor-
phometry (VBM) using the optimized method and statistical nonpara-
metric mapping analysis (Ashburner and Friston, 2000; Good et al.,
2001a). VBM was performed using SPM99 software (www.fil.ion.
ucl.ac.uk/spm), and the SnPM toolbox was used for nonparametric anal-
ysis (Nichols and Holmes, 2002). The technique has been validated with
independent region of interest measurements (Vargha-Khadem et al.,
1998; Maguire et al., 2000; Richardson et al., 2004). Images are first
normalized into a standard space and then segmented. To correct for
nonlinear spatial normalization, voxel values are multiplied by the Jaco-
bian determinants of the spatial normalization. Resultant values reflect
absolute amounts (volume in arbitrary units) of gray matter. The nor-
malized segments are smoothed with a 12 mm full width at half maxi-
mum Gaussian kernel in all three dimensions for parametric analysis and
only within plane for nonparametric analysis. Global differences in gray
matter density were treated as confounds and were removed to ensure
that differences pertained to regional rather than global changes in gray
matter. VBM analysis is sensitive to MR scan parameters. Our analysis
indicated that results from fast scans were inferior to those obtained by
no-flow scans. The no-flow scans include a saturation band inferior to
the volume acquisition to reduce artifacts caused by inflowing blood. In
addition, the no-flow data were acquired ata1mmslice thickness,
whereas the fast scans were acquired at 2 mm and interpolated to 1 mm.
Therefore, we only contrasted VBM results for no-flow scans. Regional
changes in gray matter were assessed nonparametrically to allow rigorous
cluster-based comparisons of significance. Statistical nonparametric
maps were generated using a two-group, one-scan-per-subject permuta-
tion analysis (Nichols and Holmes, 2002), in which pseudo t statistical
analysis was performed over the entire brain. Permutation-based cluster-
level inference was also done to assess significance of cluster size. Group
differences were tested against 1000 random permutations, which inher-
ently accounts for multiple comparisons. Because we observed a thalamic
decrease in gray matter density that did not pass the whole-brain
multiple-comparison threshold, we repeated the nonparametric com-
parison after spatially limiting the analysis to data within the thalami.
Regional gray matter densities were obtained from the first eigenvariate,
fora5mmradius sphere at the peak of interest, which accounted for
⬎99% of the variance. White matter changes between CBP subjects and
controls were not analyzed, because segmentation of subcortical gray
matter is not very precise and can result in contaminating white matter
volume estimates with gray matter and ventricular volumes. Similar con-
taminants may obscure estimates of ventricular volumes as measured by
VBM. For this reason ventricular volumes were only measured by
SIENAX, coupled with a specific mask.
Pain characteristics. Patients completed McGill pain questionnaire
short forms (Melzack, 1987) to derive scalars for sensory and affective
dimensions of CBP. Anxiety and depression traits were determined by
questionnaires (Beck and Steer, 1993a,b), and intensity of pain was as-
sessed on a visual analog scale (0 ⫽ no pain, 10 ⫽ maximum imaginable
pain) on scan day. Duration of CBP pain was measured in years, and drug
consumption was calculated using a validated Medication Quantifica-
tion Scale (MQS) (Middaugh et al., 1987), which reduces drugs used for
different durations and doses to a single scalar. These characteristics were
related to brain morphometry either by linear correlations or in multiple
linear regression models with stepwise elimination (supplemental Table
2, available at www.jneurosci.org as supplemental material).
Cortical gray matter volume
Skull-normalized whole-brain neocortical gray matter volume
(excluding the cerebellum, deep gray matter, and brainstem;
SIENAX analysis) was 528 ⫾ 44 cm
(mean ⫾ SD; n ⫽ 26) in the
CBP brain and 559 ⫾ 42 cm
(n ⫽ 26) in controls, matched for
age, sex, and scan type (Fig. 1A). The 30 cm
difference in gray
matter volume, a 5.4% decrease, was highly significant (paired t
test ⫽ 3.7; p ⬍ 0.001). A similar measure was derived from the
VBM regional analysis: whole-brain mean gray matter density
per voxel (VBM modulation analysis). This measure showed a
5.9% decrease in overall gray matter density (0.251 ⫾ 0.031 in
CBP subjects; 0.267 ⫾ 0.027 in controls; n ⫽ 17 pairs; paired t
Apkarian et al. •Chronic Back Pain and Brain Atrophy J. Neurosci., November 17, 2004 • 24(46):10410 –10415 • 10411
test ⫽ 2.28; p ⬍ 0.04). Whole-brain cortical gray matter volume
correlated with age in both groups (CBP subjects, age depen-
dence slope ⫽ –2.9 cm
, 0.5%; controls, slope ⫽ –2.8 cm
⫽ 0.6, p ⬍ 0.001 in both groups) (Fig. 1A). Thus, the age-
associated decrease in neocortical gray matter volume was 2.8
(0.5%) per year in both groups. This value closely agrees
with previous estimates of age-dependent gray matter atrophy
(Good et al., 2001a; Resnick et al., 2003).
Multiple linear regression indicated that whole-brain gray
matter volume in CBP depended on age, gender, and pain dura-
⫽ 28.6; p ⬍ 10
), where pain duration had a
⫺0.33 ( p ⬍ 0.008) showing that pain duration was a significant
predictor after correcting for age and gender. Consistent with
previous studies (Good et al., 2001a,b), age and gender were also
strong predictors of gray matter volume in normal subjects.
Therefore, we corrected for these confounds and then compared
the gray matter volume between controls and CBP patients (Fig.
1B). The resultant gray matter volume was 663 ⫾ 27 cm
controls and 590 ⫾ 28 cm
in CBP patients, reflecting an 11%
decrease. Mean gray matter volume was not different between
neuropathic (nuCBP) and non-neuropathic (non-nuCBP) sub-
types, but dependence on pain duration was only significant in
nuCBP, with corrected gray matter volume decreasing by 1.3 cc
(0.2%) per year of chronic pain.
Regional gray matter density
Regional specificity of brain gray matter changes was assessed
nonparametrically (Ashburner and Friston, 2000), using opti-
mized voxel-based morphometry (Friston et al., 1995; Good et
al., 2001a; Nichols and Holmes, 2002). The dorsolateral prefron-
tal cortex (DLPFC), bilaterally, was the main region that showed
local decreased gray matter density in CBP subjects relative
to controls (Fig. 2 A) (supplemental Fig. 1, available at www.
jneurosci.org as supplemental material). In the left hemisphere,
the region contained two distinct peaks [x, y, z coordinates (in
millimeters), ⫺35, 4, 55; pseudo-t
⫽ 6.38, p ⫽ 0.003; cluster
size ⫽ 25.3 cm
, p ⫽ 0.003; with a second peak at ⫺24, 5, 59;
⫽ 5.75, p ⫽ 0.006], and a single peak was seen in the
right hemisphere [x, y, z, 34, 18, 53; pseudo-t
⫽ 5.1, p ⫽ 0.034;
cluster size ⫽ 19.0 cm
, p ⫽ 0.005].
Because of the importance of the thalamus in mediating no-
ciceptive inputs to the cortex, and because of repeated reports of
decreases in thalamic activity in human brain imaging studies, a
separate nonparametric analysis was performed for the thalamus
in which the statistical comparison was performed for the space
limited by the thalami. The right anterior thalamus showed a
significant decrease in gray matter density in CBP subjects [x, y, z,
14, ⫺18, 16; pseudo-t
⫽ 4.3, p ⬍ 0.007; cluster size ⫽ 2.3 cm
p ⬍ 0.01] (Fig. 2 B) (supplemental Fig. 2, available at www.
jneurosci.org as supplemental material). The left thalamus also
shows decreased gray matter density; however, this decrease
did not pass cluster threshold (supplemental Fig. 2, available at
www.jneurosci.org as supplemental material). No significant in-
creases in regional gray matter density in CBP were observed.
For the three highly significant DLPFC peaks (showing the
largest amount of cortical atrophy), the first eigenvariate of gray
matter density was calculated, for a 5-mm-radius sphere, across
subjects. Comparisons of dependence of this regional index of
gray matter density on brain region (three locations) and on CBP
(nuCBP, non-nuCBP, and controls) with a two-way (repeated
measures for subjects) ANOVA indicated no significant differ-
ences between brain regions ( p ⬎ 0.9) and a large difference for
presence and subtype of pain (F
⫽ 26.5; p ⬍ 10
comparisons between controls and CBP patients (F
Figure 1. Decreased whole-brain cortical gray matter volume in CBP subjects. Skull-
normalized neocortical gray matter volumes are shown for CBP subjects and matched control
subjects.A, Graymattervolumes as a function of age. Thedifferenceininterceptscorrespondsto
an average decrease of 30 cm
in gray matter volume in CBP compared with the normal sub-
jects. B, Graymatter volumes asa function ofpain duration, aftercorrecting for age and gender.
Individual control subjects are shown at pain duration ⫽ 0. nuCBP and non-nuCBP CBP patient
data are presented separately. The horizontal line is the mean volume for controls. Individual
whole-brain gray matter volumes in CBP subjects are all below the mean volume for controls.
Group-averaged gray matter volumes (mean ⫾ SEM) are shown in the right bar graph, before
(top) and after (bottom) correcting for age and gender. Lines are best linear fits for each group.
Figure 2. Regional gray matter densitydecreases in CBP subjects. A nonparametric compar-
ison of voxel-based morphometry between CBP and control subjects is shown. A, Gray matter
density is bilaterally reduced in the DLPFC. The result is from a VBM permutation-based
pseudo-t test and voxel-level contrasts when all brain gray matter voxels were compared be-
tween controls and CBP subjects. Pseudocolor highly positive values indicate regions where
gray matter density was reduced in CBP subjects (controls ⫺ CBP). B, A nonparametric com-
parison spatially limited to the thalamirevealed a significant decrease in gray matter density in
the right anterior thalamus. A slice at the peak of decreased thalamic gray matter is shown.
Pseudo-t values are color coded; range is 3–6.
10412 • J. Neurosci., November 17, 2004 • 24(46):10410 –10415 Apkarian et al. •Chronic Back Pain and Brain Atrophy
p ⬍ 10
) and between nuCBP and non-nuCBP subgroups were
highly significant (F
⫽ 12.7; p ⬍ 0.0006); the average decrease in
DLPFC gray matter density was 14% in non-nuCBP subjects and
27% in nuCBP subjects relative to controls. Thus, within the DLPFC
nuCBP patients show larger decreases in gray matter density than
non-nuCBP patients (Fig. 3A).
Given that the three DLPFC regions did not differ in gray
matter density, we examined their relationship to pain character-
istics as a group. Across regions, DLPFC gray matter density in-
dex for all CBP, or for subtypes of CBP, was significantly or
borderline significantly negatively correlated with measures di-
rectly related to pain [the intensity and duration of pain and
sensory and negative-affective dimensions of CBP (p ⬍ 0.1)] and
with age and gender but not with anxiety, depression, or drug use.
In contrast, the eigenvariate for the right thalamus peak was sig-
nificantly negatively correlated only with pain duration. To dif-
ferentiate further the relationship between regional gray matter
and pain characteristics, we subtracted the gray matter index of
CBP subjects from corresponding controls. This index of change
in DLPFC gray matter, which removes age and gender con-
founds, was regressed with pain characteristics that best distin-
guish CBP subjects from controls (pain intensity, duration, and
sensory and affective dimensions of CBP). Across all CBP, the
combination of sensory and negative-affective dimensions of
CBP predicted DLPFC gray matter change (F
⫽ 5.3, p ⬍
⫽ 0.47, p ⬍ 0.004; sensory
⫽⫺0.45, p ⬍ 0.006).
When CBP was subdivided into its subgroups, we observed dif-
ferential relationships. In nuCBP subjects, pain intensity, dura-
tion, and negative affect predicted DLPFC gray matter change
⫽ 6.2, p ⬍ 0.002; pain intensity
⫽⫺0.64, p ⬍ 0.0007;
⫽ 0.62, p ⬍ 0.0009), whereas in non-nuCBP subjects all
four pain characteristics contributed to DLPFC gray matter
⫽ 12.3, p ⬍ 0.0002; pain intensity
⫽ 1.48, p ⬍
0.0002; pain duration
⫽ 1.44, p ⬍ 0.007; affect
⫽⫺1.97, p ⬍
0.0018). Thus, regional gray matter changes are strongly and spe-
cifically related to pain characteristics, and this pattern is oppo-
site for neuropathic compared with non-neuropathic types. This
dissociation is consistent with extensive clinical data showing
that neuropathic pain conditions are more debilitating and have
a stronger negative affect (Dworkin, 2002), which may be directly
attributable to the larger decrease in gray matter density that we
observe in the DLPFC of nuCBP patients.
Lateral ventricular volume
We also measured skull-normalized vol-
umes of the lateral ventricles, using a spe-
cially designed mask to isolate these struc-
tures (SIENAX analysis). There was no
difference in normalized lateral ventricu-
lar volume between CBP subjects and con-
trols (mean ⫾ SD was 21.2 ⫾ 8.2 cm
CBP and 20.5 ⫾ 7.6 cm
in controls; paired
t test ⫽ 0.34; p ⬎ 0.7). There were, how-
ever, significant positive correlations be-
tween lateral ventricle size in CBP and pain
intensity and between sensory and
negative-affective dimensions of CBP
( p ⬍ 0.05). The change in lateral ventricle
size (CBP –control) was also positively
correlated with sensory and negative-
affective dimensions of CBP ( p ⬍ 0.05).
This is the first study showing brain mor-
phometric abnormalities in chronic pain. One other study exam-
ined morphometry in pain conditions (Matharu et al., 2003), in
which migraine patients were contrasted to normal subjects, and
no significant differences were found. We observe decreased
global cortical gray matter with two independent approaches,
with both assessments showing a similar amount of overall neo-
cortical brain volume decrease in CBP. We refer to the global and
regional decreases in gray matter as atrophy. This is based on the
close agreement between the two measures used for estimating
global gray matter decrease: the validation of VBM by manual
measures and by alternate MRI measures of local gray matter
decreases (Vargha-Khadem et al., 1998; Maguire et al., 2000; Ri-
chardson et al., 2004). It should be qualified, however, that only
direct histological analysis can unequivocally confirm cellular at-
rophy. Given that normal whole-brain gray matter atrophy is
0.5% per year of aging and that atrophy caused by CBP is 5–11%,
the magnitude of brain gray matter atrophy caused by CBP is
equivalent to 10–20 years of aging. However, this analogy only
holds for the overall magnitude, because the regional specificity
of atrophy in CBP is distinct from that seen with aging (Good et
al., 2001a; Resnick et al., 2003).
After correcting for age and gender, individual CBP whole-
brain gray matter volumes were lower than the mean of controls.
Moreover, only 18% of whole-brain gray matter variance could
be explained by pain duration. Therefore, a large portion of the
whole-brain atrophy in CBP cannot be accounted for by the mea-
sured pain characteristics, implying that there may be genetic
(Zubieta et al., 2003) and experiential (Perkins and Kehlet, 2000)
predispositions contributing to the observed atrophy. In the
DLPFC, a larger proportion of the variance could be explained by
pain characteristics (40% for nuCBP; 80% for non-nuCBP), im-
plying a tighter relationship between regional brain atrophy and
perceived pain. Therefore, we suggest that the pattern of brain
atrophy is directly related to the perceptual and behavioral prop-
erties of CBP.
What does the regional pattern of atrophy imply? The ob-
served regional pattern of atrophy is distinct from that seen in
chronic depression or anxiety (Bell-McGinty et al., 2002;
Almeida et al., 2003; Yamasue et al., 2003) and shows a minimal
relationship with anxiety and depression traits. Thus, it seems to
be specific to chronic pain, especially because the regions show-
ing atrophy, the thalamus and DLPFC, participate in pain per-
Figure 3. DLPFC gray matter density as a function of subtypes of CBP. A, DLPFC gray matter density is highest in controls and
lowestinneuropathicpatients.Graymatterdensityisin arbitrary units derived from regional eigenvariates. B, Somatotopy of pain
is shown on the figurines to the right, for subtypes of CBP. Color code is the number of subjects localizing their pain to indicated
body site. Only the 17 CBP patients used in VBM are shown. The remaining CBP patients had a similar somatotopy for CBP.
Apkarian et al. •Chronic Back Pain and Brain Atrophy J. Neurosci., November 17, 2004 • 24(46):10410 –10415 • 10413
ception. The DLPFC is activated in acute pain, with responses
that do not code stimulus intensity (Coghill et al., 1999). Recent
evidence suggests that the DLPFC exerts “top-down” inhibition
on orbitofrontal activity, limiting the magnitude of perceived
pain (Lorenz et al., 2003). Thus, DLPFC atrophy may lead to a
disruption of its control over orbitofrontal activity, which in turn
is critical in the perception of negative affect in general (Small,
2002; Goel and Dolan, 2003) and particularly in pain states
(Price, 2000; Apkarian et al., 2004b). Thalamic atrophy in CBP is
important, because it is a major source of nociceptive inputs to
the cortex (although the peak decrease in gray matter seems more
anterior than the medial thalamic target of spinothalamic termi-
nations), and damage to this region may be a reason for the
generalized sensory abnormalities commonly associated with
chronic pain (Moriwaki and Yuge, 1999; Rommel et al., 2001;
Fishbain et al., 2003; Giesecke et al., 2004). Moreover, the tha-
lamic atrophy that we observe provides an explanation for re-
peated reports of decreased baseline and stimulus-evoked activity
and for abnormal chemistry within the thalamus for diverse
chronic pain states (Apkarian et al., 2004b). The dorsal anterior
cingulate is shown to be specifically involved in pain affect in
normal subjects and exhibits decreased nociceptive signaling in
various chronic pain states (Apkarian et al., 2004b), which may
again be caused by thalamic atrophy because the anterior thala-
mus is a primary input to the anterior cingulate. Therefore, we
suggest that regional atrophy dictates the brain activity observed
in chronic pain, and it may explain the transition from acute to
chronic pain by shifting brain activity related to pain affect away
from the anterior cingulate to orbitofrontal cortex.
It is possible that some of the observed decreased gray matter
reflects tissue shrinkage (changes in extracellular space and mi-
crovascular volume may cause tissue shrinkage without substan-
tially impacting neuronal properties), implying that proper treat-
ment would reverse this portion of the decreased brain gray
matter. The atrophy may be also attributable to more irreversible
processes, such as neurodegeneration, which we favor because
the main brain region involved (the DLPFC) also exhibits de-
creased N-acetyl-aspartate (Grachev et al., 2000), and decreased
N-acetyl-aspartate has been observed in most neurodegenerative
conditions, implying that it may be a marker for cell density in the
brain (Salibi and Brown, 1998), and because spinal cord neurons
undergo apoptosis in rats with neuropathic pain (Whiteside and
Munglani, 2001; Moore et al., 2002; de Novellis et al., 2004).
Similarly, within the thalamus, abnormal chemistry and de-
creased baseline and stimulus-evoked activity are all consistent
with the notion that this region may undergo atrophy. We do not
know the specific elements within the thalamus and DLPFC un-
dergoing atrophy; and whether they involve projecting neurons,
interneurons, or both, as well as glia, and in what proportions. In
the spinal cord, the evidence suggests that mainly GABAergic
inhibitory interneurons undergo apoptosis. Extrapolating from
this evidence, we assume that at the supraspinal level as well,
atrophy impinges primarily on interneurons. Recent evidence
also suggests that after nerve injury, some components of pain
behavior are a consequence of hyperactivity of spinal cord micro-
glia (Tsuda et al., 2003), and a histological study has shown a
reduction in glial numbers in the cortex in major depressive dis-
order and bipolar disorder (Ongur et al., 1998). Thus, changes
in glial numbers may be important in the atrophy we observe
in CBP as well; this remains to be determined. Because the
relationship between DLPFC gray matter decrease and pain
parameters is distinct for subtypes of CBP, it is likely that the
extent of involvement of different cellular types will also vary
with type of chronic pain.
Independent methods were used for determining global brain
gray matter volume to increase our confidence with the observed
results. We also used nonparametric analysis of VBM mainly
because this circumvents Gaussian distribution assumptions
necessary for proper calculation of cluster threshold in paramet-
ric VBM, and because the approach directly compensates for
multiple-comparison correction. We controlled for drug con-
sumption in CBP by using a unitary scale, which showed no
relationship to global or local measures of gray matter. However,
the implications of this negative result are not clear, because the
tool we used (MQS) reduces multiple drugs with multiple doses
to a unitary scale and thus may obscure effects of some drugs by
others. More-specific drug effects on brain gray matter might be
uncovered if drugs used by CBP patients are parceled into sepa-
rate categories and then related to brain morphometry. Such a
study would require a much larger cohort (we do not have
enough subjects in any particular category in the current study).
We should also state that the current study was cross-sectional in
design, which can uncover relationships but not establish causal-
ity. To achieve the latter, a longitudinal design is necessary. Our
results indicate that the type of CBP is relevant to brain atrophy.
In future studies, it will be important to delineate clinical syn-
dromes of CBP into multiple more-homogeneous categories, be-
cause this may further parcel the location and degree of changes
of brain volume and density.
Given that, by definition, chronic pain is a state of continuous
persistent perception with associated negative affect and stress, a
parsimonious mechanistic explanation for the decreased gray
matter is overuse atrophy caused by excitotoxicity and inflamma-
tory agents (Brown and Bal-Price, 2003; Mattson, 2003), poten-
tiated by predisposing factors. We cannot rule out the contribu-
tion of possible lifestyle differences between the patients and
control subjects to the observed differences in gray matter. How-
ever, we can assert the coexistence of theoretically independent
effects (i.e., group differences in global and local brain gray mat-
ter volumes) determined by two independent methods (the asso-
ciation between global and regional densities with pain duration
within the patient group and the association between regional
densities and pain parameters within the patient group that also
distinguishes between subtypes of CBP) that provide compelling
evidence for the importance of the observed morphometric
changes in the pathophysiology of CBP. We hypothesize that
atrophy of the brain circuitry involved in pain perception may
dictate the properties of the pain state, such that as atrophy of
elements of the circuitry progresses, the pain condition becomes
more irreversible and less responsive to therapy.
Almeida OP, Burton EJ, Ferrier N, McKeith IG, O’Brien JT (2003) Depres-
sion with late onset is associated with right frontal lobe atrophy. Psychol
Med 33:675– 681.
Apkarian AV, Sosa Y, Krauss BR, Thomas PS, Fredrickson BE, Levy RE,
Harden R, Chialvo DR (2004a) Chronic pain patients are impaired on
an emotional decision-making task. Pain 108:129 –136.
Apkarian AV, Bushnell MC, Treede RD, Zubieta JK (2004b) Human brain
mechanisms of pain perception and regulation in health and disease. Eur
J Pain, in press.
Ashburner J, Friston KJ (2000) Voxel-based morphometry–the methods.
NeuroImage 11:805– 821.
Beck AT, Steer RA (1993a) Beck depression inventory. San Antonio, TX:
Beck AT, Steer RA (1993b) Manual for the Beck anxiety inventory. San An-
tonio, TX: Psychological Corporation.
• J. Neurosci., November 17, 2004 • 24(46):10410 –10415 Apkarian et al. •Chronic Back Pain and Brain Atrophy
Bell-McGinty S, Butters MA, Meltzer CC, Greer PJ, Reynolds III CF, Becker
JT (2002) Brain morphometric abnormalities in geriatric depression:
long-term neurobiological effects of illness duration. Am J Psychiatry
Brown GC, Bal-Price A (2003) Inflammatory neurodegeneration mediated by ni-
tric oxide, glutamate, and mitochondria. Mol Neurobiol 27:325–355.
Cavanaugh JM, Weinstein JN (1994) Low back pain: epidemiology, anat-
omy and neurophysiology. In: Textbook of pain (Wall PD, Melzack R,
eds), pp 441–456. Edinburgh: Churchill Livingstone.
Coghill RC, Sang CN, Maisog JM, Iadarola MJ (1999) Pain intensity pro-
cessing within the human brain: a bilateral, distributed mechanism.
J Neurophysiol 82:1934–1943.
de Novellis V, Siniscalco D, Galderisi U, Fuccio C, Nolano M, Santoro L,
Cascino A, Roth KA, Rossi F, Maione S (2004) Blockade of glutamate
mGlu5 receptors in a rat model of neuropathic pain prevents early over-
expression of pro-apoptotic genes and morphological changes in dorsal
horn lamina II. Neuropharmacology 46:468–479.
Deyo RA (1998) Low-back pain. Sci Am 279:48 –53.
Deyo RA, Weinstein JN (2001) Low back pain. N Engl J Med 344:363–370.
Dworkin RH (2002) An overview of neuropathic pain: syndromes, symp-
toms, signs, and several mechanisms. Clin J Pain 18:343–349.
Fishbain DA, Cutler RB, Lewis J, Cole B, Rosomoff RS, Rosomoff HL (2003)
Is the location of nondermatomal sensory abnormalities (NDSAs) related
to pain location? Pain Med 4:238 –243.
Friston KJ, Holmes AP, Worsley KJ, Poline J-B, Frith CD, Frackowiak RS
(1995) Statistic parametric maps in functional imaging: a general linear
approach. Hum Brain Mapp 2:189–210.
Giesecke T, Gracely RH, Grant MA, Nachemson A, Petzke F, Williams DA,
Clauw DJ (2004) Evidence of augmented central pain processing in id-
iopathic chronic low back pain. Arthritis Rheum 50:613– 623.
Goel V, Dolan RJ (2003) Reciprocal neural response within lateral and ven-
tral medial prefrontal cortex during hot and cold reasoning. NeuroImage
Good CD, Johnsrude IS, Ashburner J, Henson RN, Friston KJ, Frackowiak RS
(2001a) A voxel-based morphometric study of ageing in 465 normal
adult human brains. NeuroImage 14:21–36.
Good CD, Johnsrude I, Ashburner J, Henson RN, Friston KJ, Frackowiak RS
(2001b) Cerebral asymmetry and the effects of sex and handedness on
brain structure: a voxel-based morphometric analysis of 465 normal adult
human brains. NeuroImage 14:685–700.
Grachev ID, Fredrickson BE, Apkarian AV (2000) Abnormal brain chemis-
try in chronic back pain: an in vivo proton magnetic resonance spectros-
copy study. Pain 89:7–18.
Harstall C, Ospina M (2003) How prevalent is chronic pain? Pain Clinical
Updates 11:1– 4.
Hunt SP, Mantyh PW (2001) The molecular dynamics of pain control. Nat
Rev Neurosci 2:83–91.
Julius D, Basbaum AI (2001) Molecular mechanisms of nociception. Nature
Lorenz J, Minoshima S, Casey KL (2003) Keeping pain out of mind: the role
of the dorsolateral prefrontal cortex in pain modulation. Brain
Maguire EA, Gadian DG, Johnsrude IS, Good CD, Ashburner J, Frackowiak
RS, Frith CD (2000) Navigation-related structural change in the hip-
pocampi of taxi drivers. Proc Natl Acad Sci USA 97:4398– 4403.
Matharu MS, Good CD, May A, Bahra A, Goadsby PJ (2003) No change in
the structure of the brain in migraine: a voxel-based morphometric study.
Eur J Neurol 10:53–57.
Mattson MP (2003) Excitotoxic and excitoprotective mechanisms: abun-
dant targets for the prevention and treatment of neurodegenerative dis-
orders. Neuromolecular Med 3:65–94.
Melzack R (1987) The short-form McGill pain questionnaire. Pain
Mendell JR, Sahenk Z (2003) Clinical practice. Painful sensory neuropathy.
N Engl J Med 348:1243–1255.
Merskey H, Bogduk N (1994) Classification of chronic pain: descriptions of
chronic pain syndromes and definitions of pain terms, Ed 2. Seattle: In-
ternational Association for the Study of Pain.
Middaugh SJ, Masters SL, Kee WG (1987) The medication quantification
scale: A method of quantifying medication use in chronic pain patients.
Biofeedback Self Regul 12:157–158.
Moore KA, Kohno T, Karchewski LA, Scholz J, Baba H, Woolf CJ (2002)
Partial peripheral nerve injury promotes a selective loss of GABAergic
inhibition in the superficial dorsal horn of the spinal cord. J Neurosci
22:6724 – 6731.
Moriwaki K, Yuge O (1999) Topographical features of cutaneous tactile hy-
poesthetic and hyperesthetic abnormalities in chronic pain. Pain 81:1– 6.
Nichols TE, Holmes AP (2002) Nonparametric permutation tests for func-
tional neuroimaging: a primer with examples. Hum Brain Mapp 15:1–25.
Ongur D, Drevets WC, Price JL (1998) Glial reduction in the subgenual
prefrontal cortex in mood disorders. Proc Natl Acad Sci USA
Perkins FM, Kehlet H (2000) Chronic pain as an outcome of surgery. A
review of predictive factors. Anesthesiology 93:1123–1133.
Price DD (2000) Psychological and neural mechanisms of the affective di-
mension of pain. Science 288:1769–1772.
Resnick SM, Pham DL, Kraut MA, Zonderman AB, Davatzikos C (2003)
Longitudinal magnetic resonance imaging studies of older adults: a
shrinking brain. J Neurosci 23:3295–3301.
Richardson MP, Strange BA, Dolan RJ (2004) Encoding of emotional mem-
ories depends on amygdala and hippocampus and their interactions. Nat
Neurosci 7:278 –285.
Riley JL, Robinson ME, Wade JB, Myers CD, Price DD (2001) Sex differ-
ences in negative emotional responses to chronic pain. J Pain 2:354 –359.
Rommel O, Malin JP, Zenz M, Janig W (2001) Quantitative sensory testing,
neurophysiological and psychological examination in patients with com-
plex regional pain syndrome and hemisensory deficits. Pain 93:279 –293.
Salibi N, Brown MA (1998) Clinical MR spectroscopy. New York: Wiley.
Small DM (2002) Toward an understanding of the brain substrates of re-
ward in humans. Neuron 33:668–671.
Smith SM, Zhang Y, Jenkinson M, Chen J, Matthews PM, Federico A, De
Stefano N (2002) Accurate, robust, and automated longitudinal and
cross-sectional brain change analysis. NeuroImage 17:479 – 489.
Tsuda M, Shigemoto-Mogami Y, Koizumi S, Mizokoshi A, Kohsaka S, Salter
MW, Inoue K (2003) P2X4 receptors induced in spinal microglia gate
tactile allodynia after nerve injury. Nature 424:778 –783.
Vargha-Khadem F, Watkins KE, Price CJ, Ashburner J, Alcock KJ, Connelly
A, Frackowiak RS, Friston KJ, Pembrey ME, Mishkin M, Gadian DG,
Passingham RE (1998) Neural basis of an inherited speech and language
disorder. Proc Natl Acad Sci USA 95:12695–12700.
Whiteside GT, Munglani R (2001) Cell death in the superficial dorsal horn
in a model of neuropathic pain. J Neurosci Res 64:168–173.
Woolf CJ, Salter MW (2000) Neuronal plasticity: increasing the gain in
pain. Science 288:1765–1769.
Yamasue H, Kasai K, Iwanami A, Ohtani T, Yamada H, Abe O, Kuroki N,
Fukuda R, Tochigi M, Furukawa S, Sadamatsu M, Sasaki T, Aoki S, Oh-
tomo K, Asukai N, Kato N (2003) Voxel-based analysis of MRI reveals
anterior cingulate gray-matter volume reduction in posttraumatic stress
disorder due to terrorism. Proc Natl Acad Sci USA 100:9039–9043.
Zubieta JK, Heitzeg MM, Smith YR, Bueller JA, Xu K, Xu Y, Koeppe RA,
Stohler CS, Goldman D (2003) COMT val158met genotype affects
mu-opioid neurotransmitter responses to a pain stressor. Science 299:
Apkarian et al. •Chronic Back Pain and Brain Atrophy J. Neurosci., November 17, 2004
• 24(46):10410–10415 • 10415