Alterations in resting-state regional cerebral blood flow demonstrate ongoing pain in osteoarthritis: An arterial spin-labeled magnetic resonance imaging study




Increasing evidence suggests a central nervous system (CNS) component underpinning persistent pain disease states. This study was undertaken to determine regional cerebral blood flow (rCBF) changes representing ongoing pain experienced by patients with painful osteoarthritis (OA) of the carpometacarpal (CMC) joint and to examine rCBF variability across sessions. We used pulsed continuous arterial spin labeling (pCASL), a perfusion magnetic resonance imaging (MRI) technique.


The study included 16 patients with CMC OA and 17 matched controls. Two pCASL scans and numerical rating scale (NRS) estimates of ongoing pain were acquired in each of two identical sessions. Voxelwise general linear model analyses were performed to determine rCBF differences between OA and control groups, rCBF differences between sessions within each group, and whether sessionwise rCBF differences were related to variability in perceived ongoing pain.


In the OA group, rCBF increases representing ongoing pain were identified in the primary and secondary somatosensory, insula, and cingulate cortices; thalamus; amygdala; hippocampus; and dorsal midbrain/pontine tegmentum, including the periaqueductal gray/nucleus cuneiformis. Sessionwise rCBF differences in the OA group in the postcentral, rostral/subgenual cingulate, mid/anterior insula, prefrontal, and premotor cortices were related to changes in perceived ongoing pain. No significant sessionwise rCBF differences were observed in controls.


This is the first quantitative endogenous perfusion MRI study of the cerebral representation of ongoing, persistent pain due to OA. Observed rCBF changes potentially indicate dysregulated CNS appraisal and modulation of pain, most likely the maladaptive neuroplastic sequelae of living with painful OA. Understanding the neural basis of ongoing pain is likely to be important in developing novel treatment strategies.

Persistent pain is a major health care problem. As many as 100 million people in Europe alone experience an intractable, ongoing malaise that affects quality of life, places an increasing burden on health care resources, and costs the economy in excess of €50 billion every year (1). While multidisciplinary pain management strategies help patients cope (2), there is a recognized, unmet need for the development of novel, more efficacious treatments of persistent pain (3). In the absence of a recordable biologic marker, patients' self-reports have been relied upon uniquely to quantify both the pain experience and treatment responses; the difficulties of this approach are compounded by interindividual differences in reporting (4). Novel indices for pain measurement are required, ideally ones that relate to an underlying aspect of pain transduction, take biopsychosocial factors into account, and translate between preclinical and human studies (4, 5).

Neuroimaging techniques, such as functional magnetic resonance imaging (fMRI) and positron emission tomography (PET), have shown promise in pain research, enabling noninvasive investigation of the cerebral underpinnings of pain (6–8). Studies to date have added to our knowledge of the central representation of pain as a distributed network of brain regions underpinning the sensory–discriminative, cognitive–evaluative, and affective–motivational features of pain (5). Until recently, many have focused primarily on eliciting cerebral responses to repeated, evoked painful stimuli in healthy volunteers (9). For ethical reasons, these studies are constrained to painful stimuli that do not induce lasting tissue damage; as such, functional, biochemical, and structural alterations now known to occur during the development of persistent pain states are challenging to represent in healthy volunteers (10).

Recent studies have focused on understanding disease-specific representations of individual persistent pain states themselves (for review, see refs.6 and9). However, with few exceptions (for example, refs.11 and12), most studies have examined responses to evoked stimulation in patient groups, rather than the representation of ongoing pain experienced in all chronic pain disease states. This is most likely due to technical constraints; conventional “evoked-response” blood oxygenation level–dependent (BOLD) fMRI techniques are best suited to the comparison of repeated, rapidly changing stimuli over the course of seconds, rather than longer periods, as may be required in monitoring long-term treatment effects (13). Resting-state fMRI evaluates temporal synchrony across brain regions, and recent reports have indicated sensitivity to persistent pain disease states (11, 12). However, all BOLD fMRI measurements suffer from signal losses and distortions in brain regions close to air/bone/tissue interfaces. While H215O PET does not have these constraints, its impact in pain research has been limited by comparatively poor spatial resolution, expense, availability, and safety concerns regarding repeated radiotracer administration (6, 7). In contrast, arterial spin labeling (ASL), a perfusion MRI technique, provides quantitative, reproducible whole-brain measures of regional cerebral blood flow (rCBF) without the need for exogenous tracers and with a spatial resolution superior to PET (14). The use of ASL remains comparatively novel in pain research, but results from our laboratory and elsewhere have recently shown that ASL has suitable characteristics for the study of both clinical (15, 16) and experimental phasic and tonic pain (17–19).

In this study, we applied pulsed continuous ASL (pCASL) (14) to the challenge of describing the cerebral representation of ongoing pain in participants with painful osteoarthritis (OA) of the carpometacarpal (CMC) joint of the thumb. OA is one of the most predominant causes of persistent pain in the Western world (1). While traditionally considered to be related only to biomechanical damage of synovial joints and associated tissue (20), recent evidence suggests a significant central nervous system (CNS) component in maintaining the recurrent bouts of ongoing pain characteristic of the disease (21, 22). Despite the prevalence of OA, there remain relatively few neuroimaging reports of systems-level alterations in participants with OA (22–26).

We sought to demonstrate changes in rCBF representing ongoing pain in patients with CMC OA compared to age- and sex-matched controls. We hypothesized that there would be rCBF differences between OA and control groups in a network of brain regions specified a priori to underpin the experience of pain, based on previous studies of the cerebral representation of pain (6, 8, 9). Our second aim was to examine the reproducibility of pCASL-derived rCBF in both OA and control groups. We hypothesized that between-session rCBF differences in OA participants would relate to variation in their perceived ongoing pain, whereas pain-free controls would not demonstrate significant session-to-session rCBF differences.


Subjects and materials.

Twenty-four right-handed postmenopausal female participants (with a mean age of 60.8 years) who fulfilled the American College of Rheumatology criteria for CMC OA (27) in their dominant (right) hand and 20 age- and sex-matched controls (with a mean age of 64.7 years) participated. Informed consent was obtained from all subjects. Ethical approval for the study was granted by the local NHS research ethics committee (Ref 07/H0807/69). All patients with OA reported an average estimate of ongoing pain at rest during the week prior to screening of ≥3 on a numerical rating scale (NRS) of 0–10. All participants were required to remain on a stable drug regimen throughout the course of the study. Additional use of acetaminophen (up to a maximum daily dose of 4 gm) was permitted except during the 24 hours prior to each visit. Eight OA patients and 3 control subjects were excluded due to claustrophobia, image artifacts, development of pain in other body sites, or use of analgesic medication other than the stable drug regimen. Rescue medication was available to all participants upon request.

Study design and procedure.

The study consisted of 2 identical sessions, separated by a minimum of 7 days and a maximum of 21 days. Each session involved a screening and familiarization stage prior to MRI. During screening, all participants underwent a detailed examination by a specialist clinician to assess for the presence and severity of CMC OA. Questionnaires were administered to assess pain and psychological status, as described below. Participants were introduced to a simulated mock scanning environment before their “real” MRI examination to minimize anxiety and familiarize them with the requirements of the imaging environment. Each MRI examination comprised acquisition of high-resolution anatomic images and 2 pCASL scans. Each of the pCASL scans lasted 6 minutes, spaced 20 minutes apart. The maximum session duration was <1 hour.

Baseline psychometry.

Screening assessments were performed for all participants. Subjects were screened for depression using the Beck Depression Inventory II (Pearson Assessments) and for anxiety using the State-Trait Anxiety Questionnaire (MindGarden). State anxiety was assessed prior to both sessions. Additional estimates of ongoing hand pain, on an NRS of 0 (no pain) to 10 (worst pain imaginable), were recorded at the beginning and end of each session in each group.

Imaging procedure.

Imaging was performed on a 3T Signa HDx whole-body MRI scanner (General Electric) fitted with an 8-channel, phased-array receive-only head coil. High-resolution T1- and T2-weighted images were acquired for radiologic assessment and image registration. Resting-state rCBF measurements were made using pCASL (14), with an irradiation time of 1.5 seconds and postlabeling delay of 1.5 seconds. The pCASL images were acquired using a single-shot, fast spin-echo (3D-FSE) spiral multislice readout resulting in whole-brain blood flow maps. Scanning parameters were as follows: time to echo 32 msec, repetition time 5,500 msec; echo train length 64, matrix size 48 × 64 × 60, field of view 18 × 24 × 18 cm, and number of excitations 3. Full details of the pCASL sequence are available online at

Image preprocessing.

Preprocessing was performed using FSL software version 4.1.5 ( and statistical parametric mapping software (SPM5; Preprocessing consisted of skull stripping (FSL-BET), normalization to the Montreal Neurological Institute (MNI) template (SPM-NORMALIZE), and spatial smoothing (FSL-SUSAN: λ = 8 mm full-width half-maximum). The pCASL data were masked to include gray matter voxels only. Probabilistic gray matter images in MNI space, derived from the FSL voxel-based morphometry toolbox, were thresholded to produce a mask which included all voxels from all subjects with a >20% likelihood of being gray matter. A global CBF value, defined as the mean of all gray matter voxels within the mask, was computed for each individual pCASL CBF map in each subject.

Analysis of demographic data, psychometric data, and NRS pain scores.

Demographic, psychometric, and pain score data were analyzed using SPSS version 18.0 ( Between-group differences in age, state and trait anxiety, depression, and coping strategies were assessed using independent-groups t-tests. Groupwise differences in average NRS pain scores were assessed using a Mann-Whitney U test. Intrasubject and intersession differences in NRS estimates of pain in the OA group were examined using a paired t-test.

Global CBF analysis.

A general linear mixed-effects model, with subject and subject-by-session as random effects, and group, session, and time point as a 3-way factorial, was fitted to the global CBF data using SAS version 9.1.3 ( to assess differences in CBF global mean within and between diagnostic groups, between sessions, and between scans within sessions.

Analysis of pCASL data.

The pCASL data were analyzed using a voxelwise optimized general linear model (FSL-FLAMEO). Between-group rCBF differences were examined using a 2-level, mixed-effects approach. In each group, first-level voxelwise general linear models were fitted to pCASL data, with explanatory variables defined for each subject's mean rCBF values across scans and sessions, resulting in individual contrast estimate (COPE) images describing the mean rCBF value across all 4 pCASL scans. COPE images were used to assess voxelwise differences in rCBF between OA and control groups in a second-level mixed effects analysis using an independent-groups t-test.

Session effects.

Separate voxelwise general linear model analyses were computed for OA and control groups to examine between-session rCBF differences, specifying explanatory variables for each subject, time point, session, and global CBF value. In the OA patient group only, an additional general linear model was computed including NRS pain scores as an additional nuisance explanatory variable. F tests were computed to examine between-session rCBF differences that were independent of variability in perceived ongoing pain.

Region of interest (ROI) investigations.

ROIs defined a priori to be involved in the processing of pain, namely the primary and secondary somatosensory cortices, anterior cingulate cortex, thalamus, insula, amygdala, and hippocampus, were delineated as previously described (15). In the OA group only, summary estimates of each ROI were used to explore relationships between sessionwise changes in rCBF (controlling for sessionwise changes in global CBF) and changes in perceived pain NRS score using the semipartial correlation function in SPSS version 18. In a supplementary analysis, between-session variability in rCBF values was explored in each ROI in each group, using the method described by Bland and Altman (28). Bland-Altman plots were created using Microsoft Excel for Macintosh 2011 (


In all voxelwise analyses, resulting Z statistic images were thresholded using a Gaussian random field– corrected cluster-significance threshold [FSL-CLUSTER] of P < 0.05. The anatomic locations of rCBF changes, in MNI coordinate space, were identified using neuroanatomy atlases (29). Conventional univariate comparisons were thresholded at α = 0.05.


Demographics, psychometry, and NRS pain scores.

The demographic and psychometric characteristics of the subjects and the NRS pain scores are summarized in Table 1. There were no differences identified between sessions within the same group or between the OA and control groups with regard to age, state or trait anxiety, or depression score. Groups differed significantly in NRS estimates of ongoing pain (Z = −5.105, P < 0.001 by Mann-Whitney U test) (mean NRS across sessions 0 in controls and 3.65 in OA patients). Sessionwise NRS scores in the OA group were significantly higher in session 2 (mean ± SD 4.15 ± 2.21) than in session 1 (mean ± SD 3.15 ± 1.84) (t[15] = 2.916, P = 0.011 by paired t-test).

Table 1. Demographic characteristics, NRS estimates of ongoing pain, and psychometric variables in the OA and control groups*
  • *

    Results of between-group and within-group statistical comparisons, as appropriate, are shown. The control group included 17 subjects, and the carpometacarpal (CMC) osteoarthritis (OA) patient group included 16 subjects. NRS = numerical rating scale; NA = not applicable; BDI = Beck Depression Inventory; CCSI-R = Cognitive Coping Strategies Inventory-Revised.

  • 2-tailed P = 0.01.

 Control64.171.07 (31)
 CMC OA60.87 
NRS ongoing pain  
 CMC OA3.65 
NRS ongoing pain  
 CMC OA session 13.152.92 (15)
 CMC OA session 24.15 
State anxiety (session 1)  
 Control43.41−0.18 (31)
 CMC OA44.06 
State anxiety (session 2)  
 Control39.47−0.48 (31)
 CMC OA41.56 
Trait anxiety  
 Control42.47−0.97 (31)
 CMC OA46.00 
Depression per BDI (session 1)  
 Control4.24−0.88 (30)
 CMC OA6.20 
Depression per BDI (session 2)  
 Control4.00−0.36 (31)
 CMC OA4.75 
CCSI-R distraction  
 Control1.990.71 (30)
 CMC OA1.79 
CCSI-R catastrophizing  
 Control1.440.04 (30)
 CMC OA1.44 
CCSI-R coping  
 Control2.23−0.11 (30)
 CMC OA2.26 

Mean global CBF.

Mean global CBF did not differ between the OA and control groups (mean 50.20 in controls and 52.84 in OA patients; standard error of between-group difference 2.21; P = 0.22) and there was no main effect of session across the 2 groups (P = 0.39).

Groupwise rCBF differences.

A distributed network of brain regions demonstrated local increases in CBF in participants with CMC OA compared to matched controls. Tables 2 and 3 detail the anatomic locations and Z score statistics for cluster maxima indicating relative rCBF increases in the OA group, in brain regions specified a priori and those not specified a priori, respectively. There were no increases in rCBF identified in the control group compared to the OA group.

Table 2. Cluster index, in MNI coordinate space, of differential rCBF increases between the CMC OA and control groups in brain regions specified a priori*
RegionZ scorexyz
  • *

    MNI = Montreal Neurological Institute; rCBF = regional cerebral blood flow; CMC OA = carpometacarpal joint osteoarthritis; L = left hemisphere; R = right hemisphere.

Anterior cingulate, L3.29−4486
Posterior insula/parietal operculum, L2.98−36−3420
Insula, L3.48−36−216
Posterior cingulate, L2.99−10−5812
Posterior cingulate, R3.3916−5824
Postcentral gyrus, L4.59−52−1622
Postcentral gyrus, R4.1040−3448
Hippocampus/parahippocampal gyrus, L3.31−24−16−26
Amygdala, L3.76−260−24
Thalamus, L2.58−10−260
Periaqueductal gray3.85−4−30−12
Pontine tegmentum3.552−31−26
Table 3. Cluster index, in MNI coordinate space, of differential rCBF increases between the CMC OA and control groups in brain regions not specified a priori*
RegionZ scorexyz
  • *

    MNI = Montreal Neurological Institute; rCBF = regional cerebral blood flow; CMC OA = carpometacarpal joint osteoarthritis; L = left hemisphere; R = right hemisphere.

Medial frontal gyrus/subcallosal gyrus, L3.38−646−8
Medial frontal gyrus, L3.05−10−1478
Medial frontal gyrus/subcallosal gyrus, R3.41638−10
Middle frontal gyrus, L3.68−402220
Inferior frontal gyrus, L3.71−40424
Precentral gyrus, L3.90−62034
Precentral gyrus, R3.1238−2268
Precuneus, L4.13−4−5244
Precuneus, R3.496−7844
Superior parietal lobule, L3.76−28−7848
Inferior parietal lobule, L4.64−36−3644
Inferior parietal lobule, R4.0646−4260
Superior temporal gyrus, L3.61−56−16−2
Middle temporal gyrus, L3.09−66−10−18
Inferior temporal gyrus, L3.62−62−30−22
Fusiform gyrus, L3.56−36−50−20
Cuneus, L3.89−4−8818
Cuneus, R3.8316−9232
Lingual gyrus, L3.34−6−78−2
Middle occipital gyrus, L4.03−36−9012
Inferior occipital gyrus, L3.47−44−84−18
Tuber, L3.05−52−52−32

In the OA group, several clusters of increased rCBF were identified in brain regions specified a priori to be associated with the processing of pain (Figure 1). Bilateral clusters were identified in the postcentral gyrus, corresponding to the somatotopic representation of the hand in the primary somatosensory cortex, and the caudal inferior parietal lobule. In the left hemisphere only, increases in rCBF extended into the rostral superior parietal lobule and ventrally to the secondary somatosensory cortex, including the parietal operculum, the dorsalmost aspect of the posterior insula, and a mid-insula region proximal to the central sulcus of the insula.

Figure 1.

Local increases in regional cerebral blood flow in patients with osteoarthritis of the carpometacarpal joint compared to controls. a–c, Axial sections of the dorsal pontine tegmentum (a), periaqueductal gray/nucleus cuneiformis (b), and ventral thalamus (c). d–f, Coronal sections of the bilateral primary somatosensory cortex and contralateral secondary somatosensory cortex (d), hippocampus/parahippocampal gyrus (e), and amygdala (f). g–i, Sagittal sections of the subgenual cingulate/medial orbital frontal cortex (g), posterior and mid insula cortex (h), and periaqueductal gray/dorsal pontine tegmentum (i).

Further rCBF increases in the OA group were observed in the left hemisphere rostral anterior cingulate cortex, extending ventrally into the subcallosal gyrus and the orbital portion of the medial frontal gyrus. An additional cluster was observed in the posterior cingulate cortex, dorsal to the splenium, extending posteriorly into precuneus bilaterally. In the left hemisphere temporal lobe, rCBF increases in the OA group were identified in the amygdala, hippocampal head, and surrounding entorhinal/parahippocampal cortices. Increases in rCBF in the thalamus were confined to the left hemisphere only, approximating to the ventral posterior nuclei. Increases in rCBF in the ventral posterior formed the anterior tip of a midline cluster that descended into the dorsalmost aspect of the midbrain tegmentum, periaqueductal gray, and nucleus cuneiformis, terminating proximal to the floor of the fourth ventricle in the dorsal pontine tegmentum.

Several additional increases in rCBF were observed in brain regions not specified a priori. These regions were the left hemisphere middle and inferior frontal gyri and precentral gyrus bilaterally; the superior, middle, inferior-temporal, and fusiform gyri; and the medial and lateral occipital cortex, cerebellar vermis, and lateral-anterior cerebellar lobes bilaterally.

Assessment of sessionwise variability in the OA group.

F tests revealed sessionwise differences in rCBF in the OA group (Table 4 and Figure 2a) bilaterally in the postcentral gyrus, rostral and subgenual anterior cingulate cortex, and mid/anterior insula cortex. Left hemisphere insula rCBF differences were located anteriorly toward the frontal operculum. Further bilateral contiguous clusters extended into the medial–lateral aspect of the dorsal prefrontal cortex, including superior and middle frontal gyri, precentral gyrus, right hemisphere superior and middle temporal gyri, and medial cerebellar hemispheres bilaterally.

Table 4. Cluster index, in MNI coordinate space, of differential rCBF increases between sessions in the CMC OA group, with and without the inclusion of NRS estimates of pain as a nuisance covariate*
RegionZ scorexyz
  • *

    MNI = Montreal Neurological Institute; rCBF = regional cerebral blood flow; CMC OA = carpometacarpal joint osteoarthritis; NRS = numerical rating scale; L = left hemisphere; R = right hemisphere.

  • With pain score as a covariate.

Anterior cingulate, L4.6308−8
Anterior cingulate, R4.17228−10
Cingulate gyrus, L5.21−122036
Cingulate gyrus, R4.8220250
Insula, R5.0142−24
Medial frontal gyrus, L3.36−6−2074
Superior frontal gyrus, L3.9−104438
Superior frontal gyrus, R5.33124048
Middle frontal gyrus, L4.62−323244
Middle frontal gyrus, R4.92422230
Inferior frontal gyrus, L4.47−5426−6
Inferior frontal gyrus, R4.144618−2
Precentral gyrus, L3.61−40−2266
Precentral gyrus, R5.065406
Postcentral gyrus, R4.5858−2048
Postcentral gyrus, L3.3−22−4867
Superior temporal gyrus, R3.1664−3410
Middle temporal gyrus, R3.7656−2−10
Declive, L3.86−36−88−26
Declive, R4.7416−70−16
Tuber, L3.55−40−74−32
Uvula, R3.9730−78−34
Middle occipital gyrus, L−3.95−38−924
Figure 2.

a, Axially oriented slices showing sessionwise differences in regional cerebral blood flow (rCBF) in patients with osteoarthritis of the carpometacarpal joint. Panels show slices of the brain descending inferiorly from left to right. The upper left-most panel shows the most superior slice of the brain, and the bottom right-most panel shows the inferior-most slice. Yellow and orange indicate results obtained without numerical rating scale (NRS) pain estimates as an additional nuisance covariate; blue and magenta indicate results obtained with NRS pain estimates as a nuisance covariate. b, Relationships between sessionwise changes in rCBF and sessionwise changes in NRS pain score in the left hemisphere (L) primary somatosensory cortex (S1), right hemisphere (R) primary somatosensory cortex, left hemisphere amygdala, and left hemisphere thalamus. Symbols represent individual patients.

We hypothesized that sessionwise rCBF differences in the OA group would relate to between-session variability in NRS estimates of perceived ongoing pain. In a revised general linear model analysis, including a nuisance covariate describing participants' perceived pain, sessionwise differences in frontal, parietal, and temporal lobes were no longer present. Only a single, small cluster in the lateral occipital cortex showed differences in rCBF values between session 1 and session 2. Additional investigations of the relationship between sessionwise differences in rCBF and perceived ongoing pain, explored using semipartial regression, were performed in ROIs specified a priori. Significant correlations were observed in the left amygdala (r = 0.533, P = 0.041), left and right primary somatosensory cortex (left r = −0.52, P = 0.047; right r = −0.535, P = 0.040), and left thalamus (r = 0.612, P = 0.15) (Figure 2b). No significant sessionwise differences were identified in the control group.

Supplementary analysis.

Bland-Altman plots describing between-session variability in rCBF values in each ROI in each group are available online at


We have described increases in rCBF detected by pCASL in patients with CMC OA, compared to matched controls, in a distributed network, including the somatosensory, insula, and cingulate cortices, thalamus, and midbrain/pontine tegmentum. Sessionwise rCBF variability in participants with OA pain was related to changes in their perceived ongoing pain; rCBF measures were stable between sessions in controls, demonstrating the robustness of the ASL technique. In light of the relationship demonstrated between changes in blood flow and perceived pain between sessions, we interpret rCBF differences as representing the experience of ongoing pain resulting from OA.

This is the first study to demonstrate resting-state, ASL-detected rCBF changes in individuals with persistent pain secondary to OA. Unlike previous studies, which experimentally elicited or augmented ongoing pain (16, 19), our work examined clinically relevant spontaneous pain in patients at rest. We confirmed several hypotheses specified a priori regarding the anatomic location of changes in rCBF, which were largely in brain regions in the hemisphere contralateral to the affected painful joint. In accordance with recent comments (30, 31), we have avoided explicit reference to the “pain neuromatrix,” due to dangers of nonspecificity in its definition. On the basis of a wealth of BOLD fMRI (6, 8), PET (6, 7), and very recently, ASL (15–18) studies, however, it is clear that several of our findings represent the identification of rCBF differences indicating sensory–discriminative, affective–motivational, and cognitive–evaluative aspects of ongoing OA pain (5). For example, our findings of postcentral gyrus rCBF changes in the primary somatosensory cortex, secondary somatosensory cortex, and insula are consistent with previous accounts of responses in these regions using various evoked-pain stimulation paradigms, in both healthy control and patient groups (6–8). These similarities indicate that changes in rCBF are coupled to the pain experience, rather than being solely consequences of increased levels of cytokines, chemokines, inflammatory mediators, matrix-degrading proteases, and prostaglandins readily identifiable as blood-borne markers of active OA pathophysiology (20). This coupling is highly complex, however, and increasing evidence indicates that when the immune system is strongly activated by infection, injury, or persistent stresses (for example, chronic inflammation), morphologic and functional changes occur in blood–brain barrier permeability, nonneuronal cells (including glia), and neuronal cells with immune functions (32, 33). While a detailed discussion of neuroinflammation is not possible here, it is important to acknowledge that neuroimmune interactions have the capacity to modulate CBF, either directly or indirectly, but that these interactions remain far from fully understood. From a philosophical viewpoint, immune system changes are an important component of an emergent pain experience, comprised of multiple physiologic processes and activated in response to the threat of damage to the body (34). Neuroimmune processes are likely to be a key driver of the neuroplastic changes seen in persistent pain conditions (10).

While relative rCBF decreases representing persistent pain have been described previously using PET (35), in this study we identified only increases in rCBF in the OA group compared to controls. This finding of only increases in rCBF is consistent with other recent ASL studies examining ongoing pain (15, 16, 19). Within the OA group, however, several ROIs defined a priori demonstrated relationships between changes in perceived ongoing OA pain and change in rCBF, including a negative correlation in the primary somatosensory cortex bilaterally; local reductions in rCBF were associated with an increase in perceived pain. Unlike previous ASL studies (16, 19), correlations between rCBF and NRS in the present study were identified only after accounting for sessionwise differences in global CBF; consideration of relationships between rCBF and global CBF should be employed in future ASL studies. These findings indicate that the relationship between rCBF and perceived pain is complex, both between patient and healthy control groups and within individual groups of patients with persistent pain. Patterns of rCBF changes may be disease-state specific (10) and are likely to interact with endogenous pain modulation systems (36). Finally, while correlations between rCBF and perceived pain are important, we urge caution in their overinterpretation. Given the multifaceted nature of the pain experience, seeking analogs of perceived pain scores in individual brain regions is unlikely to be wholly successful. Arguably, neuroimaging end points should be considered in terms of the information they provide over and above subjective reporting.

Sessionwise differences in rCBF identified in the OA group related to variability in subjective reports of the severity of pain experienced. In light of significant between-session differences in NRS scores, most likely a consequence of natural variability in OA pain, rCBF changes are entirely expected. The core premise of this study was that rCBF changes would relate to the pain experienced. Sessionwise rCBF differences were only independent of variability in perceived pain in the lateral occipital cortex. Pain influences both attention (37) and visuomotor outputs (34), and previous reports (38) have demonstrated that pain modulates this region of the visual cortex, governed by the rostral anterior cingulate cortex. Changes in rCBF in this region are likely to represent modulation of mechanisms of visuospatial attention, rather than pain perception per se.

Between-session rCBF variations were not identified in pain-free control group participants. This finding is consistent with those of previous studies of repeatability within and between subjects and scanning sites using both pulsed and continuous ASL techniques (39). Concerns over measurement stability are more likely to relate to reduced rCBF values in an older population, with ages characteristic of patients with OA (40, 41). Understanding the nature of sessionwise rCBF variability will be of vital importance should the efficacy of novel therapies be examined using crossover trial designs (42); for this reason, we have provided Bland-Altman plots of between-session variability as supplementary data to inform these future studies.

Consistent with previous reports using other imaging modalities, rCBF differences in participants with OA are likely to indicate altered processing of the affective qualities of the ongoing pain (23, 24, 26). However, the pattern of rCBF increases observed in a network of brain regions, including the cingulate cortex, parietal lobe, medial temporal lobe, and prefrontal cortex, may represent a broader change than solely emotional processing and self-assessment. Such a network may signify perturbed CNS functioning, representing revised, maladaptive assessment and response to potential damage or “threat” to the body from painful OA (34, 43).

While rCBF changes could be interpreted as resulting solely from tonic nociceptive input from peripheral tissue (15, 18, 29), recent evidence, including neuroimaging examination of individuals with painful OA and other diseases, suggests that maintenance of persistent pain is mediated by mechanisms such as CNS plasticity (10, 25, 44). Some CNS changes may be reversible; for example, many patients with hip joint OA experience reversal of painful neuropathic symptoms following arthroplasty. Ten percent of patients, however, continue to report ongoing hip pain for more than 1 year postsurgery (21, 45), supporting a theory of centrally mediated maintenance of persistent pain. Alterations in the functional roles of such brain regions, following neuroplastic changes initiated by tonic input from the painful OA joint, could provide the necessary biologic machinery to support a “kindled” pain appraisal and response system (46).

Our findings are compatible with a theory of maladaptive assessment and response to threats to the integrity of the body tissues in patients with persistent pain. While the posterior cingulate cortex has frequently been described as important in visuospatial attention, additional evidence supports its putative role as a component in an “early warning” system orienting attention toward pain. Previous reports have described the importance of the posterior cingulate cortex and the posterior parietal cortex in assessing the emotional relevance of stimuli and orienting attention toward threats to homeostasis (47). Differential rCBF in the posterior cingulate cortex has been described previously in individuals with traumatic neuropathic pain (35). Increases in rCBF in the posterior cingulate cortex were continuous with regions of inferior and superior parietal lobule and anterior precuneus. These regions are reported to be involved in higher-level cognitive functioning, including memory, self-related processing, and aspects of consciousness (37, 48). The posterior cingulate cortex also has reciprocal connections with the subgenual anterior cingulate cortex, a region important in the storage of negative events and memories of personal relevance (47). The subgenual anterior cingulate cortex has extensive reciprocal connections throughout the brain, including regions with autonomic projections, including the amygdala, parabrachial nucleus, and periaqueductal gray (47), and others, including the orbital medial prefrontal cortex, hippocampus, and thalamus, all brain regions which showed rCBF increases in participants with CMC OA. Together, these regions may plausibly represent a network capable of underpinning abnormal assessment and response to the threat to homeostasis presented by pain secondary to OA. Clearly, the existence of such a network in painful OA, and potentially other persistent pain states, is speculative and requires considerable further study. However, systems-level dysregulation of select cortical, subcortical, and limbic structures (including the subgenual anterior cingulate cortex/amygdala), representing disruption of homeostatic mechanisms, have also been reported in nonpainful chronic conditions such as depression (34, 43, 49), providing an appropriate rationale for further investigation.

Our findings of rCBF changes in the midbrain and brainstem, in particular periaqueductal gray/nucleus cuneiformis and dorsal pontine tegmentum, suggest that development of persistent pain in OA, and potentially in other persistent pain states, might occur via dysregulation of descending modulatory pain control systems. The findings of this study are consistent with reports implicating the brainstem in supraspinal mechanisms of sensitization, derived from both experimental pain (50) and OA (22, 25) studies. Increased rCBF in the dorsal pontine tegmentum parallels a recent report describing the importance of this region in the maintenance of central sensitization following intradermal capsaicin injection (50). Similarly, our findings are consistent with reports of differential periaqueductal gray activity following punctate stimulation of hyperalgesic areas of referred pain in patients with hip OA, compared to healthy control participants (22). A further relationship between periaqueductal gray activity during evoked pain and questionnaire-based indices of neuropathic-like symptoms was also identified (22). In this study, increased rCBF values in these regions, in the absence of evoked stimulation, indicate altered midbrain and brainstem functioning in OA. A theory of perturbed descending modulatory systems in OA pain may well predict additional changes in the hindbrain, for example, the output pathways of the rostral ventromedial medulla (36). Unfortunately, we were unable to examine rostral ventromedial medulla functioning using pCASL. The rostral ventromedial medulla was located inferiorly to the prescribed field of view, and further methodologic development will be required to include the hindbrain within the imaging volume. Extended brainstem coverage would further our ability to examine descending modulatory systems in greater depth.

In summary, we have demonstrated the first application of pCASL to the study of ongoing pain due to OA. Our findings suggest that the pattern of rCBF increases observed in a group of participants with painful CMC OA represents the experience of ongoing pain secondary to the disease. The observed pattern of rCBF changes suggests dysregulation of systems that include evaluation of threat to the body from ongoing pain and the ability of the brain to modulate pain via descending modulatory mechanisms. Improving our understanding of supraspinal mechanisms of ongoing pain in disease should assist in the development of new approaches for therapeutic intervention.


All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Howard had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Howard, Sanders, Fotopoulou, Massat, Huggins, Vennart, Williams.

Acquisition of data. Howard, Sanders, Krause, Fotopoulou, Zelaya, Choy, Daniels.

Analysis and interpretation of data. Howard, Krause, O'Muircheartaigh, Zelaya, Thacker, Massat, Huggins, Vennart, Daniels, Williams.


This project was an academic–industrial collaboration between King's College London and the study sponsor, Pfizer Global Research and Development, UK. Pfizer and King's College London scientists worked in collaboration on the following areas: study design, data analysis, interpretation of data, and the writing of the manuscript. All data collection was performed by King's College London scientists only. Pfizer approved the content of the manuscript prior to submission.


We are grateful to Ms Caroline Wooldridge and Dr. Ellen Migo for helpful comments and suggestions on the manuscript. We would like to thank all of the participants of the Chingford Women Study and Dr. Deborah Hart (King's College London) for their time and dedication. The authors also wish to acknowledge the ongoing support of the King's Centre of Excellence in Medical Engineering.