Authors: Inge Timmers (1Department of Medical and Clinical Psychology, Tilburg University, Tilburg, the Netherlands; 2Department of Anesthesiology, Perioperative, and Pain Medicine, Stanford University School of Medicine, Palo Alto, CA, United States), Emma E. Biggs (2Department of Anesthesiology, Perioperative, and Pain Medicine, Stanford University School of Medicine, Palo Alto, CA, United States), Lisa Bruckert (3Department of Pediatrics, Stanford University School of Medicine, Palo Alto, CA, United States), Alexandra G. Tremblay-McGaw (2Department of Anesthesiology, Perioperative, and Pain Medicine, Stanford University School of Medicine, Palo Alto, CA, United States), Hui Zhang (4Department of Computer Science, University College London, London, United Kingdom), David Borsook (5Center for Pain and the Brain, Boston Children’s Hospital, Boston, MA, United States), Laura E. Simons (2Department of Anesthesiology, Perioperative, and Pain Medicine, Stanford University School of Medicine, Palo Alto, CA, United States)
Categories: Article, Chronic pain, diffusion-weighted imaging, white matter, neurite orientation dispersion and density imaging, pediatric pain, pain catastrophizing
Source: Pain
Authors: Inge Timmers, Emma E. Biggs, Lisa Bruckert, Alexandra G. Tremblay-McGaw, Hui Zhang, David Borsook, Laura E. Simons
Chronic pain is common in young people and can have a major life impact. In spite of the burden of chronic pain, mechanisms underlying chronic pain development and persistence are still poorly understood. Specifically, white matter (WM) connectivity has remained largely unexplored in pediatric chronic pain. Using diffusion-weighted imaging, the current study examined WM microstructure in adolescents (age M=15.8y, SD=2.8y) with chronic pain (n=44) compared with healthy controls (n=24). Neurite orientation dispersion and density imaging (NODDI) modeling was applied, and voxel-based whole-white-matter analyses were used to obtain an overview of potential alterations in youth with chronic pain, and tract-specific profile analyses to evaluate microstructural profiles of tracts-of-interest more closely. Our main findings are 1) youth with chronic pain showed widespread elevated orientation dispersion compared to controls in several tracts, indicative of less coherence; 2) signs of neurite density tract-profile alterations were observed in several tracts-of-interest, with mainly higher density levels in patients; and 3) several WM microstructural alterations were associated with pain catastrophizing in the patient group. Implicated tracts include both those connecting cortical and limbic structures (uncinate fasciculus, cingulum, anterior thalamic radiation), which were associated with pain catastrophizing, as well as sensorimotor tracts (corticospinal tract). By identifying alterations in the biologically informative WM microstructural metrics orientation dispersion and neurite density, our findings provide important and novel mechanistic insights for understanding the pathophysiology underlying chronic pain. Taken together, the data support alterations in fiber organization as a meaningful characteristic, contributing process to the chronic pain state.
Worldwide, an estimated quarter to almost half of adolescents live with chronic pain,[20;24;28] with about 5% facing daily-life disabilities[17;21;33;64] that may persist into adulthood.[15;27;42;57] Despite the severe burden, our understanding of mechanisms underlying its development and persistence remains limited. Psychosocial factors, such as pain-related distress (e.g., fear, worries/catastrophizing), have been identified as important (risk) factors related to poorer pain-related functioning.[16;31;40;47] Neuroimaging studies have illustrated the importance of corticolimbic circuitry in encompassing individual differences in pain and its modulation,[18;51;55;61] and that chronic pain is associated with widespread neuroplasticity.[30;37;68] Although neuroimaging studies in youth with chronic pain are sparse, findings are largely in line with adult work, showing alterations in resting-state functional connectivity and grey matter volume in several networks, including sensory as well as more cognitive-affective networks (e.g., salience and default mode network).[6] Moreover, there is a similar shift in adolescents from sensory towards affective neural circuitry as pain becomes chronic.[63] Altered neural responses in fear and reward circuitry have been identified in youth with chronic pain as well,[48] especially in those individuals with elevated pain catastrophizing.[22;52] However, research in adolescents mainly focused on functional imaging and grey matter structure,[6] leaving white matter (WM) connectivity in youth with chronic pain largely unexplored.
In adults with chronic pain, diffusion-weighted imaging (DWI) has identified WM alterations in various tracts (e.g., corpus callosum, cingulum, corticospinal tract/CST, uncinate fasciculus/UF),[8;9;19;23;29;32;34;41] with some findings linked to individual difference factors such as pain intensity and anxiety.[10;35;43] Corticolimbic WM connectivity has been shown to predict the transition from acute to chronic pain.[36;56] However, DWI studies in adolescents are limited to youth with headache/migraine, showing WM differences compared to controls,[38;46] for instance in the cingulum.[39] Overall, findings are inconsistent, which could be explained by the diffusion tensor imaging (DTI) model and the non-specific nature of its derived metrics, making it difficult to attribute observed alterations to biological substrates.[4;45] More biologically informed models in adults with chronic pain have revealed lower WM fiber complexity and WM density differences in several tracts,[7] as well as widespread lower WM coherence (orientation dispersion), with some tracts also showing correlations with pain.[12] However, biophysical models (i.e., obtaining biophysically meaningful metrics instead of merely representing the signal) have not yet been applied to younger samples, nor have WM microstructural properties been linked to pain catastrophizing yet.
The current study examined microstructural properties of the WM in adolescents with chronic pain. Two complementary approaches were 1) a voxel-based whole-white-matter analysis to compare youth with chronic pain and controls, and 2) a tract-specific profile analysis for in-depth evaluation of tracts-of-interest (cingulum, CST, forceps minor, UF and anterior thalamic radiation/ATR). In addition to DTI, we applied the biophysical model neurite orientation dispersion and density imaging (NODDI), providing indices on neurite density and orientation dispersion.[26;66] We focused on examining group differences and on relations with pain catastrophizing in the chronic pain group. We hypothesized that WM alterations would be observed, especially in tracts connecting cortical and limbic structures (e.g., cingulum, uncinate fasciculus), which would be associated with pain catastrophizing.
Ninety-seven participants were enrolled in this study, of which 61 had chronic pain (56 females, 5 males) and 36 did not (27 females, 9 males). The youth with chronic pain were recruited from the Pain Treatment Service – Chronic Pain Clinic at Boston Children’s Hospital when they presented for consultation. The inclusion criteria were age 10-24 years, pain duration >3 months, and confirmed diagnosis of chronic non-disease-related pain. We excluded participants taking opioid or antipsychotic medications, but not for taking selective serotonin reuptake inhibitors. Other exclusion criteria were significant cognitive impairment, significant medical or psychiatric disorder, pregnancy, claustrophobia, and magnetic implants. The participants without chronic pain were recruited from the community using local advertisements, with current or history of chronic pain (symptoms for >3 months) being an additional exclusion criterion.
Several participants were n=3 because of chronic pain history or incidental finding (2 controls, 1 patient), n=16 because no DWI data was collected (9 patients, 7 controls), and n=4 because DWI data collection was incomplete (3 patients, 1 control). A total of 74 datasets were pre-processed (48 patients, 26 controls), of which n=6 more were excluded based on data quality (see Magnetic Resonance Imaging – Data Analyses). Thus, the final sample consisted of 68 participants (44 with chronic pain, 24 pain-free peers).
Those participants that were excluded from analyses (n=29) were slightly younger than those included in the final sample (M=14.6y, SD=2.8y vs M=15.8y, SD=2.8y; F1,95=4.22, p=.04), but did not differ in sex (X^2^1,n=97=0.56, p=.45) or group (patients or controls; X^2^1,n=97=0.32, p=.57). Within the patient group, there were also no differences in pain duration, pain intensity, pain sensitivity, pain-related disability, or pain catastrophizing (pain F1,56=0.63, p=.43; pain F1,59=0.04, p=.84; pain sensitivity/PPT: F1,59=0.05, p=.83, FDI: F1,59=0.25, p=.62; PCS-C: F1,59=0.16, p=.69).
DWI was collected as part of a larger study on threat-safety discrimination learning in youth with chronic pain. This study was approved by the Boston Children’s Hospital Institutional Review Board (#P00013786). Participants and legal guardians provided written assent/consent. Data on task-evoked functional MRI data and resting-state fMRI have been published,[22;52] but the DWI data have not yet been described.
Participants filled out questionnaires using REDCap electronic data capture tools hosted at Boston Children’s hospital. Demographics and pain characteristics included age, sex, and pain intensity (numerical rating scale from 0-10). Pain duration was calculated from the medical records.
The questionnaire battery included, but was not limited to, the Pain Catastrophizing Scale for Children (PCS-C) to asses catastrophic thinking about pain, including rumination, helplessness and magnification;[11] and the Functional Disability Inventory (FDI) to assess perceived difficulties in performing activities due to their pain.[58]
In addition, pain sensitivity was assessed by measuring pressure pain thresholds (PPTs) on the thumb of the non-dominant hand. Using an algometer (AlgoMed, Medoc, Ltd), pressure was gradually applied perpendicular to the thumb and participants were asked to indicate when they started feeling discomfort/pain. The pressure was then immediately released. This was repeated three times, and the first assessment was discarded to account for unfamiliarity effects.
Diffusion-weighted images were acquired on a 3 Tesla whole body MRI scanner (Siemens Magnetom TrioTim syngo MR B17) using a 12-channel head coil. A 2D echo-planar imaging (EPI) sequence was used with the following 70 slices with isotropic voxels of 2 mm^3^, repetition time (TR)=5200 ms, echo time (TE)=109 ms, field of view=240 mm, slice acceleration factor=2. Two different b-values were b=1000 s/mm^2^ and b=2000 s/mm^2^, both with 30 diffusion-encoding gradient directions (i.e., same directions across the two shells). In addition, 5 interspersed b=0 volumes were acquired, as well as 1 b=0 with opposite phase encoding direction (posterior to anterior). The diffusion encoding directions spanned the entire sphere.
First, the susceptibility induced off-resonance field was estimated from the unweighted (b0) volumes acquired with the different phase-encode directions (i.e., anterior-to-posterior and posterior-to-anterior) using topup as implemented in FMRIB Software Library (FSL version 6.0).[1;50] Then, using fsl’s eddy, these susceptibility-induced distortions as well as eddy current-induced and motion distortions were corrected in the full dataset.[3] Slices with signal loss caused by subject movement coinciding with the diffusion encoding were detected and these outlier slices were replaced by predictions made by a Gaussian Process.[2] The quality of the DWI dataset was assessed using the eddy QC tools.[5] Based on the quality reports, outlier subjects were identified and excluded. We defined an outlier as a participant who exceeded 2 mm for absolute motion, or who exceeded 3 standard deviations from the mean for absolute, relative displacement, number of outliers volumes, SNR (b0) and CNR (b1000, b2000).[44] This resulted in removal of 6 participants (2 controls, 4 patients). Data were visually inspected as well. Figure 1A shows the entire pipeline.
The diffusion tensor imaging (DTI) model is a single compartment model, which models the DW data as a tensor using a Gaussian model (Figure 1A). As this is a linear fit, this step was restricted to the data acquired with the lower b-value (i.e., b1000). The main derived metric is fractional anisotropy (FA), used as a proxy of white matter integrity as it reflects the degree of diffusion that is anisotropic or along a certain axis (i.e., as is the case for axons). The model was fitted using fsl’s dtifit.
Neurite orientation dispersion and density imaging (NODDI) is a multi-compartment model,[66] which requires multiple shells (i.e., in this case b1000 and b2000). Specifically, it models the intra-neurite space as restricted diffusion (i.e., sticks with a Watson distribution); the extra-neurite space as a tensor with hindered, but not restricted diffusion (i.e., anisotropic Gaussian diffusion); and the cerebral spinal fluid as free diffusion (i.e., referred to as the free water fraction/FWF; see Figure 1A). Two key derived indices are 1) the neurite density index (NDI, typically high in WM, low in GM) represented by the volume fraction of the intra-neurite space, showing high correspondence with histological measures of myelin staining intensity;[25] and 2) the orientation dispersion index (ODI), quantified by the angular variation of neurite orientation (ranging from 0 for perfectly coherently oriented structures, to 1 for isotropic structures; typically high in GM, low in WM), showing high correspondence with quantitative Golgi analyses.[14] NODDI was fitted using linear framework for Accelerated Microstructure Imaging via Convex Optimization (AMICO),[13] as implemented in python.
A tensor-based spatial alignment of the data was performed using DTI Toolkit (DTI-TK; nitrc.org/projects/dtitk).[65;67] Alignment based on tensors has been shown to be more accurate compared to standard FA-based methods.[59] We performed spatial alignment to a study-specific template in three bootstrap with an initial standard template, affine transformation, and deformable transformation. The resulting deformation fields were subsequently combined to warp the data from native space to the study-specific template space in one single transformation. This combined deformation field was also used for alignment of the scalar NODDI maps (i.e., NDI, ODI and FWF).
The framework of tract-based spatial statistics (TBSS)[49] was used to focus our analyses on the core WM tracts, summarized in a so-called WM skeleton. A customized, enhanced version of TBSS was used with improved spatial normalization, as described before.[53;54] Based on the DTI data, a mean FA skeleton was created (thresholded at FA>.2, containing 69,203 voxels; Figure 1B), and individual FA data were projected onto the skeleton. In addition, using the calculated distance maps, the NODDI metrics were projected on the same skeleton as well.
We used Automated Fiber Quantification (AFQ)[62] to identify and isolate the major fiber tracks, in order to extract tract-specific profiles. Pre-processed data, but in original space, were fed into the AFQ toolbox (github.com/yeatmanlab/AFQ). Details on AFQ can be found in [62] but in AFQ performs whole brain tractography using deterministic tractography (in a WM mask, defined as FA>.2, in all directions until FA values drop below .2), segments the fiber groups into 20 tracts (or fascicles) using regions of interest (ROIs) that are warped into native space, outlier fibers are removed, and tract properties are calculated across 30 equidistant nodes. Tract profiles for all 20 tracts were generated with respect to the NODDI and DTI metrics. Here, we focus on specific tracts-of-interest: the left and right cingulum, left and right CST, forceps minor, left and right UF and the left and right ATR (Figure 1C). We successfully identified the tracts in all participants, except for the right cingulum of 1 control. In addition, we excluded the following tracts as they had too few streamlines (<10% of the mean) or presented with outlier values in NDI, ODI and/or FA (i.e., >3*interquartile range above the third quartile): the left cingulum of 1 control and 1 patient, the right cingulum of 1 control, and the left UF of 1 control.
For the demographics, questionnaire and sensory testing data, group characteristics were compared using General Linear Models (GLM) having group (patients, controls) as between-subject (BS) factor using SPSS 25.0.
For the voxel-based analyses, permutation-based statistics were carried out on the skeletonized data using fsl 7randomize. TFCE (Threshold-Free Cluster Enhancement; 2D) was used to account for multiple comparisons (5,000 permutations), and a minimal cluster extent (k) of 10 voxels was used. In addition to group comparisons for the NODDI (i.e., NDI, ODI, FWF) and DTI (i.e., FA) metrics, correlations between NODDI metrics and pain catastrophizing were examined across the skeleton in the chronic pain group. Age was added as a covariate to account for age-related effects. Clusters with a TFCE-p < .05 and k > 10 were considered statistically significant. From significant clusters, parameter values were extracted for plotting and for further examining correlations with pain-related outcomes post-hoc using Pearson correlation analyses in SPSS (i.e., pain-related disability, pain catastrophizing, pain duration, pain intensity, and pain sensitivity).
For the tract-profile analyses, we used linear mixed effects (LME) models, as implemented in Matlab R2021a. This allowed us to account for the continuous nature of the ‘node’ predictor (i.e., n=30 nodes per tract) and the quadratic profile of the tracts (i.e., U- [or inverted-U-] shaped profiles, reflecting different microstructural properties at the ends of the tracts compared to the middle). To investigate whether the groups differed in their tract profile, the NODDI and DTI metrics (i.e., NDI, ODI, FA) were included as the dependent variable, with fixed effects predictors of group (patients, controls) and node (in quadratic node + node^2^). Additionally, age and FWF were included as covariates to account for age-related and partial volume effects (i.e., CSF contamination). Participant was included as a random effects predictor. We applied a false discovery rate (FDR) correction to account for the number of tracts that were analyzed. Effects with an FDR-p < .05 were considered statistically significant. For tracts that showed significant group-related differences in tract profiles of the NODDI metrics, we followed up with analyses in the chronic pain group only that additionally included pain catastrophizing as a fixed effects predictor, to examine whether tract profiles were related to the degree of pain catastrophizing.
The final sample consisted of 68 youth, including 44 with chronic pain and 24 without chronic pain (age M=15.8y, SD=2.8y; 57 females; Table 1). Age did not differ significantly across groups, but sex did. As expected, patients reported more pain-related symptoms, but did not differ significantly in terms of pain sensitivity. Additional details on the patient group can be found in Table S1.
The patient and control group were compared in terms of motion (i.e., average absolute and relative displacement), and signal and contrast to noise ratio (SNR/CNR) for each shell (Figure S1). No significant differences were identified (absolute F1,66=1.49, p=.23; relative F1,66=0.22, p=.64; % outlier F1,66=0.38, p=.54; SNR F1,66=0.00, p=.96; CNR F1,66=0.70, p=.41; CNR F1,66=1.63, p=.21).
Analyses showed elevated ODI in youth with chronic pain compared to controls in several clusters, including in bilateral corona radiata, superior longitudinal fasciculus, CST, tracts going into pre- and postcentral gyrus, and left UF (Figure 2A, Table 2). Elevated ODI was observed bilaterally, although it was slightly more pronounced in the right hemisphere. No group differences in NDI were observed. Also, no differences in FWF were found, indicating that there were no (potentially confounding) differences in partial volume effects across groups.
From the identified clusters, parameter values were extracted (Figure 2B). In follow-up tests, we then correlated the parameter values with pain-related outcomes in the patient group (see Table S2 for all correlations). ODI in cluster 4 (in right posterior corona radiata / cingulum) was correlated with pain catastrophizing (PCS-C: r=.33, p-uncorr=.03) and with pain-related disability (FDI: r=.32, p-uncorr=.04), in that higher ODI was associated with higher levels of pain catastrophizing and disability. Also, ODI in cluster 6 (left CST) showed a marginally significant correlation with pain sensitivity (PPT: r=.29, p-uncorr=.05), in that higher ODI was associated with higher thresholds and hence lower pain sensitivity. No other significant correlations were observed.
In addition, we investigated whether the other NODDI indices also showed differences in the identified clusters that did not show up in the skeleton-wide voxel-based analyses. No differences were observed across groups in NDI or FWF in any of the clusters (Figure S2).
Across the entire WM skeleton, there were no regions that showed a significant correlation with pain catastrophizing in the patient group, in any of the metrics (all TFCE-p > .05).
No group differences were identified across the WM skeleton in terms of FA (all TFCE-p > .05).
When zooming into the clusters that showed elevated ODI in patients, we observed that these clusters also showed lower FA (Figure S2). Note that this direction was as expected, as higher ODI is in line with lower FA.
Analyses showed no significant main effects of group in NDI or ODI in any of the tracts-of-interest. However, interactions between node (and/or node^2^) and group did reveal differences in the tract profile for ODI in the left and right CST, left and right UF, and forceps minor (Figure 3A, Table S3). Observed effects were largely driven by elevated ODI in patients compared to controls. In both left and right CST, we observed a significant interaction between node^2^ and group, indicating a sharper U-shaped profile with group differences appearing at the superior parts of the tract. A similar interaction between node^2^ and group was present in the right UF, with the sharper profile indicating higher ODI in the ends of the tract. In the left UF the significant interaction between node and group indicated a shift in ODI was elevated in the mid to frontal parts of the tract. The interaction between node and group in the forceps minor revealed a shift in ODI, with higher ODI for controls compared to patients in the left portion and higher ODI for patients in the right portion.
Follow-up tests in the chronic pain group showed significant main effects, and interactions between node and pain catastrophizing in the right and left UF (Figure 3B, Table S4). In the right UF, higher levels of pain catastrophizing were associated with higher ODI (main effect of catastrophizing), and a less sharp profile reflected in higher ODI in the mid-portions of the tract (interaction between node^2^ and catastrophizing). In the left UF, the opposite main effect was present, with higher pain catastrophizing being associated with overall lower ODI, although the interaction with node^2^ indicated that the individuals with higher pain catastrophizing showed a sharper profile, reflected in higher ODI at the ends of the tract, but lower ODI values in the mid-portion of the tract.
For NDI, significant interactions between node (and/or node^2^) and group also revealed differences in the tract profile of the right ATR, right cingulum and left CST (Figure 4A, Table S3). The differences were largely driven by elevated NDI in patients compared to controls. In the right ATR, this difference was observed in a node^2^ by group interaction, with higher NDI in the mid-portions of the tract for patients compared to controls. In the right cingulum, we observed increased NDI in the posterior to mid-portions of the tract, with a sharper decrease at the anterior portion for patients, compared to controls (interaction node by group, and node^2^ by group). The left CST showed higher NDI for patients compared to controls in the increasingly superior portions of the tract (interaction node by group, and node^2^ by group; similar to the differences in ODI).
Follow-up tests in the patient group showed significant interactions between node and pain catastrophizing in both the right ATR and right cingulum, with also a main effect of pain catastrophizing in the right cingulum (Figure 4B, Table S4). In the right ATR, higher pain catastrophizing was related to a shifted and sharper NDI profile, with elevated NDI especially in the mid-portion of the tract (i.e., interaction node by catastrophizing, and node^2^ by catastrophizing; similar to the group-related effect). In the right cingulum, higher pain catastrophizing was also associated with increased NDI (i.e., main effect of catastrophizing), with the interaction of node by pain catastrophizing indicating that this was related to a shift with higher NDI predominantly in the anterior portions of the tract.
Group-related differences were observed for FA in left and right CST, left and right UF, and left cingulum (Figure 5, Table S5). In both the left and right CST, we observed the expected inverted profile, relative to ODI, in FA, with a significant node by group interaction. Specifically, the superior portion of the tract mirrored the pattern observed in ODI, with lower FA in patients compared to controls. In the left and right UF, we also observed group-related differences in FA. These differences were largely present in node^2^ by group interactions, with flatter curves for patients compared to controls in FA (the mirrored pattern observed in ODI) resulting in lower FA at the ends of the tracts. While group differences in ODI were observed in the forceps minor, these were not present in FA. While there were no differences in ODI or NDI in the left cingulum, we did observe a significant node and group interaction for FA in this tract. This difference was characterized by a shift in profile, with higher values in FA for controls compared to patients especially in the middle and anterior portions of the tract.
We examined WM microstructure in youth with chronic pain, revealing 1) widespread elevated orientation dispersion in youth with chronic pain compared to controls (e.g., in corona radiata, CST and UF), indicative of less coherence in those tracts; 2) neurite density tract-profile alterations (e.g., in CST, cingulum and ATR), with mainly higher, but also lower density levels in patients; and 3) several associations with catastrophizing in the patients (e.g., in UF, cingulum). Overall, FA analyses corroborated NODDI findings, yet offered little unique insights, supporting the use of more biologically specific metrics to probe WM microstructure. As hypothesized, cortico-limbic tracts (e.g., UF, cingulum, ATR) were altered and associated with pain catastrophizing, although other, sensorimotor tracts (e.g., CST) also exhibited WM alterations, which potentially relate to other (chronic) pain aspects.
Our analyses included voxel-based and tract-specific profile analyses to allow an overview of WM alterations as well as focusing on tracts-of-interest. We will summarize and discuss findings per tract/region (see also Figure 6).
Bilateral microstructural alterations were identified in the CST, carrying axons from the spinal cord to sensorimotor regions, and in the associated corona radiata, carrying axons from and into the cortical layers. Specifically, ODI was higher in patients compared to controls, especially in superior parts of the CST as these tracts approach their cortical targets. Clusters of differences spanned various sensorimotor regions; from the more anterior supplementary motor area to the primary motor region (MI; precentral gyrus) to primary somatosensory cortex (SI; postcentral gyrus). Only one cluster in the posterior corona radiata/posterior cingulum showed correlations with pain catastrophizing and disabilities, although it should be noted these were not corrected for multiple testing (and would not survive a multiple comparison correction). Another cluster in left CST, however, showed a marginally significant correlation with pain sensitivity (also uncorrected), suggesting that higher ODI may be related to lower pain sensitivity in patients. Also in the superior portions of the left CST, the tract-profile analyses showed higher NDI in patients. In older adults with chronic pain, opposite directions have been reported, showing lower coherence in corona radiata (i.e., lower ODI).[12] In people with complex regional pain syndrome, findings were in opposite directions too, yet similarly complex, showing a combination of increased axial and increased radial diffusivity in corona radiata, which would be in line with lower ODI and lower NDI, respectively, although the interpretation of these DTI metrics has been challenged.[60]
Higher ODI and lower FA was also observed in bilateral UF, carrying axons that run medially from temporal and limbic structures to inferior portions of the frontal lobe. Although there were differential patterns across hemispheres, the frontal part of the UF seemed to be robustly altered in patients, and higher ODI in these tracts was related to more pain catastrophizing. In youth with migraine, the UF was also investigated, but no group differences were observed, nor any relations with headache frequency.[39] In an adult study, lower fiber complexity was observed in right UF,[7] which would be in line with higher ODI and lower FA. Yet, in a group of older adults, UF findings were opposite showing lower ODI levels in patients.[12]
The forceps minor, or anterior forceps, connect the lateral and medial parts of the frontal cortex via the corpus callosum. While higher ODI in patients was observed in the right forceps minor, the left portion showed lower ODI in the tract-profile analyses. It is unclear what this lateralization in fiber coherence differences means, and follow-up research will have to provide more information on its robustness and its meaning.
The cingulum projects from the cingulate gyrus to parts of the limbic system, therewith connecting frontal, parietal and medial temporal brain regions. The voxel-based analyses showed elevated ODI in a right posterior cluster of the cingulum, although this cannot be distinguished from posterior corona radiata. Furthermore, patients with higher levels of pain catastrophizing showed less coherent fiber bundles, which is in line with our previous findings of decreased functional connectivity during rest between amygdala and parietal lobe in those with more catastrophizing.[52] Interestingly, the tract-profile analyses only showed altered NDI in the right cingulum, which was associated with catastrophizing. The finding of higher NDI and FA levels in patients is in line with a study in youth with migraine, observing higher FA in the cingulum.[39] Especially in the anterior portions of the tract, our patients showed lower NDI, while lower NDI was associated with less catastrophizing, potentially pointing towards a compensatory or protective mechanism. In adults, findings pointed towards lower crossing fiber complexity in the cingulum,[7] which would be in line with higher FA, but not with elevated ODI.
Lastly, tract-profile analyses demonstrated group-related NDI differences in the right ATR, which connect the thalamus with the prefrontal cortex via the anterior limb of the internal capsule. Higher NDI in patients was observed in portions of the tract, which was furthermore associated with more catastrophizing. No other group-related effects were observed in this tract. Higher NDI points towards increased fiber count or increased myelination of this tract in the youth with chronic pain.
The overview in Figure 6 illustrates that the WM microstructure findings are complex. In most instances where differences between patients and same-aged peers emerged, the orientation dispersion and/or neurite density was higher among patients, which is not typical, and in part associated with pain catastrophizing. Differences in orientation dispersion were more pronounced though (i.e., emerging in voxel-based and tract-based analyses), more widespread and bilateral, compared to those in neurite density. Broadly speaking, the observed differences in microstructure are consistent with previous findings in the cingulum for youths with chronic headaches.[38;39] However, here we extend those previous findings showing alterations in multiple tracts -not limited to cortico-limbic circuitry- with complex patterns and associations with pain catastrophizing.
Our study in adolescents points towards involvement of similar tracts compared to the adult literature, including CST, corona radiata, UF and cingulum, although the directionality varied. Comparisons, however, are complicated by the usage of different analysis approaches (e.g., challenging comparison of crossing fiber complexity and ODI). Moreover, comparing people in development (when axons are pruning, myelination is still in progress, and hence coherence is increased) to adult people (when axons may be already degenerating, decreasing coherence) is challenging. Thus, even differences in findings with the other NODDI study[12] may be explained by other factors. It does support the idea that adult studies do not necessarily extend to younger populations, and we need more research in developing samples that should also address the role of age and pubertal status.
The complex pattern of WM microstructure alterations suggests various underlying pathophysiological processes. The widespread increases in dispersion (less coherence) in ODI point to altered fiber organization as a key characteristic of chronic pain and may indicate altered or delayed pruning. This may be a marker of ‘wear and tear’ on the brain due to dealing with disabling pain. The lack of associations with pain intensity or pain duration suggests that it is more about coping and giving meaning to pain than simply living with pain. NDI alterations were identified too, yet less pronounced, suggesting some involvement of myelination processes. Several tracts showed higher ODI and higher NDI (e.g., CST, cingulum), pointing to a combination of less coherence across fibers and increased number of fibers, and hence multiple pathophysiological mechanisms at play. The increases in NDI, and associations with less catastrophizing in these tracts, suggest that these alterations may be compensatory in nature, potentially the result of (successfully) coping with pain.
Our findings should be interpreted in light of some considerations and limitations. First, although our analysis approach was comprehensive, part of our analyses was limited to our tracts-of-interest, and hence we cannot exclude the possibility that we may have missed findings in other regions that were not strong enough to be uncovered in the voxel-based whole-white-matter analyses. Second, voxel-based analyses focus on a core WM skeleton and the tract-profile AFQ analyses focus on the central tract portions, and hence potential alterations in more peripheral regions cannot be excluded. Future research could move beyond the core WM portions and move into the areas where axons start entering (and exiting) the cortex, or perform tractography originating from specific subcortical nuclei (e.g., subnuclei of the amygdala or thalamus). Also, our patient sample included several pain types and while it would be interesting to explore commonalities and differences across these subgroups, our sample size did not allow any formal comparisons. Last, the sample was mostly White and female with highly educated parents, challenging its generalizability. Only few males were included in the sample, and although that is reflective of higher prevalence rates of chronic pain in girls,[28] the small number also limits any inferences on potential gender/sex contributions.
We identified widespread tracts/regions with higher orientation dispersion (or less coherence) in young people with chronic pain, in addition to tract-profile alterations in neurite density. Implicated tracts include both those connecting cortical and limbic structures (i.e., AF, cingulum, ATR), which were associated with catastrophizing, as well as sensorimotor tracts (i.e., CST). By identifying alterations in biologically informative WM microstructural metrics, our findings offer novel mechanistic insights into the pathophysiology of chronic pain, pointing towards alterations in fiber organization as a meaningful characteristic.