Chronic pain is associated with greater brain entropy in the prefrontal cortex

Abstract

Chronic pain is a debilitating clinical condition and a severe public health issue that demands to be addressed. Neuroimaging-based techniques have been widely adopted to investigate the neural underpinnings of chronic pain. Despite the efforts the complex nature of pain experience as well as the heterogeneity of chronic pain have made the identification of neuroimaging-based biomarkers extremely challenging. In this study, resting-state fMRI-based brain entropy, a measure reflecting the “irregularity” of brain activity, was adopted as a biomarker of chronic pain by comparing individuals with chronic pain and healthy controls in a sample of middle-to-old-age participants (n > 30,000) drawn from the UK Biobank database. Abnormal brain entropy is associated with altered brain dynamics and may serve as a potential marker of disrupted pain processing in individuals with chronic pain. Compared to healthy controls, individuals with chronic pain exhibited increased brain entropy in a broad set of regions including the frontal, temporal, and occipital lobes, as well as the cerebellum. In addition, individuals with a more distributed chronic pain showed increased brain entropy in occipital lobes. When examining distinct types of chronic pain individually, only participants with headache and pain all over the body showed brain entropy differences compared to a matched sample of healthy controls.

Perspective

This article investigates the neural substrates of chronic pain using brain entropy, a measure of the randomness and irregularity of brain activity. This measure could potentially aid in the assessment and treatment of chronic pain.

Introduction

Pain is a complex, multidimensional, and subjective experience that plays a key role in survival by acting as the body’s alarm system ¹. The adaptive value of pain, however, fails when it persists beyond the normal tissue healing time. Chronic pain (i.e., pain lasting ≥ 3 months) is a debilitating clinical condition that causes untold suffering and results in significative restrictions to daily life. Recent reports indicate chronic pain as one of the major sources of disability, a high socioeconomic burden, and hence a serious public health issue ^2,3. Neuroimaging has been increasingly used to identify an objective brain marker or pattern related to pain. The complex nature of pain and the multi-facet manifestations of chronic pain, however, have made the identification of neuroimaging-based biomarkers extremely challenging ¹. Previous studies have identified a few pain processing-related brain regions, including (but not limited to) anterior cingulate cortex, prefrontal cortex, primary and secondary somatosensory cortices, supplementary motor area, insula, amygdala, hippocampus, and striatum ^4–6. The task activation paradigm and the small sample size involved in these studies can often lead to non-consistent findings. Over the past decade, the task-free resting state fMRI (rsfMRI) has been adopted to investigate the neural substrate of pain using the functional connectivity (FC) analysis. While useful for inferring the functional coupling between regions, FC does not directly reflect properties of regional brain activity. In recent years, rsfMRI-based brain entropy (BEN) has been increasingly adopted to characterize the “randomness” or “irregularity” of brain activity ^7,8, with higher BEN reflecting increased randomness (i.e., reduced temporal coherence of BOLD signal time-series) and lower BEN reflecting reduced randomness (i.e., increased temporal coherence of BOLD signal time-series) ⁸.

BEN has been linked to age, education, and neurocognition ^9–11, as well as to brain diseases such as Alzheimer’s disease ¹², schizophrenia ¹³, depression ¹⁴, and substance abuse ^15–17. Notably, electroencephalography-based BEN has been found to correlate with pain ^18,19. Specifically, chronic pain patients were characterized by increased BEN compared to pain-free controls ¹⁸, while medication-sensitive compared to medication-resistant pain patients showed significantly lower BEN in central–parietal regions ¹⁹. In addition, a recent study ²⁰ reported that in middle-to-old-age participants pain severity was positively correlated to rsfMRI-derived BEN in regions involved in the sensory and cognitive processing of pain, possibly indicating reduced efficacy of pain processing and modulation ²⁰.

In this study, data from UK Biobank (UKB) ²¹ were used to investigate BEN differences between chronic pain and non-chronic pain in a sample of middle-to-old age participants. In addition, the relationship between BEN and chronic pain widespreadness, which reflect the extent to which pain is diffused across multiple body areas, was tested. Based on previous findings involving chronic pain patients ¹⁸ as well as middle-aged and elders healthy individuals ²⁰, it is hypothesized that individuals with chronic pain, and similarly individuals with a more widespread chronic pain, will exhibit increased BEN in regions typically involved in pain processing ⁶. Finally, BEN differences between individuals with chronic pain in different body areas and healthy controls were examined. To the best of our knowledge, the sample size included in this study is the biggest relative to what was reported in current imaging-based pain research. This study is also the first to study rsfMRI-derived BEN in chronic pain.

Materials and methods

Data selection and pain assessment

rs-fMRI and sociodemographic data were selected from the UKB ^21,22. UKB has approval from the North West Multi-centre Research Ethics Committee (MREC) as a Research Tissue Bank (RTB) approval. Informed consent was obtained from each participant. Neuroimaging data were originally acquired during two distinct sessions, namely instance 1 and instance 2. Only data from instance 1 were used in the following analyses. Sociodemographic data were selected from instance 1 and included age (field ID 21003), sex (field ID 31), and income (field ID 738) (1 = less than 18,000; 2 = 18,000 to 30,999; 3 = 31,000 to 51,999; 4 = 52,000 to 100,000; 5 = greater than 100,000). Pain was also assessed during instance 1 by means of the following question: “In the last month have you experienced any of the following that interfered with your usual activities? (You can select more than one answer)” (field ID 6159). Possible answers included: headache, facial pain, neck or shoulder pain, back pain, stomach or abdominal pain, hip pain, knee pain, pain all over the body, none of the above, prefer not to answer. An additional question was posed only to individuals who experienced one or more of the previous pain types, asking if the pain lasted for more than 3 months (e.g., a participant who reported to have experienced headache in the last month was asked “Have you had headaches for more than 3 months?”) (field ID 100048). Note that this information was collected for each pain type participants reported to have experienced in the last month. Individuals were assigned to the Chronic Pain (CP) group if they answered at least one time “Yes” to this question (N = 13,132); if they did not experience any type of pain during the last month, they were assigned to the healthy control (HC) group (N = 18,173); if they experienced pain during the last month, but not for more than three months, they were assigned to the Non-chronic Pain (NCP) group (N = 4922). Participants with any missing data were excluded from the analyses as well as those who responded “Prefer not to answer”. Sociodemographic information of each group is reported in Table 1.

Table 1.

Sociodemographic information for chronic pain (CP) group, healthy controls (HC), and non-chronic pain (NCP) group, including age (mean ± standard deviation, range), sex (Males/Females), and income (mean ± standard deviation, range).

	CP (N = 13,132)	HC (N = 18,173)	NCP (N = 4,922)
Age (mean ± std, range)	63.54 ± 7.74 45–82	64.12 ± 7.59 46–82	62.86 ± 7.73 46–82
Sex (M/F)	5,862/7,270	9,154/9,019	2,640/2,282
Income (mean ± std, range)	2.77 ± 1.12 1–5	2.92 ± 1.11 1–5	2.92 ± 1.13 1–5

For CP participants, a pain widespreadness score was calculated based on the number of body areas experiencing pain for more than 3 months, reflecting the extent to which chronic pain was diffused across multiple body areas or more localized. Participants with “pain all over the body” (N = 272) were not considered for the calculation of pain widespreadness, since this category already implied a generally distributed rather than localized pain (also note that all participants suffering from chronic pain all over the body had pain widespreadness = 1, see Table 2). The distribution of participants across different numbers of affected areas was as follows: one body area (N = 7966 participants), two body areas (N = 3152), three body areas (N = 1171), four body areas (N = 423), five body areas (N = 114), six body areas (N = 30), and seven body areas (N = 3). Because very few individuals had CP in six or seven body areas and none who were included in our study had CP in eight body areas, the pain widespreadness score was categorized from 1 to 5 or more.

Table 2.

Number of participants with different types of chronic pain (CP) among individuals experiencing CP in only one body area (pain widespreadness = 1).

	CP (N = 8238)
Knee pain	2736
Back pain	1732
Neck/Shoulder pain	1558
Hip pain	868
Headache	745
Stomach/Abdominal pain	293
Pain all over the body	272
Facial pain	34

rs-fMRI

rs-fMRI data were acquired by instructing participants to keep their eyes fixated on a crosshair, relax and ‘think of nothing in particular’. The run lasted 6 minutes and included 490 timepoints. Acquisition parameters included: TR = 735ms, TE = 39ms, resolution = 2.4×2.4×2.4mm³, number of slices = 64, flip angle = 52°. Data preprocessing was carried out by UKB using FSL packages, and included: motion correction with MCFLIRT, grand-mean intensity normalization, high-pass temporal filter, field map unwarping, gradient distortion correction, and artefacts removal using FSL ICA-FIX. Detailed information about acquisition parameters and pre-processing steps can be found in ²². Before BEN calculation, rs-fMRI data were smoothed using fslmath with a Gaussian filter with full-width-at-half-maximum (FWHM) = 2.5 mm to mitigate the residual inter-brain registration discrepancy.

BEN calculation

BEN maps were obtained for each rsfMRI scan session using the BEN mapping toolbox (BENtbx) ⁸⁠, which is based on the calculation of the sample entropy (SampEn) ⁸. The software used for entropy mapping is freely available from cfn.upenn.edu/zewang/BENtbx.php.

Denote the rsfMRI data of one voxel by $x=[x_1,x_2,\dots x_N]$, where $N$ is the number of time points. SampEn starts with forming a series of vectors, called embedded vectors, each with m consecutive points extracted from $\mathrm{x}:u_i=[x_i,x_i+1,\dots x_i+m-1]$, where $i=N-m+1$, and $m$ is a pre-defined dimension. Using a pre-specified distance threshold $r,B_i^m\left(r\right)$ counts the number of $u_j(j=1,toN-m,andj\ne i)$ whose distances (Chebyshev distance is generally used though any other distance can be used as well) to $u_j$ are less than $r$, so does $B_i^m+1\left(r\right)$ for the dimension of $m+1$. By averaging across all possible vectors, it is obtained:

$$B^m\left(r\right)=\frac{1}{(N-m)(N-m-1)}\sum _{i=1}^{N-m}{B}_i^m\left(r\right)$$

$$A^m\left(r\right)=\frac{1}{(N-m)(N-m-1)}\sum _{i=1}^{N-m}{B}_i^m\left(r\right)$$

And the SampEn is calculated as:

$$SampEn\left(m,r,N,x\right)=-\mathrm{l}\mathrm{n}\left[\frac{A^m\left(r\right)}{B^m\left(r\right)}\right]$$

In summary, BEN was calculated as the logarithmic likelihood that a small section (within a window of a length $m$) of the data that “matches” with other sections will still “match” the others if the section window length increases by 1. A “match” was identified when the distance between two compared time segments was smaller than the threshold $r$. The window length $m$ is widely set to be from 2 to 3. The embedding vector matching cut-off should be selected to avoid “no matching” (when it is too small) and “all matching” (when it is too big) ²³. Both parameters have been assessed in previous publications ^8,24⁠. In this study, a window length of 3 (rsfMRI image volumes or TRs) and a cut-off threshold of 0.6 were adopted ⁸. Before statistical analyses, BEN maps were smoothed with a FWHM of 4 mm.

Statistical analyses

All analyses were performed using SPM12 running on MATLAB v2021a.

To test whether BEN differences existed between CP and HC, a general linear model (GLM) was estimated including BEN maps as dependent variable. The groups of interest were modelled as a dummy variable (CP = 1, HC = 0), with age, sex (F = 0, M = 1), and income as covariates of no interest.

Similar GLMs were estimated to test differences between CP and NCP (CP = 1, NCP =0), as well as NCP vs HC (NCP = 1, HC = 0).

The previously calculated index of pain widespreadness, ranging from 1 to 5, was used to test whether BEN in CP participants was related to a more localized or widespread pain experience. The pain widespreadness score was included in a GLM as regressor, with age, sex, and income as covariates of no interest.

Finally, to explore differences between individuals with CP in different body areas, a subsample including only CP individuals experiencing pain in one body area (pain widespreadness = 1) was considered. The distribution of CP types is reported in Table 2. For each type of CP, the function pairmatch of the optmatch library in R was used to generate a sample of HC matched for age, sex, and income, with a case:control ratio of 1:3 ²⁵. Then, independent-sample T-tests were run to test for differences between CP types and HC.

All results were corrected for multiple comparisons using the family-wise error (FWE) method and considered significant if p_peak < 0.001 and p_cluster-FWE < 0.05.

Results

Compared to HC, CP participants displayed significantly increased BEN in the dorsolateral prefrontal cortex (dlPFC), dorsomedial prefrontal cortex (dmPFC), precuneus, inferior parietal lobules, inferior and middle temporal gyri, occipital cortex, and cerebellum (p_cluster-FWE < 0.05) (Figure 1, Table S1).

Figure 1. Chronic Pain vs Healthy controls.

Differences in brain entropy (BEN) between participants with chronic pain (CP) and healthy controls (HC). Results are showed overlaid on the standardized inflated surface (dorsal view at the center) and volumetric image (axial view). Results were considered significant if ppeak < 0.001 and pcluster-FWE < 0.05. The colorbar is based on T-values. Positive T-values (red-yellow) indicate higher BEN in CP with respect to HC. Details of the significant clusters are presented in Table S1. A = Anterior; P = Posterior; L = Left; R = Right.

No significant effects were observed between CP and NCP as well as between NCP and HC (p_cluster-FWE > 0.05).

The widespreadness of pain was related to increased BEN in the occipital lobes (p_cluster-FWE < 0.05) (Figure 2, Table S2).

Figure 2. Pain Widespreadness.

Effect of pain widespreadness on brain entropy (BEN). Results are showed overlaid on the standardized inflated surface (dorsal view at the center) and volumetric image (axial view). Results were considered significant if p_peak < 0.001 and p_cluster-FWE < 0.05. The colorbar is based on T-values. Positive T-values (red-yellow) indicate a positive association between BEN and pain widespreadness. Details of the significant clusters are presented in Table S2. A = Anterior; P = Posterior; L = Left; R = Right.

Comparisons of individuals with CP in different body areas with matched HC revealed that only participants with chronic headache and pain all over the body were significantly different from HC (p_cluster-FWE < 0.05). Compared to HC, individuals with chronic headache displayed higher BEN in the superior/inferior parietal and occipital cortex (Figure 3, Table S3). Moreover, individuals with chronic pain all over the body showed higher BEN in the dorsolateral/medial frontal regions, left superior parietal lobule and precuneus, as well as temporal and occipital cortices (Figure 4, Table S4).

Figure 3. Chronic Headache vs Healthy Controls.

Brain entropy (BEN) differences between individuals with chronic headache and a matched sample of healthy controls. Results are showed overlaid on the standardized inflated surface (dorsal view at the center) and volumetric image (axial view). Results were considered significant if p_peak < 0.001 and p_cluster-FWE < 0.05. The colorbar is based on T-values. Positive T-values (red-yellow) indicate higher BEN in participants with chronic headache. Details of the significant clusters are presented in Table S3. A = Anterior; P = Posterior; L = Left; R = Right.

Figure 4. Chronic Pain all over the body vs Healthy Controls.

Brain entropy (BEN) differences between individuals with pain all over the body and a matched sample of healthy controls. Results are showed overlaid on the standardized inflated surface (dorsal view at the center) and volumetric image (axial view). Results were considered significant if p_peak < 0.001 and p_cluster-FWE < 0.05. The colorbar is based on T-values. Positive T-values (red-yellow) indicate higher BEN in participants with chronic pain all over the body. Details of the significant clusters are presented in Table S4. A = Anterior; P = Posterior; L = Left; R = Right.

Discussion

In this study, rs-based BEN was used to investigate the brain functional correlates of chronic pain in a cohort of over 30,000 middle-to-old-age participants. Our main finding showed that, compared to HC, individuals with CP exhibited greater BEN in several brain regions including PFC, inferior and middle temporal cortex, inferior parietal lobules, precuneus, occipital cortex, and cerebellum.

The PFC is a functionally heterogeneous area that serves different high-level brain functions such as top-down modulation ^26–28, performance monitoring ²⁹, attention ^30,31, working memory ³², social cognition ³³, and emotion regulation ^34–36.

Previous neuroimaging studies have evidenced the role of the PFC in different aspects of pain processing, such as pain suppression and inhibition ^37–42, cognitive modulation of pain ^4,43, and pain catastrophizing ^38,44.

Moreover, several works point towards altered activity, including brain blood flow ^45–47, oscillatory activity ^48,49, task-related activity ^50,51, and resting-state functional connectivity ^6,52–54, as well as abnormal structure of the PFC (e.g., reduced gray matter volume; GMV) ^55,56 in patients with chronic pain. Changes in the activity and structure in the PFC have been proposed to represent a marker of successful intervention for clinical pain conditions ^37,38.

Increased BEN indicates randomness and reduced auto-correlation of brain BOLD signal over time ⁸. Notably, prior findings indicate that a more regular and coherent brain activity (i.e., reduced BEN) in regions of the executive control network, including the PFC, is correlated to higher fluid intelligence and better functional task performance ¹⁰. Moreover, [46,57] preliminary results indicate that BEN of the PFC is also negatively correlated with GMV and surface area ⁵⁷.

Increased randomness of brain activity (i.e., greater BEN) in individuals with CP might reflect a reduced pain inhibition and modulation capability. Consistently with this idea, different works have reported poorer executive functions in CP patients, although the directionality of this relationship is not clear ^58,59. Interestingly, these results are also consistent with a recent study where increased BEN in several brain regions, including the PFC, was correlated to pain severity in a cohort of middle-aged participants and elders, although the data in that study did not allow to characterize the type of pain experienced (acute or chronic) ²⁰.

Notably, the PFC has been considered a promising therapeutic target for treating chronic pain through, e.g., non-invasive brain stimulation ^{37,38,43,60,61}. Recent studies also indicate that BEN is modifiable through repetitive transcranial magnetic stimulations ^15,62–64, aiding for the potential translational value of BEN measurement in chronic pain.

In addition to the PFC, increased BEN was observed in other regions related to pain processing, such as precuneus, inferior parietal lobule, temporal cortex, occipital cortex, and cerebellum. The cerebellum might be involved in the integration of multimodal aspects of pain processing due to the presence of both ascending and descending pain-related pathways ^65–67. The precuneus and inferior parietal lobules are key regions of the default mode network, whose alteration in individuals with CP has been repeatedly reported ^68,69. Although not well understood, some evidence also indicates the involvement of the occipital cortex in chronic pain ^70–73. It has been proposed that the activation of visual networks in chronic pain may result from a compensatory plasticity response of the brain to long-term pain and might serve, for example, as a distraction by diverting one’s attention away from pain ⁷¹. Interestingly and possibly consistently with these findings, a more widespread rather than localized pain was related to increased BEN in the occipital cortex, perhaps reflecting a higher cognitive load to deal with a more diffused pain ⁷⁰. The functional meaning of these findings, however, is less clear and should be further investigated in future studies.

Compared to HC and CP participants, individuals with NCP did not show any BEN difference. In addition to the smaller sample size of the NCP group, it is worth noting that information in the dataset (e.g., duration of the pain experience over the one-month period) was not sufficient to adequately characterize the pain experience of these individuals.

Finally, compared to matched HC, individuals with chronic headache showed increased BEN in parietal and occipital regions, while individuals with chronic pain all over the body displayed higher BEN in the PFC, temporal, and occipital lobes.

Even though the information included in the dataset did not enable to distinguish between individuals with migraine with or without aura, dysfunctions of the visual networks (including not only occipital regions, but also parietal areas as part of the dorsal visual stream) have been repeatedly observed in individuals with migraine ^74,75. Increased BEN might reflect hyperexcitability of these regions, as suggested in a previous work ²⁰.

Participants with chronic pain all over the body included in this study might be assimilated to individuals with fibromyalgia, a chronic pain disorder characterized by widespread musculoskeletal pain and tenderness. However, this comparison should be considered with caution due to the lack of more detailed clinical data. The pattern of brain regions displayed in Figure 4 is highly similar to that observed in Figure 1, depicting BEN differences between participants with CP and HC. While all considerations made previously regarding potential mechanisms underlying increased BEN in individuals with CP remain valid, it is interesting to note that several studies have highlighted the importance of the PFC in fibromyalgia. For instance, fibromyalgia patients showed increased high and low oscillatory activity in the PFC ⁷⁶ as well as lower GMV ⁷⁷. In addition, there is evidence supporting the efficacy of repetitive transcranial magnetic stimulation of the PFC in the treatment of fibromyalgia ^78,79.

This study has several limitations. First, chronic pain was self-reported and not clinically assessed. In addition, despite the considerable size of the sample, chronic pain includes a variety of heterogeneous disorders that was not possible to properly characterize given the available data. The null effects found between most of CP types (except headache and pain all over the body) and HC highlight the importance of investigating specific types of CP rather than relying on the broader CP classification. Finally, given the cross-sectional nature of this work, longitudinal studies are needed to understand the temporal dynamics of the relationship between BEN and CP.

Conclusions

In this study, it was examined the relationship between BEN and CP in a sample of over 30,000 participants. Compared to HC, individuals with CP showed increased BEN in several brain regions including the frontal, temporal, and occipital lobes, as well as the cerebellum. Based on previous findings showing involvement of the PFC in the cognitive modulation of pain ³⁸, impairment of executive functions in individuals CP ⁵⁹, correlation between BEN and both neurocognition ¹⁰ and GM morphology (Del Mauro et al., 2024), it was hypothesized that increased BEN in the PFC might be related to a compromission of pain modulation and inhibition capability in individuals with CP. The widespreadness of CP (i.e., more diffused or localized pain) was related to increased BEN in the occipital cortex, possibly reflecting a higher cognitive load to deal with a more diffused pain. When examining individuals with CP in different body areas, only participants with headache and pain all over the body displayed significant differences from matched HC. While providing interesting and meaningful insights into the neural mechanisms underlying these disorders, these results also highlight the importance of focusing on specific types of chronic pain.

Data Statement

Data used in this study are publicly available from the UK Biobank (https://www.ukbiobank.ac.uk/).