Chronic pain afflicts over 20% of the world’s population 4 and has massive economic and societal ramifications. Debilitating pain can hijack patients’ lives, preventing them from freely being able to work, exercise, and socialize. Chronic low back pain (CBP), for example, is the leading cause of disability in the world 5. Until the mid-2010s, prescription opioids were commonly used as a first-line treatment for chronic pain, making chronic pain a natural entry point for opioid use and, thus, potential downstream misuse, opioid use disorder (OUD), and even overdose 6-8. Despite marked decreases in opioid prescription rates within the healthcare system over the past ten years, opioids are still used by 20–50% of patients suffering from chronic (non-cancer) pain (around 15 million patients in the US) 1-3. The neuropsychological implications of long-term opioid use in patients with chronic pain are largely unknown.
Chronic non-cancer pain (the population of interest throughout this study) and opioids have been extensively studied on their own. Each imparts a massive societal cost 9,10, and their combination may impart an even greater cost. Our group has demonstrated that the development of chronic pain involves mesocorticolimbic (MCL) plasticity 11-16. Interestingly, these same circuits are strongly implicated in OUD 17,18. From a high level, MCL helps code emotional valence and modulate behavioral responses, such as appetition and aversion 19-21. Pain is an aversive experience, and patients actively avoid triggers. In contrast, opioids, while commonly said to be rewarding, may induce aversion and negative affect during periods of withdrawal—so-called hyperkatifeia 22. As a result, the mechanisms of chronic pain may interact with long-term opioid use, exacerbating patients’ pain experience when off opioids to create a vicious cycle. However, there is little data on the impact of long-term opioid consumption in chronic pain patients 23-28.
In this work, we aimed to answer three questions: (1) Is long-term opioid prescription use in patients with chronic pain associated with better (or worse) clinical outcomes? (2) Is long-term opioid use in patients with chronic pain associated with differential brain activity and structure, especially within the MCL system? And (3) how do the results of (1) and (2) inter-relate? To answer these questions, we first compared the clinical characteristics, brain anatomy, and brain function of CBP prescribed long-term stable doses of opioids (CBP+O, n=70; mean = 7.8 years of opioid consumption) to CBP patients managing their pain without opioids (CBP−O, n=70). The two groups were one-to-one matched (from a 150 CBP−O dataset) based on their age, sex, pain intensity (mean = 5.4 on a 0–10 numeric rating scale), and pain duration (>0.5 years, mean = 17 years) (Table S1). We reasoned that long-term opioid use should also lead to a reorganization of the relationship between neurotransmitter receptor-related activity, given that opioid exposure leads to receptor desensitization and tolerance 29, which in turn must interact and modify the influence of other receptors on neuronal activity. Moreover, we recently showed that perceptual states, especially pain perception, engage neuronal activity throughout the brain 30, prompting us to develop an approach to examine whole-brain multi-receptor-related adaptations. Therefore, we used the neurotransmitter atlas recently developed by Hansen et al. 31 and modified their methods to relate whole-cortex activity for long-term opioid exposure to the chemoarchitecture of the human brain, based on the brain normative map for 19 receptors and transporters across 9 different neurotransmitter systems. We then compared receptor-related activity (i) between subgroups of CBP+O (high and low opioid consumption), (ii) within CBP+O subjects before and after brief opioid abstinence, and (iii) between CBP+O subjects who tapered or did not taper opioid use. With back translational, we validated the results in a mouse model (sham or chronic pain) before and after the mice self-administered vaporized fentanyl for multiple weeks. Our results lead to a novel multi-receptor model for long-term opioid use and identify a specific target receptor and a molecule that may aid opioid tapering.
Clinical phenotypic dimensions of long-term opioid use
Chronic pain and opioid use are independently associated with multiple psychological comorbidities, including worse negative affect, sleep disturbance, and diminished social interactions 8,18,32. Are these outcomes worse in patients managing their chronic pain with long-term opioid consumption? Currently, there is little data to answer this question 23,33-35. There is clinical equipoise: On one hand, prescription opioids may improve patients’ psychological characteristics via pain relief. On the other hand, if prescription opioids enhance the likelihood of OUD, they may exacerbate rather than relieve the psychological challenges associated with chronic pain, for example, prescription opioids are associated with worse depression 36.
Using validated self-report questionnaires, we cross-sectionally compared physical function, pain, and mental health between CPB+O and CBP-O. Compared to CBP-O, CBP+O exhibited worse outcomes for 7 of the 17 measures; only anxiety (PROMIS) was similar, and all other outcomes tended to show worse values in CBP+O relative to CBP−O, albeit with appreciable uncertainty (Fig. 1a, Table S2). We performed a principal components analysis (PCA) to reduce the dimensionality of these outcomes. Three principal components (PC1–3; , variance > 10%) explained 69% of the total variance: PC1 (which we call functional disability) included decreased physical function, less social activity, increased general disability, greater pain interference, and greater fatigue; PC2 (pain quality) was mainly composed of pain descriptors, including sensory and affective pain scales, neuropathic pain symptoms, and pain catastrophizing; PC3 (negative affect) included increased depression, anxiety, negative affect, decreased mental well-being, and less sleep (Fig. 1a, Table S3). These PCs were stable no matter how they were derived (Fig. S1, Table S3) and closely resemble the results we have previously reported 23.
Group (CBP+O vs. CBP−O) differences for the three PC scores were assessed using analysis of covariance (ANCOVA) with sex, race, age, pain intensity (NRS), pain duration (log), body mass index (BMI), and medication quantification scale (MQS) included as covariates. CBP+O exhibited higher functional disability (PC1) and worse pain quality (PC2) scores compared to CBP−O. The groups had similar negative affect (PC3) (Fig. 1b, Table S4).
Even though back pain intensity was matched between CBP−O and CBP+O and PC2 was correlated with back pain intensity in both groups, the consumption of non-opioid medications (MQS) was twice that in CBP+O compared to CBP−O and correlated with back pain intensity only in CBP+O (Fig. S2). Overall, the clinical phenotyping indicates that CBP+O 1) report more severe pain qualities, particularly neuropathic-like pain with a larger sensory component; 2) show higher functional disability; 3) do not differ in negative affect from non-opioid users, although their depression ratings were one standard deviation higher than CBP-O; and 4) consume more non-opioid medications than those not on opioids, in proportion with their reported pain intensity. Although opioid consumption was not associated with an improvement in any of the 17 clinical parameters examined, it is essential to note that many of the differences between CBP+O and CBP−O were modest.
Relating opioid use outcomes to clinical phenotypes of long-term opioid use
Opioid dosage is a strong determinant of medication safety and is considered when deciding to taper or reduce prescribed dose(s) 37. We examined how opioid use in CBP+O relates to clinical measures, including withdrawal symptoms and misuse risk. Opioid use was quantified using three measurements: 1) the daily prescription converted into morphine milligram equivalent (MME); 2) blood levels of opioids converted to a relative opioid equivalent (ROE, mg/L); and 3) the duration of opioid use (DOU) (Fig. 1c). MME was used to subdivide CBP+O by OUD risk as defined by the Centers for Disease Control and Prevention (CDC) guidelines 38 (Fig. 1c top panel). All three measures—MME, ROE, and DOU—were right-skewed and thus log-transformed. Blood opioids (ROE) in CBP+O reflected their prescription dose (MME) (r=0.48, p<0.001; Fig. 1d-e scatter plot), but neither blood opioid levels (ROE) nor prescription dose (MME) was strongly associated with the duration of opioid use or with non-opioid medication use (MQS) (Fig. 1d).
We assessed patients’ withdrawal symptoms and misuse risk using validated scales (subjective opiate withdrawal scale, SOWS; current opioid misuse measure, COMM). On these scales, 19 CBP+O (27%) showed high misuse risk (COMM > 9) and 7 (10%) high withdrawal symptoms (SOWS > 20). We investigated these scales’ relationships with back pain intensity, functional disability (PC1), pain quality (PC2), and negative affect (PC3). Opioid use characteristics (MME, ROE, and DOU) were not consistently associated with withdrawal or misuse. Only ROE showed a statistically significant positive correlation with functional disability (r=0.47, p<0.01; Fig. S3). There was no statistically significant association between opioid usage parameters with pain intensity, pain quality, or negative affect scores (PC1–3, Fig.1f).
Brain structure with long-term opioid use
We next examined the impact of long-term opioid use on brain structure. Chronic pain and OUD are each associated with global and local gray matter reorganization 39-42. We computed normalized peripheral gray matter volumes (PGMV) for all participants. After adjusting for covariates of no interest (age, sex, race, NRS, pain duration, BMI and MQS), CBP+O showed lower PGMV than CBP−O (p<0.05) (Fig. S4a, Table S5). However, PGMV was not related to opioid use (MME, ROE, and DOU), clinical parameters (PC1–3), or withdrawal and misuse scales (Table S6). We also investigated subcortical volumes, which did not statistically significantly differ between CBP+O and CBP−O (Table S7).
We used whole-brain voxel-based morphometry (VBM) to investigate regional grey matter density (GMD) differences between CBP+O and CBP−O. After adjusting for covariates of no interest, we found two clusters localized to (1) the left primary sensorimotor cortex (S1/M1) and (2) the mid-anterior cingulate cortex (mACC), which showed decreased GMD in CBP+O. The S1/M1 cluster was associated with sensorimotor function, while mACC with pain, nociception, and arousal, using Neurosynth reverse inference decoder (Fig. S4b, Table S8). In addition, mACC GMD was negatively correlated with both back pain intensity (p=0.02) and duration (p=0.01) in both groups (Fig. S5, Table S9). Similar to our PGMV analysis, localized GMD changes in mACC and S1/M1 were not statistically significantly associated with opioid use, withdrawal, misuse, or clinical parameters (PC1–3) in CBP+O patients (Table S10). Overall, whole-brain gray matter decreases were modest, and focal decreases in cortical gray matter volume were associated with long-term opioid use in CBP+O.
To examine the impact of opioid exposure on the white matter, we contrasted a subsample of the subjects, 58 CBP+O and matched 58 CBP−O, white matter properties using a whole-brain skeletal fractional anisotropy (FA) contrast, which did not yield any statistically significant differences between the two groups (data not shown).
Brain activity changes with long-term opioid use
A primary hypothesis of this study was that the groups will differ in ongoing brain activity. We used used resting state fMRI signal to test this hypothesis. The power spectrum of brain activity signals is related to various brain functional properties 43,44. We used resting-state fMRI to localize voxel-wise differences in the amplitude of low-frequency fluctuations (ALFF, which reflects the low-frequency energy of spontaneous neural activity 45) between CBP+O and CBP−O while adjusting for age, sex, pain intensity, pain duration, BMI, MQS, and head motion (ΔALFF, CBP+O > CBP-O). We also included voxel-wise corrections for scanner signal-to-noise ratios and GMD. Compared to CBP−O, CBP+O showed greater ALFF in five distinct clusters, the largest of which (5,050 voxels) encompassed multiple MCL structures, including bilateral nucleus accumbens (NAc), amygdala, subgenual cingulate, medial/orbital prefrontal cortex, hippocampus, and brain stem. Other clusters in which CBP+O had greater ALFF included the lateral occipital cortex, the middle prefrontal cortex, and the right and left posterior portions of the inferior temporal gyrus. CBP+O patients had lower ALFF in 3 clusters, including the left dorsolateral prefrontal cortex (dlPFC, 3,955 voxels) and the right and left anterior part of the mid-temporal gyrus (Fig. 2a, Table S11). Greater ALFF in CBP+O was localized to brain regions involved in reward, motivation, incentive, and value processing, while lower ALFF in CBP+O was localized to brain regions involved in language and mental states (Fig. 2b). Similar to our morphological results, localized ALFF changes were not statistically significantly associated with opioid use, withdrawal, misuse, or clinical parameters (PC1–3) in CBP+O patients (Table S12).
Relating cortical receptor distributions to brain activity
Given that the CBP+O cyclically and daily perturb their brain molecular properties by regularly ingesting opioids and also our recent evidence shows that pain perception is better conceptualized as a whole-brain process 30, we developed a new methodology to relate the brain activity of regular opioid use to cortical receptor distribution-dependent activity. We studied how receptor density distributions mapped onto CBP+O vs. CBP−O activity differences. Specifically, how much of the difference in brain activity (ΔALFF between CBP+O and CBP−O) can be explained by neurotransmitter receptor densities? To address this issue, we used multiple regression to model ΔALFF using 19 neurotransmitter receptor density maps when the cortex is divided into 100 regions (constructed by amalgamating PET images from 1,238 healthy participants across nine neurotransmitter systems 31) (Fig 2c). The receptor density maps explained 51% of the variance of ΔALFF (p<0.01)(Fig 2d). The serotonin (5-HT1A and 5-HT1B) receptors and the µ-opioid (MOR) receptor accounted for the largest explanatory variance (ΔR2 for 5-HT1A receptor map was 19.6%, for 5-HT1B it was 17.6%, and for MOR it was 13.7%; together their ΔR2 was 27.4%). After adjusting for the remaining receptors, the 5-HT1A and 5-HT1B receptor distributions in the cortex were strongly negatively correlated with ΔALFF, while the MOR receptor cortical distribution was strongly positively correlated with the ΔALFF (Fig 2e). The strength of these correlations highlights that their influence on ΔALFF follows an almost uniform proportionality throughout the cortex (e.g., the 5-HT1A receptor expression anywhere in the cortex accounts for a decrease of ~2/3 AU in local ALFF).
Whole-cortex receptor-related activity distinguishes opioid use and reflects clinical phenotypes
The spatial uniformity of the relationship between ΔALFF and all three receptor density maps enables the derivation of a unitary measure of receptor-related activity across the cortex. We used this measure to test for group differences and to examine relationships with clinical characteristics. Cortex-wide receptor-related activity for each subject was computed as the normalized dot product between regional ALFF and the corresponding receptor density distribution. This resulted in one value per subject representing cortical receptor-related activity, which we designate as 5-HT1AR*ALFF, 5-HT1BR*ALFF, and MOR*ALFF. Overall, across healthy subjects and CBP−O and CBP+O, the 5-HT1AR*ALFF and 5-HT1BR*ALFF were positive (enhancing brain activity, or excitatory), while the MOR-specific activity (MOR*ALFF) was negative (decreasing brain activity, or inhibitory). Both serotonin-related activities were less in CBP+O than in CBP−O and healthy controls, while MOR-related activity was higher (less inhibition) in CBP+O (Fig 3b). 5-HT1AR*ALFF and 5-HT1BR*ALFF were positively correlated with each other and negatively with MOR*ALFF in both patient groups, but not in the healthy (Fig 3c).
We examined the relationship between receptor-related activity and clinical parameters using multiple regression analysis. Across all 140 CBP-O and CBP+O patients, 5-HT1AR*ALFF was negatively related to the PC1-functional disability (Fig 3d), while MOR*ALFF was positively related to the PC3-negative affect (Fig 3e). PC2-pain quality was not related to any of the three receptor-related activities. These results can be summarized in a model with three nodes (Fig 3f), which shows that long-term opioid use predominantly impacts serotoninergic activity and partially renormalizes opioidergic activity.
Receptor-related activity responses to various modulations of opioid exposure
Our results so far demonstrate a strong relationship between the three receptor-related activities and the states of CBP-O and CBP+O. To establish a causal relationship concerning how these receptors interact, we assessed their response to various opioid conditions and perturbations. We investigated 5-HT1A, 5-HT1B, and MOR-related brain activity in 4 different conditions: (1) high vs. low opioid consumption, (2) before and after brief abstinence from opioid use, and (3–4) responses to a non-pharmacological, multi-disciplinary intervention leading to either successful (3) or unsuccessful (4) tapering.
1. High vs. low opioid consumption: The effect of opioid consumption dose was determined by computing ΔALFF between high opioid (n =12 , MME > 50) and low opioid ( n=12, MME < 20) CBP+O patients matched for age, sex, pain intensity, and duration as well as opioid use duration (Table S13). Localized regions that showed ALFF changes with long-term opioid use (CBP+O > CBP-O) did not statistically significantly differ between high and low opioid users (Table S14). Despite the paucity of salient local differences, ΔALFF strongly negatively related to 5-HT1A and 5-HT1B and positively correlated with the MOR density map (Fig 4a, first row). In addition, patients with high doses of opioids showed lower 5-HT1AR*ALFF and higher MOR*ALFF than patients on low doses of opioids (Fig 4b). A pattern that closely recapitulates what we observe in the overall group of CBP+O in comparison to CBP-O (Fig 3f).
2. Brief opioid abstinence: In a subsample of 14 CBP+O patients, we perturbed their stable opioid status by requesting the participants who were taking short-acting opioids only to briefly refrain from taking their opioids overnight (19.4 ± 6.7 hours) and undergo a second brain resting state scan. Blood samples confirmed patients’ abstinence, as there were no or minimal opioids detected in the blood, especially in comparison to their first scan, which was collected within 3.01 ± 2.75 hours of opioid consumption (Table S15). Pain intensity did not differ between baseline and abstinence, but signs of withdrawal increased with abstinence (Table S15). Most brain regions that showed significant ALFF differences between CBP+O and CBP-O showed minimal changes following opioid abstinence. Only the left pITG showed decreased ALFF, while the right aMTG showed increased ALFF following abstinence (Table S16). The within-subject ΔALFF between before and after abstinence was strongly negatively associated with 5-HT1A receptor distribution and positively associated with the MOR receptor distribution in the cortex (Fig 4a, second row). In addition, opioid abstinence was associated with a decrease in 5-HT1A R*ALFF and an increase in MOR*ALFF, but no changes in 5-HT1BR*ALFF (Fig 4c). This result sheds light on the temporal dynamics of opioid exposure’s effects on the brain, suggesting that 5-HT1B receptor-related activity is a consequence of long-duration adaptations.
3. Successful, and 4. Unsuccessful tapering: we examined the effect of opioid tapering in 21 CBP patients following a 4-week multi-disciplinary non-pharmacological chronic pain rehabilitation program (Table S17). Out of the 21 patients, 8 patients decreased opioid use by more than 30% (tapered group), while 13 patients showed minimal/no change in medication use (non-tapered group, Table S18). Both groups’ pain decreased following the treatment (Fig 4a, last column, rows 3,4, Table S18), and most brain regions that showed significant ALFF differences between CBP+O and CBP-O showed minimal changes with treatment except for left pITG (Table S18). In the patients who tapered their opioid use, within-subject ΔALFF between before and after treatment tightly negatively correlated with 5-HT1A and 5-HT1B receptor distributions and positively correlated with MOR receptor distribution in the cortex (Fig 4a, third row). In addition, post-treatment, we observed increased 5-HT1A R*ALFF and 5-HT1BR*ALFF values and decreased MOR*ALFF values, as compared to the pre-treatment values (Fig 4d). In contrast, patients who did not taper their opioid use showed minimal to no changes in the three receptor-based parameters (Fig 4a,e).
Brain-wide receptor-specific adaptations to opioid exposure in a mouse model
To establish across-species and experimental generalizability of our results, we tested whether the three receptor-based adaptations can be observed in a mouse model of long-term opioid exposure. We collected resting-state brain activity in 9 chronic neuropathic pain model mice (spared nerve injury, SNI) and 7 sham injured mice before and after 20 days of daily fentanyl vapor self-administration (fentanyl exposure started 1 month after the peripheral injury). Pain-like behavior was assayed by testing tactile withdrawal thresholds for the injured paw in SNI and Sham mice. Withdrawal thresholds for the injured paw were lower in SNI compared to sham and did not change following fentanyl (tested 24 hours after fentanyl vaping) (Fig S6a).
We obtained the mouse brain expression profiles from the Allen Atlas 46, for 18 of the 19 receptors and transporters that we studied in humans. We calculated within mouse ΔALFF separately for Sham and SNI animals (Fig S6b) and modeled the result in multiple regression using all 18 receptor expression maps for the entire brain (92 regions). Similar to our human results, in SNI animals (chronic pain model) ALFF changes following fentanyl exposure were negatively related to 5-HT1B and positively related to MOR expressions in the brain. However, ΔALFF was not associated with 5-HT1A in SNI mice (Fig S6c), which may be at least partially attributable to the effects of anesthesia on serotonin-related activity 47. In contrast to SNI, ΔALFF changes in sham animals were only related to MOR, but not to 5-HT1A and 5-HT1B (Fig S6c).
A molecular target to modulate brain-wide receptor-specific activity
Our perturbations show that serotonin receptor- and MOR-related activities are divergent. Such results imply that the 5-HT1AR neuronal population may inhibit the MOR population. We tested this concept in 5 healthy volunteers by examining brain activity ΔALFF (post – pre) after ingesting a single dose of a 5-HT1A agonist (vortioxetine). In all subjects, we observed increased 5-HT1A R*ALFF and decreased MOR*ALFF, and no change in 5-HT1BR*ALFF (Fig 5a). The specificity of the result is remarkable, as 15 of the remaining 16 receptor-related activities remained unchanged (only dopamine D1R*ALFF changed with vortioxetine ingestion). The data confirms our hypothesis and suggests that 5-HT1AR neuronal populations compete with MOR neuronal populations. Thus, serotonergic agonists may counteract the effects of long-term opioid use and help in opioid tapering and even in improving chronic pain.
A minimal receptor-specific model of long-term opioid use
Our results can be summarized in a circuit motif model of a receptor-specific brain-wide dynamical system of long-term opioid exposure (Fig 5b). The primary concept is that long-term opioid use in chronic pain patients drives cortical (a) hyperactivity of opioidergic circuits and (b) hypoactivity of cortical serotonergic (5-HT1A R and 5-HT1bR) circuits. It is well-established that opioid use drives mu-opioid receptor (MOR) downregulation. Since MORs are inhibitory, downregulation hyperactivates MOR-containing neurons. We observed that this hyperactivity is associated with blunted 5HTergic activity. The competitive interaction between MOR and 5-HT1A R neuronal pools is rapid (within a day), as we observed in our abstinence and vortioxetine results, while longer-term adaptations (weeks) control the 5-HT1BR neuronal pool interaction with MOR and 5-HT1A R neurons, as we saw for successful tapering.