Investigating Neuroepigenetic Alterations in Chronic Low Back Pain with Positron Emission Tomography

Abstract

Epigenetics has gained considerable interest as potential mediators of molecular alterations that could underlie the prolonged sensitization of nociceptors, neurons, and glia in response to various environmental stimuli. Histone acetylation and deacetylation, key processes in modulating chromatin, influence gene expression; elevated histone acetylation enhances transcriptional activity, whereas decreased acetylation leads to DNA condensation and gene repression. Altered levels of histone deacetylase (HDAC) have been detected in various animal pain models, and HDAC inhibitors have demonstrated analgesic effects in these models, indicating HDACs’ involvement in chronic pain pathways. However, animal studies have predominantly examined epigenetic modulation within the spinal cord after pain induction, which may not fully reflect the complexity of chronic pain in humans. Moreover, methodological limitations have previously impeded an in-depth study of epigenetic changes in the human brain. In this study, we employed [¹¹C]Martinostat, an HDAC-selective radiotracer, positron emission tomography to assess HDAC availability in the brains of 23 chronic low back pain (cLBP) patients and 11 age- and sex-matched controls. Our data revealed a significant reduction of [¹¹C]Martinostat binding in several brain regions associated with pain processing in cLBP patients relative to controls, highlighting the promising potential of targeting HDAC modulation as a therapeutic strategy for cLBP.

Introduction

Chronic pain is characterized as persistent pain lasting longer than three months which further unfavorably affect the quality of life [8,52]. In the United States, low back pain affects 80% of adults, and 15–20% of LBP sufferers develop chronic LBP (cLBP), causing significant suffering and socio-economic burden [12,32,53]. Although prescribing opioids for pain management is not preferred, moderate-to-severe pain, including cLBP, remains one of the leading causes of opioid prescription [11,29] because nonpharmacologic treatments or nonsteroidal anti-inflammatory drugs have been reported to be less effective than opioids-based options [20]. Moreover, prescription opioids have undesirable side effects, such as constipation, respiratory depression, and the potential for abuse and dependence [4,29,31]. Therefore, identifying alternatives to current treatment options is an urgent clinical need that can be achieved by advancing our comprehension of pathophysiological mechanisms underlying chronic pain.

Nociceptors within the peripheral nervous system (PNS) and neurons and glial cells in the central nervous system (CNS) undergo a process of sensitization in response to environmental stimuli like tissue damage, inflammation, and infection [27,54]. These stimuli provoke morphological/functional changes based on sustained molecular alterations associated with prolonged pain [16]. Such changes are often amplified by self-perpetuating mechanisms that modulate gene expression [16]. Within this context, the field of epigenetics, which includes DNA methylation, histone acetylation, and non-coding RNA interactions, has become increasingly relevant due to its role in regulating gene expression without altering the underlying DNA sequence. Histone acetylation and deacetylation, as determine by the interplays between histone acetyltransferases and histone deacetylases (HDACs), govern the extent of chromatin relaxation. Elevated histone acetylation enhances transcriptional activity, while reduced acetylation leads to DNA condensation and gene repression. Different levels of HDAC expression and histone acetylation have been observed across various animal models of pain, reflecting dynamic epigenetic regulation [6,9,16,17,28,33,49]. The analgesic effects of HDAC inhibitors in these models [2,9,14] underscore the relevance of HDACs in neuroplastic changes and highlight their potential as therapeutic targets for pain management [9,15,16,25,40]. However, research on histone modifications in response to pain within the CNS is still in its early stages [6], with most research to date focusing on the spinal cord in animal models. It is important to note that these animal models may not fully recapitulate chronic pain in humans. Methodological limitations have impeded explorations of epigenetic changes in the human brain.

Recently, the availability of HDACs can be quantified using an HDAC-selective radiotracer, [¹¹C]Martinostat, with positron emission tomography (PET). [¹¹C]Martinostat exhibits selectivity for a subset of class I/IIb HDACs (class I: HDAC 1, 2, 3 and class IIb: HDAC 6) [39,48,50,51]. [¹¹C]Martinostat showed high binding and appropriate characteristics for imaging, as shown in our previous studies [18,50]. Importantly, the regional distribution of [¹¹C]Martinostat binding aligned well with their relative ranking of HDAC expression levels as measured in post-mortem tissue samples [18,50]. This study aims to explore potential alterations in HDAC availability within the brains of cLBP patients relative to age- and sex-matched healthy controls. Post hoc analyses further examine associations between altered HDAC levels and patient-reported outcomes.

Methods

Study participants

Twenty-three cLBP patients (11 males and 12 females; age: 46.0 ± 14.7 years old) and 11 healthy controls (8 males and 3 females; age: 48.5 ± 17.6 years old) were included in this study. Participants provided written informed consent, which was approved by the Institutional Review Board and the Radioactive Drug Research Committee at Massachusetts General Hospital, to take part in the study. Study data were collected and managed using REDCap (Research Electronic Data Capture) electronic data capture tools hosted at Massachusetts General Hospital [21,22]. Healthy controls were matched to cLBP patients by age (within two years), sex, race, and exclusion criteria, except they have no documentation of low back pain via medical records. Potential cLBP subjects were excluded if they experienced psychiatric impairment that could prevent completion of study procedures, had an active substance use disorder based on DSM-V criteria, had a history of myocardial infarction or other cardiovascular conditions, underwent surgery within six months of contact, tested positive on a urine drug screen (with the exception for cannabis), experienced peripheral neuropathy, migraines, or fibromyalgia, or were contraindicated for magnetic resonance imaging (MRI) or PET scanning. For their pain management, nine opioid-naive cLBP patients reported the use of NSAID medications at the time of screening, and three cLBP patients reported the use of gabapentin for nerve control and back pain, but no subjects reported the use of opioid-based analgesics, which is confirmed by a negative urine drug test. In addition to the above criteria, opioid-naive cLBP patients were excluded if they did not experience moderate-to-severe pain (i.e., above 4/10 on a visual analog scale), if they reported back pain resulting from cancer or bodily injury, had extensive back surgery that would complicate any interpretation of their data, underwent interventional procedures for back pain within two weeks of contact, or could not provide documentation of low back pain via medical records or a note from a clinician.

Each cLBP subject underwent a comprehensive medical history, physical examination, and low back pain evaluation by a medical doctor or nurse practitioner during enrollment. The clinician inquired about the approximate date of pain onset and duration, as well as any associated injuries or acute trauma. Each cLBP subject should have at least 6 months of pain duration. The examination characterized each subject’s pain as axial, pseudo-radicular, or radicular, and determined affected dermatome (if radicular or pseudo-radicular) or location. Current subjective pain levels (0 = “no pain” and 10 = “the most intense pain imaginable”) were also assessed. The clinician reviewed the results with the study team to ensure that no exclusion criteria were met. All cLBP participants also completed a set of questionnaires and psychological assessments, including the Brief Pain Inventory (BPI; for pain severity), the Oswestry Disability Index (ODI; for self-reported function and disability), the Hospital Anxiety Depression Scale (HADS total score; for emotional function)[35], painDETECT (for neuropathic pain elements), the Screener and Opioid Assessment for Patients with Pain, Revised (SOAPP-R; for estimating the risk of opioid misuse)[5], and the Mini International Neuropsychiatric Interview (MINI; to rule out current substance abuse disorder). Additionally, each participant completed the “minimum dataset” suggested by the NIH Back Pain Consortium (BACPAC)[34], as part of the NIH HEAL Initiative, consisting of two questionnaires (Baseline Demographics and Outcomes Assessment) concerning demographics, pain level, pain interference, sleep disturbance, physical function, and emotional function. Healthy controls did not complete assessments on pain but did complete surveys on demographics.

Radiosynthesis of [¹¹C]Martinostat

[¹¹C]Martinostat is a hydroxamic acid-based HDAC inhibitor that contains an adamantyl group and radiolabeled with 11C. Briefly, the synthesis of [¹¹C]Martinostat was initiated by reductive amination, followed by conversion into a hydroxamic acid in the presence of hydroxylamine and sodium hydroxide in accordance with cGMP guidelines as previously described [48,50].

Scan procedures: PET/MR imaging

All participants had no MRI or PET contraindications to undergo brain imaging safely. PET/MRI scans were acquired on a 3T Siemens TIM Trio with a BrainPET insert (Siemens, Erlangen, Germany). A PET-compatible circularly polarized transmit coil, and an eight-channel receive array coil were used for MRI data acquisition. An intravenous catheter was placed in the antecubital vein of the other arm, and a licensed nuclear medicine technologist performed a manual bolus injection (9.23 ± 4.37 mCi) of [¹¹C]Martinostat into the intravenous catheter. Dynamic PET scan (90 min) was started concomitantly with radiotracer injection. Our previous study with healthy individuals reported that image-based outcome measures, i.e., standard uptake values (SUV) and SUV normalized to the white matter (SUVR), are well correlated with the volume of distribution values (V_T) derived from kinetic modeling with low inter-subject variability [50]. It is important to confirm whether a simplified outcome measure is valid in cLBP patients. Therefore, in a subset of cLBP patients (n=11), an arterial line was placed in the radial artery of one arm, and blood samples were collected from the arterial line by an experienced nurse practitioner. Participants were instructed to remain still during the scan. Due to a long scan duration, we registered the MR-based Attenuation Correction (AC) maps to preliminary PET recons spanning 60–90 mins after radiotracer administration and carefully performed motion corrections for the reconstructed PET images. Images were reconstructed using the three-dimensional ordinary Poisson ordered-subset expectation maximization (3D OP-OSEM) algorithm with detector efficiency, decay, dead time, attenuation, and scatter corrections applied. An atlas-based MR AC method using Statistical Parametric Mapping (SPM) was applied[24] to derive a linear attenuation coefficient map. The final PET images with the units of SUV (mean radioactivity per injected dose per weight) and radioactivity concentrations (becquerels per milliliter) were reconstructed into 153 slices with 256 × 256 pixels 1.25-mm isotropic voxel size. A high-resolution structural scan using a multi-echo magnetization-prepared rapid acquisition gradient-echo sequence (MEMPRAGE) was acquired during the PET scan using the following parameters: repetition time (TR) = 2530 ms; echo time 1 (TE1) = 1.64 ms; TE2 = 3.49 ms; TE3 = 5.35 ms; TE4 = 7.21 ms; inversion time (TI) = 1200 ms; flip angle = 7°; and isotropic resolution = 1 mm. List mode PET data were stored for 90 min and binned into 30 frames of progressively longer duration (10 s × 9, 20 s × 3, 30 s × 3, 60 s × 1, 120 s × 1, 180 s × 1, 300 s × 8, and 600 s × 4).

Arterial Blood Sampling and Analysis

The arterial blood samples were acquired during the dynamic PET scans at ~10 s intervals beginning at the start of the scan for 3 min (~2 ml each), which was followed by additional samples at 5, 10, 20, 30, 50, 70, and 90 min (~6 ml each) after radiotracer injection for plasma and radiometabolite analyses. In brief, the collected blood samples were centrifuged. The resulting 200 μl (for the samples obtained within the first 3 min) or 600 μl (for other samples) of plasma samples were placed in a gamma counter cross-calibrated with the PET scanner. For the radiometabolite high-performance liquid chromatography (HPLC) analysis, the samples obtained at (5, 10, 20, 30, 50, 70, and 90 min) were centrifuged at 4,000 rpm at 4 °C, for 4 mins. The centrifuged plasma (1 ml) was extracted from the sample tube and added to another tube containing 1 mL of acetonitrile. The resulting tube was briefly vortexed to mix the contents and then it was centrifuged for a second time with the same parameters. The liquid contents (~1 ml) of that tube was extracted and mixed in a scintillation vial containing 4 mL of deionized water with 0.1% Formic Acid. The mixture was gently shaken, and the total 5mL solution was injected on the HPLC (Agilent 1260 HPLC with Fraction Collector, Agilent Technologies, CA, USA). Total plasma radioactivity was linearly interpolated and corrected for the fractions of radiometabolites over time. The metabolite-corrected arterial plasma input curves were used for kinetic modeling (see below).

MR image processing and analysis

Each subject’s high-resolution T1-weighted images were analyzed with automatic segmentation and parcellation using FreeSurfer (version 6.0; http://surfer.nmr.mgh.harvard.edu/). From the segmentation outcomes, subject-specific binary masks for a set of region-of-interests (ROIs) containing a total of 27 brain regions were obtained for the region-based analysis. The set of ROIs includes whole brain, white matter, deep white matter, pons, cerebellum, brain stem, thalamus, caudate, putamen, pallidum, hippocampus, amygdala, nucleus accumbens, periaqueductal gray (PAG), orbitofrontal cortex (OFC), anterior cingulate cortex (ACC), posterior cingulate cortex (PCC), insula, precentral gyrus (primary motor cortex), postcentral gyrus (somatosensory cortex), paracentral lobule, rostral-middle-frontal gyrus (dorsolateral prefrontal cortex; dlFPC), precuneus, inferior-frontal gyrus, supramarginal gyrus (somatosensory association cortex), middle-temporal gyrus, superior-parietal gyrus. The selection of this set of ROIs was driven by two objectives: 1) whole brain, white matter, deep white matter, pons and cerebellum were chosen as potential normalization references for image analysis; 2) other brain regions were selected because they were implicated in the processing of pain, as indicated by extensive neuroimaging studies (broadly considered as part of the pain matrix) [47]. Because certain brain structures scale with general head size, we employed estimated total intracranial volume (eTIV), scaling the volume of brain structures based on the individual’s intracranial volume [7]. We obtained the eTIV for each subject to investigate whether a group difference of regional [¹¹C]Martinostat uptake was driven by ROI volumes relative to individual subjects.

PET image processing and analysis

For the full cohort, PET data in units of SUV spanning 60–90 mins after radiotracer administration (SUV_60–90min) were used, while the full 0–90 mins data were used for subjects who have arterial blood data available. Preprocessing of the PET data included the following steps: Reconstructed PET images were first spatially smoothed by a Gaussian kernel of 6 mm full width at half maximum (FWHM). For the motion correction, we binned 30 minutes into 6 frames of 5 minutes and aligned them to a mean volume image using rigid body linear registration (6 d.o.f) using FSL MCFLIRT function (FMRIB (Oxford Centre for Functional MRI of the Brain) Software Library, Oxford, UK; https://fsl.fmrib.ox.ac.uk/fsl/) [26]. Using this linear transformation, unsmoothed PET data were aligned and averaged into a single frame. The frame-averaged PET image for each subject was then used to derive a linear transform to the T1-weighted structural scan using the boundary-based registration (bbregister) from FreeSurfer (version 6.0; http://surfer.nmr.mgh.harvard.edu)[13,19]. To account for partial volume effects in the PET images, a region-based, voxel-wise (RBV) method was applied with a geometric transfer matrix (GTM) derived from the segmentation of the T1-weighted image for each subject with tools implemented in PETSurfer. Lastly, the MEMPRAGE was registered to Montreal Neurological Institute (MNI152) atlas using linear (FLIRT (FMRIB’s linear image registration tool) and nonlinear (FNIRT (FMRIB’s nonlinear image registration tool) algorithms implemented in FSL [42]. Combining these transformation matrices, PET images were normalized to the MNI152 space and spatially smoothed by a Gaussian kernel of 8 mm (FWHM) for further voxel-based analyses. For region-based analyses, regional SUV or radioactivity values were extracted from the final unsmoothed, PVC-corrected PET data using the set of subject-specific ROIs in the native PET/MRI space. To facilitate intersubject comparison of regional HDAC distribution, we normalized regional SUV_60–90min to individual subjects’ pons SUV_60–90min as normalized SUV_60–90min ratio (SUVR_{60–90min, Pons}), similar to our previous study in healthy volunteers, which showed less variability in test/retest scans [50].

Kinetic modeling analysis

Full kinetic analyses were performed to estimate the V_T in the subset of subjects who underwent blood draws from the arterial line. Our previous study with healthy individuals reported that these image-based outcome measures are well correlated with the V_T with low inter-subject variability [50]. However, the validity of SUV_60–90min and SUVR_{60–90min,Pons} have not been tested in cLBP cohort. Kinetic modeling was conducted using PMOD 3.9 (PMOD Technologies Ltd, Zurich, Switzerland), with 27 ROIs to derive V_T. Time activity curves (TACs) for each subject were extracted using subject-specific ROIs derived from FreeSurfer (as described above). In our previous work, a two-tissue compartment model (2TCM) was shown as the best model to estimate V_T (ml/cm³) for [¹¹C]Martinostat [50]. In this work, we used the 2TCM with the metabolite-corrected arterial plasma as an input function, fitting the blood volume fraction and a delay term for the plasma input function. Parametric images were generated with the pixel-wise modeling tool (PXMOD) of PMOD 3.9. The radioactivity concentration images were registered to the MNI152 atlas space, spatially smoothed (8 mm Gaussian kernel), and analyzed with the metabolite-corrected arterial plasma using the voxel-based graphical analysis (Logan invasive; t* = 40 min), then the resulting V_T parametric maps (n=5) were averaged.

Statistical analysis

For region-based analysis, statistical tests were performed using R software (Version 4.2.3) and GraphPad Prism (Prism9, GraphPad Software Inc., San Diego, California, USA). The statistical analysis was performed in stepwise levels: 1) We compared SUV_{60–90 min} of potential pseudo-reference regions (whole brain, white matter, deep white matter, pons, and cerebellum) between cLBP patients and controls, respectively, using a one-way analysis of variance (ANOVA) test. The aim is to identify a brain region with no significant group differences. 2) We assessed the association of SUV_{60–90 min} or SUVR_{60–90min, Pons} from the target ROIs with demographic variables (age, sex, BMI, and injected dose of the radiotracer) across the entire cohort using a non-parametric multivariate analysis of variance (MANOVA) with 5000 permutations, identifying demographic variables linked to PET measures (SUV_{60–90 min} or SUVR_{60–90min, Pons}). 3) We then examined the main effects of group (cLBP vs. controls) on SUV_{60–90 min} or SUVR_{60–90min, Pons} in the target ROIs through MANOVA with 5000 permutations. Any demographic variables showing significant associations with SUV_60–90min or SUVR_{60–90min, Pons} were factored in as covariates of no-interest in both MANOVA and subsequent post hoc analysis. 4) Should the MANOVA indicate significant group effect on PET measures (p < 0.05), we proceeded with post hoc multivariate analyses to delve deeper into these findings. 5) In cLBP patients, the relationship of SUV_{60–90 min} or SUVR_{60–90min, Pons} in the target ROIs with covariates of interest (pain, HDAS depression, and anxiety, ODI, and SOAPP-R) was tested using non-parametric MANOVA with 5000 permutations. 6) As an exploratory analysis, we conducted a multivariate post hoc analysis of the correlation between PET measures (SUV_{60–90 min} or SUVR_{60–90min, Pons}) in selected core ROIs (PAG, NAc, somatosensory, dlPFC) and pain intensity. 7) Lastly, a Pearson correlation analysis was performed to compare V_T and image-derived PET measures (SUV_{60–90 min} and SUVR_{60–90min, Pons}) for all ROIs to assess validity of mage-based outcome measurements as a appropriate surrogate for those estimated with the full kinetic modeling.

A voxel-wise comparison for [¹¹C]Martinostat uptake (SUV_60–90min and SUVR_{60–90min, Pons}) between cLBP patients and controls was performed using FSL’s FEAT (FMRIB software library) with an unpaired t-test (a significance threshold of Z > 2.3 with the cluster-based multiple comparison correction p_cluster < 0.05). Whole-brain voxel-wise analyses correlating SUV_{60–90 min} and SUVR_{60–90min, Pons} values with the pain intensity were performed using FSL’s FEAT (Z > 2.3, p_cluster < 0.05), with BMI added to the model as regressors of no interest.

Results

Demographics and clinical characteristics

Demographic information and clinical characteristics of participants are provided in Table 1. There were no significant differences in age, sex, or body mass index (BMI) between groups. However, the [¹¹C]Martinostat injected dose differed significantly between cLBP and controls (p < 0.005). To address the potential confounding effect of the injected dose on group comparison, we performed an additional analyses. These consisted of comparing cLBP patients who received a lower dose (~5 mCi) with those injected with a higher dose (~15 mCi) (Supplementary Materials), and comparing a subset of matched cohorts (8 cLBP vs. 5 controls) with comparable injected doses (Supplementary Materials). With these additional analyses, we reassured that the variance in injected dose has minimal impact on our overall results.

Table 1.

Demographic characteristics, administered radiotracer dose, and cognitive metrics of study participants (mean ± SD).

Demographic or survey metric	cLBP (n=23)	Control (n=11)	p-value
Age (year)	47.96 ± 15.17	46.91 ± 15.26	0.8390
Sex (M/F)	11/12	7/4	0.6193
Body Mass Index	26.68 ± 4.91	26.73 ± 5.92	0.9798
Injected dose (mCi)	10.72 ± 4.32	6.10 ± 2.47	0.0026*
Pain Intensity	4.48 ± 1.73	n/a
HADS - Depression (out of 21)	3.87 ± 3.76	n/a
HADS - Anxiety (out of 21)	5.78 ± 4.19	n/a
ODI (out of 49)	11.48 ± 6.1	n/a
SOAPP-R (out of 96)	10.78 ± 9.19	n/a

Abbreviation: cLBP = chronic lower back pain patients; SD = standard deviation; HADS = Hospital Anxiety Depression Scale; ODI = Oswestry Disability Index; SOAPP-R= Screener and Opioid Assessment for Patients with Pain, Revised.

^*p<0.005.

[¹¹C]Martinostat uptake between cLBP patients and matched controls

PET images of [¹¹C]Martinostat are shown in Figure 1 to illustrate the spatial distributions of HDAC availability of cLBP patients and controls. Analysis revealed no significant correlation of either SUV_60–90min or SUVR_{60–90min, Pons} with age (F = 0.3933, p = 0.5557; F = 0.5739, p = 0.4965), sex (F = 1.4233; p = 0.2322; F = 0.3173, p = 0.7159), or injected dose (F = 1.0458, p = 0.3167; F = 0.6683, p = 0.4471). However, a significant association was found between BMI and both PET measures (SUV_60–90min, F = 4.9892, p = 0.0288; SUVR_{60–90min, Pons}, F = 4.4578, p = 0.0260). Therefore, we included BMI as a covariate alongside group effects in subsequent analyses. MANOVA demonstrated a significantly lower SUV_60–90min (F = 5.5079; p = 0.0252) in cLBP patients compared to healthy control. A post hoc multivariate analysis indicated the regions showing significantly lower SUV_60–90min included the thalamus, caudate, putamen, pallidum, hippocampus, amygdala, OFC, ACC, PCC, primary motor cortex, somatosensory cortex, dlPFC, precuneus, inferior frontal gyrus, somatosensory association cortex, middle temporal, superior parietal gyri (Figure S2(a)). Similarly, we found a significantly lower SUVR_{60–90min, Pons}, in cLBP patients relatie to controls (F = 3.9708, p = 0.0368), with affected regions including ACC, PCC, primary motor cortex, somatosensory cortex, dlPFC, precuneus, inferior frontal gyrus, somatosensory association cortex, middle temporal, superior parietal gyri (Figure 2). Notably, estimated total intracranial volume (eTIV) and eTIV-normalized regional brain volumes did not differ between groups (Figure S3), which confirms that the observed [11C]Martinostat uptake differences were not driven by brain volumes (F = 0.5176, p = 0.5131).

Figure 1.

Group average maps of [¹¹C]Martinostat uptake in SUV. SUV at 60–90 min post radiotracer injection (SUV_60–90min) for cLBP patients (n=23) and age- and sex-matched healthy controls (n=11) are shown in the MNI152 atlas space. Standard uptake value, SUV; chronic low back pain, cLBP.

Figure 2.

Regional SUVR_{60–90min,Pons} comparison between cLBP patients and matched controls (n=11). SUV at 60–90 min post radiotracer injection normalized to the pons SUV (SUVR_{60–90min,Pons}) was plotted from selected brain regions. The main group effects (cLBP vs. controls) in SUVR_{60–90min, Pons} in the target region of interests (n=22) were tested by using non-parametric multivariate analysis of variance (MANOVA) with 5000 permutations with the BMI as a covariates. If MANOVA indicated a significant group effect (p < 0.05), we performed addtional post hoc multivariate analyses. The asterisk indicates statistical significance, *p < 0.05, **p < 0.01. Standard uptake value, SUV; chronic low back pain, cLBP; analysis of variance, ANOVA.

In voxel-wise group comparison of SUV_60–90min (Z > 2.3, p_cluster < 0.05), cLBP patients showed significantly lower regional [¹¹C]Martinostat uptake in multiple clusters within the somatosensory cortex, temporal, frontal cortex, cerebellum, and amygdala when compared to controls (Figure S2(b)). Nevertheless, no significant differences were observed in SUVR_{60–90min, Pons} between cLBP patients and controls (Z > 2.3, p_cluster < 0.05) when cluster-based multiple comparison corrections were applied. Moreover, no regions showed increased uptake (SUV_60–90min and SUVR_{60–90min, Pons}) in cLBP patients relative to controls.

Kinetic analysis

Due to equipment malfunction, blood data from six out of the eleven subjects who underwent arterial blood sampling were unable to be properly For the remaining subjects, radioactivity in the arterial plasma peaked within 1 min after radiotracer injection and decreased rapidly thereafter (Figure S4(a)). The averaged parent fraction ratios from cLBP patients (n=5) were about 0.5 at 30 min after radiotracer administration and eventually reached 0.2 by the end of the scan (Figure S4(b)). Consistent with our previous study in healthy individuals, the two-tissue compartment model (2TCM) best described dynamic [¹¹C]Martinostat PET data obtained from cLBP patients [50]. Figure S4(c) shows representative time activity curves (TACs) and compartment model fitting results of the putamen from a representative cLBP patient. The 2TCM-estimated volume of distribution (V_T) values ranged from 4.7 to 38.6 ml/cm³ (mean ± SD = 16.9 ± 7.1 across analyzed ROIs) (Table 2). Logan graphical analysis estimated regional V_T values, using metabolite-corrected arterial blood as an input function (t* of 40 mins), yielded regional V_T values that were positively correlated with those derived with 2TCM (Pearson r = 0.81, two-tailed p < 0.0001) (Table 2 and Figure 3 (a)). The V_T ratio of Logan graphical analysis to 2TCM was 1.01 ± 0.17 (mean ± SD across all ROIs), indicating comparable estimates by the two methods. As illustrated in Figure S5, whole brain V_T parametric map was calculated from the subset of the cLBP patients (n=5). We found a significant positive correlation between V_T values and SUV_60–90min (Pearson r = 0.72, two-tailed p < 0.0001; Figure 3(b)) and with SUVR_{60–90min, Pons} (Pearson r = 0.72, two-tailed p < 0.0001; Figure 3(c)) in the subset of the cLBP cohort, confirming that SUV_60–90min and SUVR_{60–90min, Pons} are appropriate image-based substitutes for V_T in cLBP patients. Similar to the previous study in healthy controls, SUVR_{60–90min, Pons} demonstrated lower inter-subject variability, with a coefficient of variation (CV) ranging from 6.02 to 12.82%, compared to blood data–derived V_T values (CV of 34.23 to 65.38%) across brain regions.

Table 2.

The volume of distribution (V_T) estimated by using the 2TCM and Logan invasive method in the brain of cLBP patients (n=5).

Brain regions	2TCM		Logan invasive
	VT [mL/cm3]	SD	VT [mL/cm3]	SD
Whole brain	15.20	6.15	16.88	8.73
White matter	14.28	5.32	16.05	7.61
Deep white matter	9.03	3.09	10.38	4.60
Pons	14.32	6.04	14.05	6.20
Cerebellum	22.63	11.21	22.33	13.12
Brain stem	13.50	5.37	13.28	5.33
Thalamus	17.08	7.68	19.20	11.64
Caudate	18.98	12.41	17.92	12.40
Putamen	22.27	9.56	23.49	12.41
Pallidum	17.94	7.82	17.94	9.05
Hippocampus	15.14	6.00	14.52	6.59
Amygdala	16.72	6.72	12.21	7.74
Nucleus Accumbens	21.20	10.14	20.57	11.56
Periaqueductal gray	15.33	6.69	12.38	3.34
Orbitofrontal cortex	19.73	11.37	21.52	14.72
Anterior Cingulate Cortex	18.63	7.69	20.25	10.29
Posterior Cingulate Cortex	17.76	6.18	18.67	8.48
Insula	18.85	7.78	19.37	8.86
Precentral	15.66	5.79	16.99	8.03
Postcentral	15.51	5.83	16.96	8.20
Paracentral	16.87	7.52	17.14	8.06
Rostralmiddlefrontal	17.37	7.10	19.16	9.83
Precuneus	17.44	6.25	19.17	8.80
Inferiorfrontal gyrus	16.91	6.19	17.38	7.61
Suparmarginal	17.51	6.48	18.93	9.07
Middletemporal	18.55	7.89	19.52	10.58
Superiorparietal	15.13	5.32	16.99	8.00
Averaged across ROIs	17.02	7.31	17.53	8.90

Abbreviation: 2TCM = Two tissue compartment model; cLBP = chronic lower back pain patients; SD = standard deviation.

Figure 3.

Scatter plots of (a) regional V_T values estimated using the 2TCM and Logan plot (t*=40 mins) and (b) regional V_T values from the 2TCM and SUV_60–90min or (c) SUVR_{60–90min, Pons} in the subset of cLBP patients (n=5). Correlation analyses are conducted using a Pearson correlation analysis. Gray lines indicate standard deviation. Volume of distribution, V_T; two tissue compartment model, 2TCM; standard uptake value at 60 to 90 minutes after a radiotracer injection, SUV_60–90min; pons brain normalized standard uptake value, SUVR_{60–90min,Pons}; chronic low back pain, cLBP.

Post hoc analyses

Post hoc analyses were performed to assess the relationship between regional [¹¹C]Martinostat uptake from a priori ROIs and patient-reported outcome measures. We included the BMI as a covariate due to its established correlation with SUV_60–90min and SUVR_{60–90min, Pons}. Neither SUV_60–90min nor SUVR_{60–90min, Pons} showed a significant association with any self-report outcomes. However, considering potential region-specific associations, we conducted a focus examination of PET measures from selected brain regions (NAc, PAG, somatosensory cortex, dlPFC) and their relationship with pain intensity with a post hoc multivariate analysis. Interestingly, a trend level of higher reported pain intensity corresponded with reduced SUVR_{60–90min, Pons}, particularly within the PAG (p = 0.0423) (Figure S6).

Discussion

Chronic pain is a complex disorder that affects both physical and mental functioning that could compromise quality of life. Current therapeutic strategies remain inadequate. Building on previous preclinical studies that implicate histone modifications are involved in animal pain models [16,49], we employed [¹¹C]Martinostat PET to measure and compare in vivo HDAC expression levels in the brain of cLBP patients with age- and sex-matched healthy controls. In addition, we investigated the association between various patient-reported outcomes and HDAC availability in cLBP. Our findings indicated a pronounced decrease in HDAC availability in the cLBP patients compared to controls, particularly within the ACC, PCC, primary motor cortex, somatosensory cortex, dlFPC, precuneus, inferior frontal gyrus, somatosensory association cortex, middle temporal and superior parietal gyri. These results supported the hypothesis that HDAC availability was altered in cLBP, consistent with preclinical animal studies [2,9,16,49], providing evidence of HDAC alterations in vivo in the brains of cLBP patients.

We found that [¹¹C]Martinostat SUV_60–90min are significantly lower in the brain of cLBP patients than controls globally. A majority of the target ROIs, except for 5 brain regions (brain stem, NAc, PAG, insula, paracentral lobule), demonstrated significant group differences. When assessing SUVR_{60–90min, Pons}, which is expected to have lower inter-subject variability than SUV_60–90min and V_T, we could identify the affected brain regions corresponding to the pain matrix [47]. Specifically, the primary motor cortex, somatosensory cortex, dlPFC, precuneus, inferior frontal gyrus, somatosensory association cortex, middle temporal, and superior parietal gyri in cLBP patients exhibited lower SUVR_{60–90min, Pons} compared to controls. The thalamus, somatosensory cortex, superior parietal gyrus, somatosensory association cortex, cingulate cortices, and OFC are traditionally associated with the sensory component of pain [30,40,47]. The basal ganglia have also been implicated in pain processing based on preclinical and clinical evidence [3]. To ensure our results were valid, we explored potential confounding factors, including anatomical intracranial brain volumes, injected dose, age, and sex between cLBP and controls and confirmed no significant group differences. In addition, we found no significant correlation between PET measures (SUV_60–90min and SUVR_{60–90min, Pons}) and any patient-reported outcomes, potentially due to our relatively small sample size (n=23 cLBP). This is supported by our exploratory post hoc analysis which showed a trend level negative correlation between pain intensity and SUVR_{60–90min, Pons} in the PAG (p = 0.0423). The PAG is a brain region with a high density of mu-opioid receptors and is considered a key component in the transmission and regulation of pain within the descending pain modulatory system and the processing of pain signals at the supraspinal level. While the MANOVA did not show a statistically significant association between pain and SUVR_{60–90min, Pons} across the target ROIs (F = 0.8607, p = 0.3743), this trend in the PAG and pain intensity warrants further validation in a larger cohort.

Previous research has demonstrated a downward trend in histone modifications (H3K4me3, H3K27Ac, and H3K4me1) within the PAG, lateral hypothalamus, and NAc after a 5-week period post-sciatic nerve injury in mice [6]. This trend, specifically lower H3K4me1 and H3K4me3 levels in the PAG, correlates with individual variations in pain intensity [6], hinting at the enduring nature of epigenetic modifications in limbic brain structures in response to nerve injury and their potential role in sustained pain modulation. Inflammation induced by a complete Freund’s adjuvant (CFA) injection has been shown to the marginally reduce class I HDACs expression in the lumbar dorsal spinal cord [2]. Concordantly, other animal models have demonstrated differences in HDAC expression in response to nerve injury; for example, increased HDAC1 expression in the nuclear fraction of the ipsilateral lumbar spinal dorsal horn in a rat model of spinal nerve ligation (SNL) model [9], and a reduction of HDAC2 in the spinal dorsal horn 7 days after a spared nerve injury [41]. These findings supported the therapeutic exploration of HDAC inhibitors [1,16,17,36], which have demonstrated analgesic efficacy in animal pain models [2,9,14]. Collectively, these preclinical studies showed that efforts to examine changes in HDAC levels (or histone markers) primarily focus on the spinal cords while the supraspinal regions remain largely unexplored; furthermore, no study has previously investigated HDAC alternations in the brain of cLBP patients. Our study expands the understanding of HDAC involvement in chronic pain and reinforces the potential for HDAC modulation as a therapeutic target in humans.

Previous studies on histone modifications related to pain have yielded inconsistent results, although the trend has generally been towards elevated HDAC levels and reduced histone acetylation in animal models. This seems to contradict our findings in cLBP patients. However. direct comparisons are confounded not only by species differences but also by the chronicity of pain in our human cohort, as opposed to the acute or sub-acute pain often induced in animal studies. Furthermore, our study focused on cerebral HDAC, in contrast to the spinal-centric emphasis of most animal studies. It’s worth noting that our findings of altered HDAC availability in patients with various neurological and psychiatric disorders do not necessarily aligned with the directionality of HDAC levels suggested by preclinical research [37,45]. Beyond global alterations in HDAC expression and histone acetylation, targeted analyses focusing on the acetylation patterns of gene-specific promoters of pain-related receptors, ion channels, and cytokines, could provide additional insights on the role of histone modification in pain [10,28,29]. For example, neuropathic pain symptoms have been correlated with decreased expression of μ-opioid receptors and Nav1.8 sodium channels in the dorsal root ganglion [46], coinciding with enhanced histone acetylation in neuron-restrictive silencer factor promoter regions [33]. In addition, enhanced acetylation in the promoters of the genes like chemokine CC motif receptor 2 (CXCR2) was observed under pain conditions [43]. Regrettably, our current approach using [¹¹C]Martinostat PET does not provide mechanistic evidence supporting specific modulation of gene expression. Nevertheless the observed reduction in HDAC availability within pain-associated brain regions of cLBP patients present new empirical evidence supporting the involvement of HDACs in chronic pain in the human brain. Future in vitro experiments elucidating the functional consequences of HDAC modulations on specific gene promoters related to pain-modulated receptors, ion channels, and cytokines may better illustrate the involvement of histone acetylation in different kinds of pain conditions.

We have validated the use of SUV_60–90min and SUV_{60–90min, Pons} as reliable image-based outcome measures to compare HDAC density between groups. Despite the challenge of variable injected doses, our analyses confirmed this factor has minimal impact on our overall results. However, we acknowledge that due to a relatively small sample size, sex differences in HDAC expression were not explored despite there are known differences in HDAC availability and the prevalence of chronic pain between males and females.

In conclusion, our study reported significant alteration in HDAC availability, indicated by [¹¹C]Martinostat PET, in several brain regions associated with pain processing in cLBP patients, underscoring the potential of HDAC modulation as a target for chronic pain management. Future research with larger cohorts or in other pain conditions are essential to further elucidate the therapeutic implications targeting HDACs.