Genetic and epigenetic regulation of Catechol-O-methyltransferase in relation to inflammation in chronic fatigue syndrome and Fibromyalgia

Patients with a clear diagnosis of CFS and FM according to the Centre for Disease Control and Prevention Criteria for CFS, [10] and to the American College of Rheumatology criteria for FM were enrolled [51]. Healthy controls were recruited during the same period among friends or acquaintances of patients or of people involved in the study and matched for sex, age, and body-mass index. Because of the higher prevalence of CFS/FM in women, [39] and because important sex differences exist for pain and immune function, [40] we included only women in the present study. Matching was performed via frequency-matching approach, and checking for frequency of the variables of interest (age and BMI) every 5 patients enrolled in the study. The ratio between patients and controls was 1:1. Participants were excluded if they reported other neurological, psychiatric, cardiopulmonary, or oncological comorbidities. In addition, as physical activity is known to influence both the autonomic nervous system and the immune system, we asked all participants to avoid engaging themselves in physical activity the day before the first assessment or during the study period. As no previous study explored COMT DNA methylation in people with CFS/FM and a repeated measure design, we used one study by Kyoung-Sae et al. [29] to calculate the effect size. Authors explored DNA methylation of COMT in healthy controls and people with clinical depression using our same methodology (PCR-pyrosequencing in the promoter region of MB-COMT). They found a significant between-group mean difference of 1.36% +/- 0.9 (F-test: 28.024, partial n ²: 0.167). Based on that information, G*Power 3.1 calculated an effect size of f = 0.45. We then calculated the sample size needed to detect an effect size of f = 0.45 at the significance level of α = 5% and a power of 80%, using a repeated-measure design with 2 groups and 2 measurements [42]. The total sample size needed was 52 participants. This is also in line with a number of previously published studies measuring variability and clinical relevance of DNA methylation using a candidate gene approach. As summarised by Jones et al [16], most studies show that a mean difference of 1–2% in DNA methylation is sufficient for inducing a significant downstream effect on gene expression [16].

Participants visited the Department of Internal Medicine and Endocrinology of the University Hospital Brussels twice within four days, between 9 and 11 am. Informed consent form was signed before the assessment started. During the first assessment, participants were interviewed. Information regarding general health, comorbidities, and drug intake was noted. Next, they filled in questionnaires assessing physical activity, fatigue, pain, sleep, cognitive and psychological symptoms. Then, a nurse was in charge of blood withdrawal. Samples were processed within one hour from sampling and stored at -80 °C. Afterwards, patients underwent a neurophysiological assessment, as described below. Patients visited our department again after three days, and underwent a second assessment, identical to the first one – questionnaires, neurophysiology, and blood withdrawal. This was done in order to measure reliability of the assessments, as well as to include variability of the biological measures in the statistical models. Assessments prone to assessor-bias were performed blinded to group allocation.

Clinical characteristics of participants were investigated through the International Physical Activity Questionnaire to measure physical activity, and CFS Symptom List sub-domains to measure the different symptoms associated with CFS: pain, fatigue, sleep disturbances, cognitive problems, and immune aspects (flu-like symptoms). Questionnaires explore clinically relevant aspects for patients with chronic pain and fatigue [33, 48]. They are validated for Dutch-speaking people and show excellent psychometric properties [33, 48]. Neurophysiological assessment included pain sensitivity, endogenous pain facilitation and descending modulation. Neurophysiological mechanisms were assessed by an experimental protocol (lasting approximately 40 min). First, we assessed pain sensitivity by measuring pain threshold to cold, heat and mechanical stimuli. Heat and cold pain thresholds were assessed at three body parts in random order (Neck [on the trapezius superior muscle belly], Hand [on the first inter-digital space of the non-dominant hand], and Leg [on the tibialis anterior muscle belly]) using the TSA-II Neurosensory Analyzer (Medoc, Ltd. Israel). TSA-II is a computerized device designed for measuring sensory thresholds by delivering thermal stimuli to the skin through a 30 × 30 mm thermode. To assess mechanical thresholds, pressure pain thresholds (PPTs) were measured at the hand site only using an analog pressure algometer (Force Dial models FDK 10 Push Pull Force Gate, Wagner Instruments, Greenwich, CT, USA). Three consecutive stimuli were delivered on each body part; participants were asked to stop the stimulation as soon as it was perceived as painful. The thresholds, expressed in °C for the thermal stimulations, and in kg/cm² for mechanical stimulation, are determined as the mean of the last two. They also had to rate the intensity of each stimulus by drawing a vertical line on a visual analogue scale. VAS was represented as a 10-cm horizontal segment going from 0, No pain, to 10, Worst pain imaginable. These procedures were found to be reliable for measuring hyperalgesia in patients with chronic pain [11]. Endogenous pain facilitation was assessed through a temporal summation protocol, which started 2 min after the final PPT measurement [44, 53]. The measure was taken at the hand site only. Participants were given 10 pulses to the previously determined mean PPT intensity. Pressure was maintained for one second and then released, with 1-second inter-stimulus interval. After the first and tenth pulse, the participant had to rate her pain on the visual analog scale. Temporal summation was calculated as the difference in the pain rate between the 10th and the 1st pulse. Finally, descending modulation was assessed through a conditioned pain modulation paradigm [44, 53]. Participants dipped the dominant hand into mildly-moderate painful (rated between 4/10 and 7/10 on a visual analogic scale) hot water (45 to 46 °C). While the hand was in the water, pain sensitivity was assessed again on the non-dominant hand as described above. Tonic painful stimulation (the hot bath) is the conditioning stimulus and is thought to engage descending inhibitory mechanisms. Descending inhibition is reflected by an increase in pain thresholds on the non-dominant hand while the dominant hand is placed in hot water. Descending modulation is thus measured as the difference in pain threshold (in °C) between before and during hot water immersion. The Polystat Isotemp 4100 C (Fisher Scientific, Aalst, Belgium) was used for accurate controlling of water temperature. The procedure was found to be reliable and valid in healthy people and chronic pain patients, including those with fibromyalgia [3]. Pre-processing of data. Pain threshold measurements for cold and heat stimulation were highly correlated to one another. However, they showed clear ceiling effect. In order to extract the most information from our assessment, we calculated the predicted values imputing the three cold thresholds and the three heat thresholds in two separate regression models. The predicted values were then considered a measure for cold sensitivity and heat sensitivity, respectively (see supplementary material S1 for details).

Blood was collected in 4 tubes – 15 ml in 3 EDTA tubes for blood collection and 5 ml in one serum tube – centrifuged (3000 rpm for 10 minutes at 4°C), and stored at -80°C within 1 hour from sampling. DNA was extracted from blood using QIAamp DNA Blood Mini Kit (Qiagen, Hilde, Germany). DNA methylation and polymorphism analyses. 200 ng of DNA was incubated with sodium bisulfite using the EZ DNA Methylation-Gold kit (Zymo Research, the Netherlands). Bisulfite-converted DNA sequences of interest were then amplified using polymerase chain reaction (PCR). Genetic and epigenetic analyses were carried out by pyrosequencing – using a Q24 Pyrosequencer device (Qiagen, Hilde, Germany). Pyromark Q24 Analyses software (Qiagen, Hilde, Germany) then measured the polymorphisms and the average methylation in each C-G dinucleotide (CpG) and returned a percentage (from 0–100% methylation) as a result. DNA methylation was measured in MB-COMT promoter region (both before and after the transcription start site in Exon I), and S-COMT promoter region (Exon III, overlapping with translational start site for both MB-COMT and S-COMT – this region was shown to hold marked regulatory function on S-COMT activity according to the most recent and comprehensive interrogation of the COMT gene) [46]. Three polymorphisms were also assessed via pyrosequencing analyses: rs4633, located in the S-COMT promoter, as well as rs4818 and rs4680, located in Exon IV. More details can be found in the Supplementary Materials (S2). Haplotype estimation. Haplotype analysis was performed using Haploview v4.2 [1]. The genetic polymorphisms rs4818, rs4680 and rs4633 passed the default cut-off of the data quality metrics (Hardy-Weinberg equilibrium p-value > 0.001, percentage of non-missing genotypes > 75%, minor allele frequency > 0.001) and were thus all included for haplotype analysis. Markers of linkage disequilibrium (D’, r² and LOD) were calculated automatically by Haploview v4.2, which derives these markers from maximum-likelihood values of the four gamete frequencies estimated via a two marker Expectation – Maximization (EM) algorithm [1]. The genetic polymorphisms were manually grouped into one block after which haplotype association analysis was performed. Differences in genotype or haplotype frequency distribution were tested using the Chi Square test. See Supplementary Materials (S2) for details. Inflammatory cytokines. Enzyme-Linked Immunosorbent Assay (ELISA) human assay kit (Thermofisher, Invitrogen Inc., USA) was used to measure TNF-α, IL-6, IFN-γ, and TGF-β in serum. Invitrogen products ensure high reliability (average intra-assay coefficient of variation: 4.5%; inter-assay variability: 5.7%; no detectable cross-reactivity). Details on the validation procedure can be found in the Supplementary Materials (S2). Of note, a number of samples could not be detected with ELISA, possibly due to degradation of the samples. IL-6 was detected in 99, INF-γ in 94, and TGF-β in 102 out of 108 samples. TNF-α was not detectable.

Explorative analyses were performed to assess distribution, skewness, variance of continuous variables, and to identify possible covariates. Between-assessment stability of neurophysiological and biological measures were assessed using the interclass correlation coefficient (ICC). Then, repeated-measure linear mixed models (RM-LMM) were employed. RM-LMM allowed us to increase power and precision of our observations, as they take within-subject measure variability into account. Separate models were built to assess between-group differences and the effect of the haplotypes on symptoms, hyperalgesia, descending modulation, inflammation, and mean DNA methylation (for each primer separately), as well to assess the association between DNA methylation and symptoms, inflammation, hyperalgesia, and descending modulation. As commonly recommended, [47] we started with a basic model including Subjects as a random term, Time as repeated measure, Group as categorical factor, age, BMI, and symptoms as covariates. Additional covariates (haplotype, DNA methylation, cytokines) were then progressively added. Each time, model’s goodness-of-fit was assessed by the − 2 Restricted Log Likelihood. Only models with the best fit were considered in further analyses. Bonferroni correction was applied to reduce the risk of false positives for multiple comparisons. P-values of Fixed-Effect tests were also adjusted, using the formula: Corrected P-value = 0.05/n° of predictors [47]. P-values were thus set at 0.008. All data were analyzed with SPSS 27 (Chicago, IL, USA).