The present trial was based on a single group pre-test and post-test design to examine the effects of an 8-week TC exercise intervention on pain, physical function, stiffness, brain functional connectivity, and plasma metabolites including amino acids and lipids. Blood samples were collected at baseline and after 8 week of TC intervention. Plasma samples were stored at −80°C for later metabolites analysis. This study was approved by the Institutional Review Board at the Texas Tech University Health Sciences Center (ClinicalTrials.gov Identifier: NCT04046003).

Postmenopausal women >50 years of age with knee pain were recruited from clinics and community centers by flyers and advertisements through newspaper. The prevalence of OA increases rapidly with age, beginning at about age 40–50 years in women, but less so in men. The hormone deficiency increases the risk of OA severity in postmenopausal women. Women older than 50 have a higher prevalence and are also more likely to report joint symptoms for the same level of radiographic severity in knee OA (37). Thus, in this study, the postmenopausal women were our target study population. If women reported bilateral symptomatic knees, only the most symptomatic knee was included in the study. Written informed consent was obtained from all subjects. Participants were selected based on the following including and excluding criteria. Inclusion criteria: (1) Postmenopausal women, (2) WOMAC pain score with at least two items out of five items are rated as moderate, severe, or extreme, respectively, (3) English literacy, (4) Able to undergo an MRI scan for subjects having MRIs, (5) current pain in the knee, and (6) medical diagnosis of knee OA or knee(s) exhibited symptoms based on American College of Rheumatology clinical classification criteria for OA (38). Exclusive criteria: (1) Prior experience with mind-body practice (e.g., TC, Qi Gong, yoga, or acupuncture) or physical therapy programs for knee OA within the past 3 months, (2) Severe medical limitations (i.e., dementia, symptomatic heart or vascular disease, or recent stroke) precluding full participation, (3) Medical/neurological or other systemic diseases affecting the musculoskeletal systems (i.e., polio/Parkinson's/multiple sclerosis, rheumatoid arthritis, uncontrolled gout, etc. in addition to cerebral vascular accident or stroke) and diabetes with peripheral neuropathy affecting their sensory/balance, (4) Intra-articular steroid injection or reconstructive surgery on most severely affected knee in the past 3 months, (5) Intra-articular hyaluronic acid injections on most severely affected knee in the past 6 months, and (6) Inability to walk without an assistive device.

The 24-form Yang style TC was employed in this study. The 24-form Yang style TC is one of the most widely practiced TC styles worldwide and one of the TC styles most widely adopted in clinical studies, with all movements well-standardized and publicized (39). The 8-week 24-form group TC program included instructed TC group classes three times per week on 3 non-consecutive days, 60 min each time at Gym of Department of Kinesiology and Sport Management, Texas Tech University, Lubbock (40, 41). Compliance of TC classes was assessed by TC class attendance record for each TC session.

Tai Chi was taught by a Master TC teacher for an 8-week period. This time selection was based on our previous studies that evaluated patients with knee OA and the effects of TC on various parameters including pain, function, range of motion, and gait parameters in 6- to 12-week trials (6, 42). The TC instructor and his assistant paid close attention to each participant during every class and made sure everyone performs TC correctly within 8 weeks. Based on our previous TC studies, all participants were able to perform the 24-form Yang style TC by themselves after 1 month. We have demonstrated in a previous study that we can teach TC and show positive effects in a group of OA participants in an even shorter period - 6 weeks (6). However, participants may subjectively expect that the intervention (Tai Chi) can improve pain, and reflect this in the subjective outcome measures (i.e., surveys), partly to please the researcher. Thus, the Hawthorne effect may be a confounder to the surveys, although not for the objective measures (e.g., fMRI, blood tests).

The Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) questionnaire was used with a scale from 26 (no difficulty) to 130 (extreme difficulty) indicating the level of difficulty associated with overall functional activities due to knee pain, including subscale of knee pain (35 points), stiffness (10 points), and physical function (85 points) (6). Pain was assessed using WOMAC-pain subscale (6), visual analog scale (VAS) (6), and Brief Pain Inventory (BPI)-pain scale (43). The BPI is a validated self-reported questionnaire that assesses pain severity using the Numerical Rating Scale for Pain Intensity (NRS-PI, 0–10 scale, where 0 = no pain and 10 = worst possible pain) for the conditions of worst, least, and average pain, as well as “pain right now” (6, 43). Physical function was assessed using WOMAC-physical function and WOMAC-stiffness subscales (6) as well as BPI-interference scale (43). These measures, WOMAC (44), VAS, and BPI (6, 45), are commonly used in knee OA studies with good validity, internal consistency, and reliability.

Laboratory lipid panel including total cholesterol, triglycerides, high-density lipoprotein (HDL)-cholesterol, non-HDL cholesterol, low-density lipoprotein (LDL)-cholesterol, very low-density lipoprotein (VLDL)-cholesterol, and Cholesterol/HDL ratio were assessed in plasma samples taken at baseline and 8 weeks. All samples were processed and analyzed in a certified diagnostic laboratory (Covenant Laboratory, Lubbock, TX).

The levels of metabolites in the plasma were determined using the AbsoluteIDQ™ p400 HR kit (BIOCRATES Life Sciences AG, Innsbruck, Austria) according to the manufacturer's instructions at the Center for Biotechnology & Genomics, Texas Tech University, Lubbock, TX. This assay allows the identification and quantification of more than 400 endogenous metabolites including 21 amino acids, 21 biogenic amines, 55 acylcarnitines, 18 diglycerides, 42 triglycerides, 24 LPC, 172 PC, 31 sphingomyelins, ceramides, and 14 cholesteryl esters. Identification and quantification of the metabolites was done using multiple reaction monitoring according to internal standards. Samples were analyzed by Q Exactive HF mass spectrometer coupled with a Vanquish ultra-high performance liquid chromatography (UHPLC) system (Thermo Scientific, USA). The amino acids and biogenic amines were separated using UHPLC with a C18 column (Biocrates, Part 9120052121032). Analytes were separated using a total 6 min gradient from solvent A (0.2% formic acid in water, to 95% Solvent B (0.2% formic acid in acetonitrile) per sample.

For lipid analyses, acylcarnitines, monosaccharides (hexose), diglycerides, triglycerides, LPC, PC, sphingomyelins, ceramides, and cholesteryl esters were analyzed by flow injection analysis (FIA) with total analysis time of ~3.8 min per sample with specific FIA mobile phase buffer (Part 9120052121018) provided in the kit, which was diluted into LC–MS grade methanol for use with the kit per manufacturer instructions. Using electrospray ionization in positive ion mode, samples for both UHPLC and flow injection analysis were introduced directly into Q Exactive Orbitrap MS systems (Thermo Fisher Scientific, Waltham, MA, USA) operating in the full scan or parallel reaction monitoring (PRM) mode. Acquisition methods and tune parameters for all instruments were provided by Biocrates as part of the p400HR kit. The concentrations of metabolites were calculated in μM. The LC–MS data were imported into the QuanBrowser module of the Thermo Xcalibur software (Thermo Fischer Scientific) for peak integration and quantification, then imported into MetIDQ^TM software package (Biocrates AG).

MRI scans were performed pre-and post-intervention with the intention of modeling the resting state functional connectivity (as measured using temporal correlations between the regions) as well structural connectivity (as modeled using diffusion parameters of water in the brain) between left and right amygdala and the mPFC. Each participant underwent two MRI scanning sessions scheduled pre- and post-intervention (0 and 12 weeks, respectively). All MRI scans were performed using a 3.0 T Siemens Skyra scanner with a 20-channel head coil located at Texas Tech Neuroimaging Institute, Lubbock, TX. Each scanning session included a scout MRI scan; an rs-fMRI scan (10 min) conducted while participants kept their eyes open; a DTI scan (12 min) and a T-weighted structural MRI scan (4.5 min). Rs-fMRI data were acquired using an echo planar imaging sequence with the following parameter settings: repetition time = 3,000 ms; echo time = 30 ms; flip angle = 90; field of view = 220 mm; matrix = 64 × 64; slice thickness = 3.4 mm; and 48 ascending axial slices. Slices were tilted ~30° from the anterior commissure—posterior commissure line to minimize orbitofrontal cortical signal dropout. Diffusion weighted images were acquired using a spin-echo-based EPI sequence with the following parameters: repetition time = 10,200 ms; echo time = 82 ms; field of view = 250 mm; matrix = 125 × 125; slice thickness = 2 mm; and 80 contiguous axial slices (a non-diffusion weighted image at b = 0 s/mm² and 62 volumes with uniformly distributed diffusion gradient directions at b = 1,000 s/mm²). The structural scan was performed using the following parameters: repetition time = 2,300 ms; echo time = 2.96 ms; flip angle = 9; field of view = 256 mm; matrix = 256 × 256; slice thickness = 1.2 mm; and 176 slices.

Descriptive statistics were calculated to inspect the distributional properties of pain, physical function, stiffness, blood lipids, and plasma metabolites. In addition, parametric and non-parametric tests were performed to examine the changes between pre- and post-intervention—paired-samples t-test for pain, physical function, and stiffness parameters and Wilcoxon signed-rank test for metabolite parameters. Prior to analysis, normality of pain and lipid outcomes was confirmed by skewness of the distributions (−0.80 to 1.50 at pre-intervention; 0.47–2.64 at post-intervention), as well as visual inspection of the Q-Q plots. The fMRI-related outcomes were normalized by using a Gaussian kernel. For pain, lipid, and metabolite outcomes, statistical significance was determined at an alpha level adjusted for multiple comparisons (i.e., Benjamin-Hochberg adjustment), and an effect size was computed for each comparison. All analyses were conducted using R (46).

Structural, resting-state functional and diffusion weighted raw data images were converted to NIfTI format using the dcm2nii converter (47). Structural images were pre-processed using Freesurfer (autorecon1) (48, 49). Functional images were preprocessed using tools in the FMRIB Software Library (FSL; Version 6.00, Oxford, UK). The following pre-processing steps were applied: motion correction by aligning each functional volume to the center volume within each functional run with 6-degrees of freedom (DOF) sinc interpolation using FSL's MCFLIRT tool (50); skull-stripping using FSL's BET tool (51); registration to high resolution structural space by the BBR algorithm and subsequently to the standard space by 12-DOF using FSL's FLIRT tool (52); spatial smoothing using a Gaussian kernel of FWHM 8.0 mm; grand-mean intensity normalization of the entire 4D dataset by a single multiplicative factor; high-pass temporal filtering (Gaussian-weighted least-squares straight line fitting, with sigma = 50.0 s); and FILM pre-whitening (53).

Resting-state fMRI data were analyzed using seed-based correlations. For the analyses, probability masks of each of left and right amygdala regions were created (54–59) using the Harvard-Oxford subcortical structural atlas in FSL and were converted to functional space of each participant using the FNIRT tool in FSL (60). Time-course of each seed region of pre- and postintervention scan was extracted from the pre-processed data using the fslmeants command in FSL. Resting state fMRI data were analyzed using a standard two-level pipeline in FSL. In level 1 analyses, subject level correlations between each seed region's time series and rs-fMRI data of the entire brain were analyzed using the Feat tool in FSL. An autocorrelation correction was included to account for serial dependencies between samples Woolrich 2001 (53). Level 2 analyses were equivalents of paired t-tests comparing the pre- vs. post-intervention level 1 estimates for each seed region (i.e., left or right amygdala). Considering the exploratory nature of the study and the limited sample size, the analyses were performed using ordinary least squares (i.e., linear approach).

Pain-associated functional dissociation of amygdala and the broader mPFC have been a consistent finding in both animal and human studies, however, there are some differences in specific sub-regions within the mPFC that showed functional dissociations. Generally, altered amygdala connectivity mainly with the infra- and pre-limbic mPFC, which are equivalent to Brodmann areas 25 and ventral 32 is reported in human subjects (13, 61–64), while human fMRI studies reported dissociated connectivity of amygdala with ventromedial PFC, including Brodmann area 25 (65) and anterior and cingulate cortical regions representing Brodmann areas dorsal 24 and 25 (64). To address potential challenges in translating the preclinical brain connectivity pattern to the human brain, we searched for clusters showing intervention-related alterations of functional connectivity with amygdala within broader target regions of mPFC (defined using a mask combing the frontal medial cortex, subcallosal cortex, paracingulate cortex and anterior cingulate cortex using Harvard-Oxford Subcortical Structural Atlas in FSL) representing Brodmann areas 32, 25, 24, and 33. Final statistical maps were thresholded using this broad mPFC mask for a z-threshold of 1.96 corresponding to a FWER corrected p-value of 0.05. Irrespective of the outcomes of the primary analyses, subject-level contrasts of parameter estimates of the mPFC of each level 1 analysis were extracted using the featquery tool in FSL (66). These contrasts of parameter estimates were used to calculate effect sizes. Furthermore, Spearman correlation analyses were performed using these parameter estimates between pre- vs. post-intervention changes in amygdala-mPFC functional connectivity and the corresponding changes in behavioral measures (i.e., VAS, WOMAC, and BPI) using R statistical software (4.0.2). fMRI data of one participant had to be excluded from the effect-size calculations and correlation analyses due to the contrasts of parameter estimates of both pre- and post-intervention scans being zero.

DTI data were subjected to distortion correction and brain extraction and were subsequently corrected for eddy currents and head motion using tools in the FDT Toolbox in FSL (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FDT). The diffusion parameters of the images were estimated using a Bayesian approach (i.e., BEDPOSTX) in FSL (67, 68). Waypoint network connectivity mapping was performed using probabilistic tractography (i.e., PROBTRACX) in FSL to estimate the connectivity between each amygdala and the mPFC which defined using Harvard Oxford Cortical and Subcortical Structural Atlases in FSL (69). While not ideal given the likelihood of regression of the post-intervention outcomes toward the mean (70), considering the pilot nature of the design and the limited sample size, pre-post comparisons of the extracted total structural connectivity were performed using Wilcoxon signed-rank tests in R statistical software (4.0.2) and the corresponding effect-sizes were calculated. Moreover, Spearman correlation analyses were performed between pre- vs. post-intervention changes in amygdala-mPFC functional connectivity and the corresponding changes in behavioral measures (i.e., VAS, WOMAC, and BPI) as well as multiple biochemical markers.

A total of 42 participants were prescreened. Among them, 17 participants met the criteria and enrolled into the study. Five participants dropped from the study due to conflict of time (n = 1), loss of interest with low compliance (<15%, n = 3), and steroid injection in knee during the study (n = 1). A total of 12 participants completed the 8-week study. All subjects were instructed to maintain their pre-existing physical activity, dietary habits, and medications, if any, throughout the study. A subgroup of participants (n = 7) received fMRI scanning for brain functional connectivity. Throughout the study, the compliance rate for TC classes was 93%. Table 1 lists demographic characteristics and medical history of study participants.

Figure 1 presents the effect of 8-week TC on pain, physical function, and stiffness. After 8-week TC exercise, subjects with knee OA had significant pain reduction in OA-affected knee as assessed by WOMAC-pain scale (p = 0.021), VAS with most pain (p = 0.018) and overall amount in the last week (p = 0.018) and right now (p = 0.068), and BPI with worst level (p = 0.018) and least amount (p = 0.033) in the past 24 h (p = 0.018). Compared to the baseline, after 8-week TC exercise, the subjects reported significantly reduced stiffness at OA-affected knee by 50%, as demonstrated by WOMAC-stiffness subscale (p = 0.020). In terms of physical function, after 8-week TC group exercise, the subjects significantly improved their physical function as shown by WOMAC-physical function subscale (p = 0.018), and reduced pain-caused interference with her ability, such as general activity (p = 0.022), mood (p = 0.033), walking ability (p = 0.018), normal work (p = 0.022), relations with other people (p = 0.047), sleep (p = 0.018), and enjoyment of life (p = 0.018).

Table 2 lists the effect of 8-week TC on plasma lipids. TC exercise significantly led to lower non-HDL-cholesterol levels (p = 0.049). Changes in non-HDL cholesterol and LDL-cholesterol were also large (Cohen's d ≥ 0.712), but they were not statistically significant at the adjusted alpha level. Also, TC exercise had no effect on plasma triglycerides, HDL-cholesterol, VLDL-cholesterol, and cholesterol/HDL ratio (all p > 0.05).

Here, we leveraged an untargeted metabolomics approach to investigate TC effect on 336 identified plasma metabolites. These metabolites comprised several chemical classes, including acylcarnitines, diglycerides, triglycerides, LPC, PC, sphingomyelins, ceramides, and cholesteryl esters. Among all metabolites, 295 metabolites were detected in at least half of the analyzed samples, which were further used in the analysis. Principal component analysis (PCA) of the plasma metabolome profiles revealed a slight clustering of the metabolome profiles before vs. after intervention (Figure 2). Moreover, we observed a shift of the metabolome profiles along the PC1 axis, which explained 28.6% of the variance, which was indicative of changes in the concentration of the metabolites caused by the intervention.

To identify the individual metabolites of which their concentration was altered by the intervention, we used the Wilcoxon signed-rank test. The change in the concentration of 36 metabolites was statistically significant (p < 0.05). We identified 11 and 25 metabolites that decreased and increased after intervention, respectively. Interestingly, the metabolites decreasing after intervention belonged to the chemical classes of biogenic amines and amino acids (Figure 3). The eight biogenic amines included serotonin (p = 0.031), cis-4-hydroxyproline (c4-OH-Pro) (p = 0.040), spermidine (p = 0.040), putrescine (p = 0.035), sarcosine (p = 0.035), symmetric dimethylarginine (SDMA) (p = 0.023), asymmetric dimethylarginine (ADMA) (p = 0.031), and kynurenine (p = 0.040). The three amino acids included methionine (p = 0.028) (71), alanine (p = 0.030), and lysine (p = 0.031) (Figure 3). In contrast, all metabolites that increased after intervention were lipids; 17 of which were LPC [namely, LPC(14:0), p = 0.035; LPC(15:0), p = 0.028; LPC(16:0), p = 0.030; LPC(16:1), p = 0.031; LPC(17:0), p = 0.030; LPC(17:1), p = 0.031; LPC(18:0), p = 0.028; LPC(18:1), p = 0.038; LPC(18:2), p = 0.040; LPC(20:1), p = 0.023; LPC(20:3), p = 0.023; LPC(20:4), p = 0.030; LPC(22:5), p = 0.031; LPC(22:6), p = 0.023; LPC-O(16:1), p = 0.040; LPC-O(18:1), p = 0.031; LPC-O(18:2), p = 0.023], five PC [namely, PC(36:5), p = 0.040; PC(38:6), p = 0.031; PC(40:9), p = 0.031; PC(42:10), p = 0.031; PC(43:6), p = 0.035)], one diglyceride (34:1) (p = 0.023), one cholesteryl ester (18:1) (p = 0.044), and an acylcarnitine (18:2) (p = 0.040) (Figure 4).

After 8 weeks of TC intervention, clusters in mPFC showed increased rs-fMRI connectivity with the left amygdala as shown in orange-yellow color (Figure 5A) and the mPFC clusters showed increased rs-fMRI connectivity with the right amygdala in blue color (Figure 5B) in the subgroup (n = 7) of subjects with knee OA. However, the pre-post difference was not significant (left amygdala with d = 0.988, p = 0.355, right amygdala with d = 0.600, p = 0.282). Similarly, DTI analyses revealed a pre-post increase in structural connectivity between bilateral mPFC and each of left amygdala (p = 0.156, d = 0.720) and right amygdala (p = 0.219, d = 0.528).

Figure 6 shows the correlation between post- vs. pre-TC intervention changes in fMRI connectively and physical function parameters. We found that there were moderate-strong non-linear correlations between pain and physical function parameters and the behavioral parameters, although these correlations were not significant, except for R/amygdala-mPFC DTI connectivity and stiffness with a negative correlation trend (unadjusted p = 0.063). For instance, there was a positive non-linear correlation between post- vs. pre-Tai Chi intervention change in physical function and L/amygdala-mPFC rs-fMRI connectivity (ρ = 0.577, unadjusted p = 0.231), suggesting that improvements in physical function may be associated with increased rs-fMRI connectivity (i.e., reversal of the pain-induced functional disconnect) between the L/amygdala and mPFC. Similarly, there was a negative non-linear correlation between post- vs. pre-TC intervention change in stiffness and R/amygdala-mPFC DTI connectivity (ρ = −0.727, unadjusted p = 0.063), suggesting that decreased stiffness with exposure to TC intervention may be associated with increased structural white matter connectivity between the R/amygdala and mPFC.