Eighteen healthy subjects (mean age 25 ± 4.74, male/female: 9/9, right/left-handed: 17/1) were recruited via local and online advertisements. Volunteers with a history of psychiatric, neurological, chronic pain condition, or history of hand/thumb trauma were excluded. Volunteers with major medical illness requiring medication, magnetic resonance imaging (MRI) contraindications or history of substance or alcohol abuse were also excluded from the study. Previous work has indicated variability in the pain responses of females depending on the phase of the menstrual cycle (29). Female volunteers completed test- retest sessions within the equivalent phase of consecutive months and those with irregular menstrual cycles were excluded. To minimise the influence of diurnal variations on pain (30, 31), participants were all tested at the same time in the day. Before each visit, participants were required to abstain from paracetamol or non-steroidal anti-inflammatory drugs for 12 h. Prior to any experiments all volunteers were informed about the study procedures, potential risks and provided written informed consent. The study was approved by the local Ethics Committee (Reference HR-18/19-8825).

All participants attended four separate sessions. In the first session participants underwent a thresholding procedure for pressure pain stimulation, which took place in a mock scanner environment. This was followed by the first imaging session that took place within 1 week of the thresholding session, where the pressure pain paradigm was administered in an MRI scanner. The third and fourth sessions were repeats of the first and second sessions respectively (i.e., second thresholding and second imaging session), and occurred after an interval of 4 weeks. The second thresholding session was performed to account for potential changes in individuals pain threshold over time. The mean interval between the two imaging sessions was 28.44 days, SD = 3.50. The mean interval between the thresholding and imaging sessions was 3.92 days, SD = 2.56.

Pressure stimuli were applied to the thumbnail of the non-dominant hand using a custom-made, automated, pneumatic, computer-controlled stimulator with a plastic piston that applies pressure via a 1.13 cm² hard rubber probe (26, 32). The thumb was inserted into a cylindrical opening and positioned such that the probe applied pressure directly to the nail bed. To maintain consistency, the same pressure device was used for both the thresholding sessions and the pressure pain paradigm in the imaging sessions.

The pressure pain thresholding procedure (26) involved one ascending series and one randomised series of pressure stimuli. During the ascending series of pressure stimuli, participants first received stimulation at 55 kilopascals (kPa) for 2 s duration. Subsequent pressure stimulations increased in steps of 4 kPa, delivered at 4 s intervals, again each lasting 2 s duration. First, participants were required to inform the investigator when they had reached their minimum pain threshold (pain score > 0 on a pain scale where 0 = “no pain” and 100 = “worst pain imaginable”). Second, participants were asked to inform the investigator when they reached their moderate pain threshold [pain score = 70 (70%)]. These values from the ascending series were then used to compute the magnitude of five different pressure intensities within the range of each subject's minimum and moderate pain thresholds. For example, if a subject's minimum pain threshold was represented by a pressure of 200 kPA and moderate pain threshold [pain score = 70 (70%)] was reached at a pressure of 600 kPA then the randomised series would consist of pressures of 200, 300, 400, 500, and 600 kPa. Subsequently, during the randomised series, each of these five pressure stimulations were delivered three times (15 stimuli in total) for a duration of 2 s in a pseudo-randomised order at 24 s intervals. During each interval participants were required to rate the intensity of their pain on a computerised pain visual analogue scale (VAS) that was presented on screen for 7 s, using their contralateral hand. Finally, a linear function was used to determine each individual's representation of a pain score of 60 (60%), built on the 15 ratings from the randomised series.

Participants first underwent localiser and structural scans followed by the pressure pain paradigm (Figure 1), utilising the same pressure probe device as used in the thresholding sessions. The paradigm was a block design consisting of sixteen alternating blocks of rest (no stimulation, passive visual fixation) and pressure pain stimuli delivered at each individual's pain score of 60% determined from the previous thresholding session. The rationale for delivering pressure set for a pain score of 60%, rather than 70% (moderate pain threshold), was to account for the repeated pressure stimuli applied in the paradigm and to make the experiment more tolerable for participants. Each block was 48 s in duration. In the pressure pain stimulus block, each pressure stimulus had a duration of 2 s, separated by a 6 s interval and for each stimulus block there were six pressure stimuli in total. At the end of each pressure pain stimuli block, participants were presented with a computerised VAS of the pain scale, as in the thresholding session, but displayed for 8 s. Apart from when the VAS was on screen, a fixation cross was displayed for the rest of the paradigm. The total duration of one run of the pressure pain paradigm was 832 s. Participants completed one run per voxel (both on the non-dominant hand) with a 5-min break to rest their hand, during each of the imaging sessions. The procedure for the second imaging session was identical to the first apart from the pressure pain stimuli delivered that depended on the values determined from each individual's second thresholding session.

The data were acquired using a GE Discovery MR750 3.0T scanner equipped with a 32-channel receive-only head coil at the Centre for Neuroimaging Sciences, King's College London, UK. A high-resolution structural 3D IR-SPGR sequence (T_R = 7.31 ms, T_E = 3.02 ms, T_I =400 ms, FOV = 270 mm, flip-angle (a) = 11°, matrix size = 256 × 256, slice thickness = 1.2 mm, 196 slices) in the sagittal plane was first acquired for localisation of the spectroscopy voxels. ¹H-fMRS spectra were acquired in two separate brain regions of interest (ROI) during the pressure pain paradigm (Figure 2). ROI-1 was an anterior cingulate cortex (ACC) region where the voxel measured 20 × 20 × 20 mm with the centre of the voxel placed 16 mm above the genu of the corpus callosum perpendicular to the anterior commissure- posterior commissure line (Figure 3A). This non-rotated ACC voxel aligns with that used in our previous ¹H-MRS studies in patient groups (33, 34). ROI-2 was a dorsal ACC (dACC) region where the voxel measured 40 × 20 × 15 mm and was positioned at an angle to sit immediately above the corpus callosum, with the anterior edge of the voxel aligned with the anterior edge of the genu of the corpus callosum (Figure 3B). The order of ¹H-fMRS acquisition was fixed with data collected from ROI-1 (ACC) first and ROI-2 (dACC) second for all participants.

The ¹H-fMRS spectra were acquired individually throughout the pressure pain paradigm using Point RESolved Spectroscopy (PRESS), with CHEmical Selective Suppression for water suppression and outer volume suppression (OVS) with Very Selective Suppression (VSS) pulses (T_R= 2,000 ms, T_E = 105 ms, phase cycle length= 2, 420 acquisitions, 16 water unsuppressed acquisitions). Before each ¹H-fMRS scan, shimming was optimised with an auto-prescan performed twice. A modified version of GE's PROBE (proton brain examination) sequence was used to commence the pressure pain paradigm and send a trigger pulse at the start of every T_R.

Analysis of ¹H-fMRS data was performed using the TARQUIN software package, version 4.3.10 (35). The TARQUIN algorithm performed a fully automated fit to the ¹H-fMRS data using a predefined basis set comprising of the following components: alanine; aspartate; creatine (Cr); gamma-aminobutyric acid; glucose; glutamine; glutathione; glutamate (Glu); glycerophosphorylcholine; myo-inositol; lactate; lipid peaks at 0.9, 1.3a, 1.3b, and 2.0 ppm; macromolecules at 0.9, 1.2, 1.4, and 2.0 ppm; N-acetyl-aspartate (NAA); N-acetyl- aspartateglutamate; phosphorylcholine; phosphocreatine (PCr); scyllo-inositol; and taurine.

¹H-fMRS spectra acquired during the “REST” and “PAIN” conditions (8 blocks of each) from each ROI were averaged within subjects and estimates of Glx (Glu + glutamine) were determined. Only Glx values were reported as with the T_E used in this study glutamate and glutamine may not be reliably differentiated. Averaged ¹H-fMRS spectra had to meet the following quality control measures to be included in subsequent analysis: (1) Cramer-Rao minimum variance bounds (CRMVB) estimate of the standard deviation of Glx < 20%; (2) signal-to-noise ratio (SNR) > 5; (3) full-width half- maximum (FWHM) < 0.1 ppm and (4) fit quality (Q) < 2.5. No spectra had to be discarded following the quality control process. Total creatine (tCr = Cr + PCr) was used to provide a reference for Glx values in each sub spectrum to alleviate possible confounds of scanner drift that would not be adequately corrected if a single water reference that was acquired at the end of the scan was used instead.

To assess consistency of voxel tissue composition, IR-SPGR images were segmented using SPM12 (http://www.fil.ion.ucl.ac.uk/spm/software/spm12/) into grey matter (GM), white matter (WM) and cerebrospinal fluid (CSF) and mean tissue proportions and standard deviations were determined across subjects and scanning sessions.

Normality of data was examined using Shapiro–Wilk tests. To evaluate a priori hypotheses, two-way repeated measures analyses of variance (ANOVAs) were used to examine differences in Glx/tCr, with CONDITION (REST and PAIN) and SESSION (Session A and Session B) as within-subject factors, including CONDITION^*SESSION interaction terms, for each ROI individually. As future research applications may wish to evaluate Glx/tCr responses during a single scanning session (for example in case-control designs), and to facilitate qualitative comparison with previous research using similar sample sizes that applied only a single session (21, 22), we additionally investigated whether effects of CONDITION reached significance in session A alone, using paired samples t-test.

To assess the test-retest reliability across scanning sessions for the change in Glx/tCr in the PAIN compared to REST condition, the intra-class correlation coefficient (ICC), was computed using a two-way mixed effects model, single measurement for absolute agreement alongside the within-subject percentage coefficient of variation (%CV). Test-retest variability (VAR) was also computed to assess reproducibility. Reliability and reproducibility tests was only evaluated where significant main effects of CONDITION were detected.

Exploratory analyses examined for potential adaptation effects for Glx/tCr across individual blocks of the pressure pain paradigm using three-way repeated measures ANOVAs with BLOCK, CONDITION and SESSION as within-subject categorical factors (including BLOCK^*CONDITION, BLOCK^*SESSION, CONDITION^*SESSION and BLOCK^*CONDITION^*SESSION interaction terms). These analyses were performed separately for each ROI and all Glx values for individual blocks, including those with CRMVB >20%, were included.

Finally, Pearson's product moment correlation coefficients were used to examine any relationships between ΔGlx/tCr between REST and PAIN conditions and VAS pain scores (averaged within-subject and compared across subjects) for each scanning session. For all analyses, a significance threshold of p < 0.05 was specified. Analyses were carried out using R version 3.6.3 and IBM SPSS Statistics version 27.

The mean stimulus pressures delivered (determined from thresholding sessions) and mean pain VAS scores from the ¹H-fMRS pressure pain paradigm for each ROI are shown in Table 1. The mean stimulus pressure delivered and mean pain VAS scores for each ROI remained stable with no statistically significant differences across sessions (Table 1). Mean pain VAS scores were higher for the second run, dACC (ROI-2), than the first run, ACC (ROI-1) for both sessions. Mean pain VAS scores across individual PAIN blocks and across sessions for each ROI are shown in Supplementary Figure 1.

Two-way repeated measures ANOVA demonstrated a significant main effect of CONDITION for Glx/tCr in the dACC (ROI-2) [F_(1,17) = 6.075, p = 0.025], with higher Glx/tCr during PAIN compared to REST. There was no significant main effect of SESSION [F_(1,17) = 0.672, p = 0.424] and no significant CONDITION ^* SESSION interaction [F_(1,17) = 0.028, p = 0.870] for dACC Glx/tCr (Table 2). dACC Glx/tCr during REST and PAIN is presented in Table 3 and Figure 4. The increase in dACC Glx/tCr during PAIN compared to REST did not reach statistical significance for session A alone [t₍₁₇₎ = 2.012, p = 0.060].

For ΔGlx/tCr in the dACC ROI between the averaged blocks of REST and PAIN there was poor test-retest reliability (ICC = 0.272, %CV = 10.5%) and limited reproducibility (VAR = 10.3). There was no significant relationship between ΔGlx/tCr in the first and second sessions (r = 0.264, p = 0.289).

For Glx/tCr in the ACC (ROI-1), there was no significant CONDITION ^* SESSION interaction [F_(1,17) = 0.202, p = 0.659] and no significant main effects of CONDITION [F_(1,17) = 2.620, p = 0.124] or SESSION [F_(1,17) = 2.703, p = 0.119] (Table 2).

Exploratory analyses with a three-way repeated measures ANOVA, examining Glx/tCr across individual blocks, demonstrated a significant main effect of CONDITION for Glx/tCr in the dACC [F_(1,17) = 10.407, p = 0.005] with no significant interactions (i.e., BLOCK^*CONDITION, BLOCK ^* SESSION, CONDITION^* SESSION or BLOCK^*CONDITION^*SESSION) and no significant main effects of BLOCK or SESSION (Supplementary Table 1). For Glx/tCr in the ACC (ROI-1) there were no significant interactions and no significant main effects of BLOCK, CONDITION or SESSION for the repeated measures ANOVA (Supplementary Table 1).

Uncorrected tCr signal values were found to be stable for averaged REST and PAIN conditions and across sessions with no significant CONDITION ^* SESSION interactions and no significant main effects of CONDITION or SESSION for either ROI (Table 2). Furthermore, tCr signal values remained stable across individual blocks of REST and PAIN (Supplementary Figure 2). Three-way repeated measures ANOVAs revealed no significant interactions and no main effects of BLOCK, CONDITION or SESSION on tCr values in either ROI (Supplementary Table 1).

For the ACC ROI, the mean ± standard deviation tissue proportions were 65 ± 4% GM, 10 ± 3% WM, 25 ± 6% CSF and 65 ± 4% GM, 10 ± 3% WM, 25 ± 5% CSF for session A and B respectively. For the dACC ROI, the mean tissue proportions were 63 ± 4% GM, 21 ± 6% WM, 16 ± 5% CSF and 62 ± 4% GM, 21 ± 6% WM, 16 ± 5% CSF for sessions A and B, respectively. The relatively small spread of proportions indicates consistency of voxel position across subjects. Furthermore, there were no significant differences in GM, WM and CSF tissue proportions between sessions indicating consistency of voxel positioning across scans (Supplementary Table 2). Example voxel positioning and corresponding ¹H-fMRS spectra are shown in Figures 3A,B.

For total averaged block conditions, mean SNR for the ACC ROI was 31.4 ± 5.7 for REST and 31.5 ± 5.8 for PAIN. For the dACC ROI, these values were 34.9 ± 5.0 and 35.4 ± 4.6 for REST and PAIN conditions, respectively. There were no significant differences in SNR between REST and PAIN conditions in either ROI (Supplementary Table 3).

Correlations between ΔGlx/tCr values and pain VAS are shown in Table 4. There were no consistent significant associations between ΔGlx/tCr (PAIN–REST) conditions and pain VAS for all blocks combined across each scanning session.