The reliability of two prospective cortical biomarkers for pain: EEG peak alpha frequency and TMS corticomotor excitability

Abstract

Background.

Many pain biomarkers fail to move from discovery to clinical application, attributed to poor reliability and an inability to accurately classify at-risk individuals. Preliminary evidence has shown that high pain sensitivity is associated with slow peak alpha frequency (PAF), and depression of corticomotor excitability (CME), potentially due to impairments in ascending sensory signalling and descending motor pathway signalling respectively

New Method.

The present study evaluated the reliability of PAF and CME responses during sustained pain. Specifically, we determined whether, over several days of pain, a) PAF remains stable and b) individuals show two stable and distinct CME responses: facilitation and depression. Participants were given an injection of nerve growth factor (NGF) into the right masseter muscle on Day 0 and Day 2, inducing sustained pain. Electroencephalography (EEG) to assess PAF and transcranial magnetic stimulation (TMS) to assess CME were recorded on Day 0, Day 2 and Day 5.

Results.

Using a weighted peak estimate, PAF reliability (n = 75) was in the excellent range even without standard pre-processing and ~2 minutes recording length. Using a single peak estimate, PAF reliability was in the moderate-good range. For CME (n = 74), 80% of participants showed facilitation or depression of CME beyond an optimal cut-off point, with the stability of these changes in the good range.

Comparison with existing methods.

No study has assessed the reliability of PAF or feasibility of classifying individuals as facilitators/depressors, in response to sustained pain. PAF was reliable even in the presence of pain. The use of a weighted peak estimate for PAF is recommended, as excellent test-retest reliability can be obtained even when using minimal pre-processing and ~2 minutes recording. We also showed that 80 % of individuals exhibit either facilitation or depression of CME, with these changes being stable across sessions.

Conclusions.

Our study provides support for the reliability of PAF and CME as prospective cortical biomarkers. As such, our paper adds important methodological advances to the rapidly growing field of pain biomarkers.

Many researchers have shifted focus to the development of biomarkers that can be used in the diagnosis, prevention and treatment of chronic pain [1, 2]. However, many prospective pain biomarkers fail to move from the initial discovery stage to clinical application [1]. Reason for this failure include poor biomarker reliability and an inability to accurately classify at-risk individuals [1, 3]. Overcoming these limitations is an important step in biomarker development, particularly prior to subsequent clinical validation and intervention studies.

Recent work [4-10] has shown promise for a cortical biomarker pain signature involving two measures: peak alpha frequency (PAF), which is measured using resting-state electroencephalography (EEG) and refers to the dominant oscillatory frequency in the 8-12Hz range [11-13] and; corticomotor excitability (CME), which is measured using transcranial magnetic stimulation (TMS), and can be indexed as the volume of the primary motor cortex (M1) representation of a peripheral muscle [14]. Using intramuscular injection of Nerve Growth Factor (NGF), which induces clinically meaningful and sustained pain for 2-3 weeks, studies [5, 6, 8, 9] have shown that individuals with reduced CME during pain (“depressors”) and slow PAF prior to pain onset, experience more pain than those who show increased CME during pain (“facilitators”) and fast PAF prior to pain onset. These findings suggest that slow PAF and reduced CME may reflect an increased sensitivity to prolonged pain that predates chronic pain onset. Further evidence to support this possibility is drawn from studies that have shown slower PAF (lower frequency) [15-17] and reduced CME [18, 19] in clinical pain populations relative to healthy controls. As there are currently no clinically meaningful brain biomarkers available for predicting the transition to chronic pain [20], such results are promising. Indeed, there has been a recent increase in research using PAF [21-24] and CME [9, 25, 26] in the context of predicting and/or modulating pain sensitivity, suggesting evaluations of the reliability of these measures is critical.

The mechanisms through which PAF and CME are associated with pain remain unclear. Alpha oscillations are believed to inhibit incoming sensory signals [12, 27] – the speed of these oscillations varies considerably between individuals, contributing to individual differences in various cognitive and perceptual processes such as working memory and attention [12, 28, 29]. As such, one explanation for the association between PAF and pain is that faster PAF reflects a greater capacity to inhibit incoming sensory input [30]. In contrast, variations in CME during pain may reflect individual differences in descending motor pathway signalling [9]. During sustained pain, individuals show large variability in motor adaptations, as individuals are required to balance the risk of increased pain against motor task performance [5, 31]. Prior studies have shown a relationship between reduced CME and the adoption of less variable movement strategies [31] that could increase tissue loading [32-35] and predispose to the maintenance and reoccurrence of pain [35, 36].

While CME and PAF hold promise as biomarkers, there are still outstanding questions regarding their reliability. The first question is whether PAF remains stable during prolonged pain. Though several studies have shown that PAF remains stable over time [37-40]. experimental pain has been shown to influence both PAF and alpha power [41, 42]. As such, the presence of pain may introduce fluctuations in PAF that may impact the usefulness of pre-pain classification methods. The second question is whether individuals show two stable and distinct CME responses during pain. Only one study has characterised the individual level responses in CME during pain sustained over several days [9]. However, this study focused on upper limb pain, whereas corticomotor adaptations in other body regions may differ. One such region is the masseter muscle, which in contrast to the upper limb, is involved in different motor functions and is subserved by trigeminal motoneurons rather than spinal motoneurons [25, 43]. Demonstrating two distinct responses to pain (facilitation and depression) in the masseter muscle would improve the generalizability of CME classifications.

The aim of the present study was to a) evaluate the reliability of PAF over several days and b) determine whether individuals show two stable and distinct patterns of CME change during sustained pain: corticomotor facilitation and depression. This is the first major step in the overall goals of the PREDICT study, an ongoing initial biomarker validation study for CME and PAF [44]. To address our research aims, we used a subsample of participants from PREDICT. Participants were administered with an intramuscular injection of NGF to the right masseter muscle at baseline (Day 0), and two days after baseline (Day 2). CME and PAF were measured on Day 0 (prior to the injection), Day 2 and Day 5. This paper only assessed the reliability of the candidate biomarkers. Future studies will evaluate the performance of these combined biomarkers in classifying pain-sensitive individuals, as outlined in our protocol paper [44].

Methods

Design

This paper is an interim analysis which aims to assess biomarker reliability, and not their relationship to pain. The latter will be the goal of the PREDICT trial [44], which will undertake analytical validation of the PAF and CME biomarkers in a human model of temporomandibular disorder (TMD). This study used a longitudinal experimental design where participants experienced the development and complete resolution of NGF-induced right masseter muscle pain, with outcomes collected over a period of 30 days. The PREDICT trial is prospectively registered on ClinicalTrials.gov (NCT04241562) and a protocol paper has been published [44].

Participants

The study included the first 85 participants who completed the PREDICT study. The inclusion criteria was healthy male and female participants between the ages of 18-44. Participants were excluded for the following reasons: 1) presence of acute pain, 2) history or presence of chronic pain, 3) history or presence of a medical or psychiatric condition, 4) frequent use of alcohol, opioids, or illicit drugs in the past 3 months, 5) pregnant or lactating women, 6) contraindicated for TMS (e.g., metal implants, epilepsy). The final sample consisted of 38 females and 47 males, with an age range from 18-43 and a mean age of 25.72 ± 5.96. Participants were recruited via advertisements placed on community notice boards, social media platforms and a healthy participant volunteer database. Ethical approval was obtained from the University of New South Wales (HC190206) and the University of Maryland Baltimore (HP-00085371). All procedures were conducted in accordance with the Declaration of Helsinki. Written, informed consent was obtained, and participants were free to withdraw from the study at any time.

Experimental Protocol

Participants attended the laboratory on Day 0, Day 2, and Day 5, with facial pain, PAF and CME measured at each session. NGF was injected into the right masseter muscle at the end of the Day 0 and Day 2 laboratory sessions. As part of the larger PREDICT study, other data (not reported here), included electronic pain diaries collected from Day 0 to Day 30, questionnaire data collected on Day 0, Day 2 and Day 5, and pressure pain thresholds obtained on Day 0, Day 2 and Day 5.

Data collection procedures

Facial Pain Questionnaires

At the beginning of each session, participants rated their facial pain on an 11-point numerical rating scale at rest, chewing, swallowing, drinking, talking, yawning and smiling. The scale ranged from 0 = “no pain” to 10 = “worst pain imaginable”.

Peak alpha frequency (PAF)

Participants were seated in a comfortable chair. Scalp EEG was recorded using the Brain Products platform (BrainVision Recorder, Vers. 1.22.0101 with actiCHamp Plus, Brain Products GmbH, Gilching, Germany). Recording was initially done at a sampling rate of 25000 Hz (participants 1-16), though this was subsequently reduced to 5000 Hz (participants 17-85) as the higher sampling rate was deemed unnecessary to obtain reliable resting state alpha activity. Signals were recorded from 63 active electrodes (actiCAP slim, Brain Products GmbH, Germany), embedded in an elastic cap (EASYCAP, EASYCAP GmbH)in line with the 10-10 system. Recordings were referenced online to ‘FCz’ and the ground electrode placed on ‘FPz’. Electrode impedances were maintained below 25 kOhms during setup. Based on previous research [45, 46] and the actiCHamp manual, we anticipated negligible signal loss with this impedance cut-off. Once setup was complete, the lights were switched off with ambient noise reduced to a minimum. Participants were instructed to relax their muscles and keep their eyes closed while remaining awake. The resting-state EEG signal was then recorded for 5 minutes.

Corticomotor excitability (CME)

Electromyography.

Participants were seated in a comfortable chair and viewed a monitor that displayed electromyographic (EMG) signals used for live feedback. Bipolar surface electrodes were used to record EMG from the right masseter muscle. A belly-tendon montage was used, with the active (belly) and reference (tendon) electrodes placed along the mandibular angle, and ground electrode placed on the right acromion. EMG signals were amplified (x 1000) and filtered (16 to 1000Hz), and digitally sampled at 2000 Hz.

Maximum Voluntary Contraction.

It is common to assess CME from the masseter muscle during active contraction that is at a certain percentage of the maximum voluntary contraction (MVC) (see [25, 43]). This is because it is challenging to obtain reliable motor evoked potentials (MEPs) from the masseter muscle at rest [47, 48]. We instructed participants to contract at 20% MVC as this provided a balance between obtaining sufficiently large MEPs for analyses [49], while also minimizing fatigue during potentially lengthy mapping sessions. To measure MVC, participants were actively encouraged to clench their jaw as hard as possible for 3 seconds on 3 separate trials, with a 1-minute rest break in between trials. The MVC was computed based on the average of the 3 trials, and 20% MVC was computed based on this average. Participants were provided with practise at maintaining a masseter contraction of 20% MVC by receiving live feedback from the EMG signal.

Transcranial Magnetic Stimulation Hotspot.

Single, monophasic stimuli were delivered using a Magstim unit (Magstim Ltd., UK) and 70mm figure-of-eight flat coil. In line with previous studies investigating optimal coil orientation for inducing masseter MEPs, an angle of 90 degrees between the anterior-posterior line and the coil handle was used [50]. This orientation induced a current in the lateral-to-medial direction. Participants wore a swim cap with a grid of 1cm x 1cm resolution. The scalp site evoking the largest MEP (i.e., the “hotspot”) during active contraction was then determined.

Thresholding.

The TMS motor threshold assessment tool [51] was used to determine the active motor threshold (AMT), defined as the minimum intensity required to evoke a reliable masseter MEP during active contraction. A reliable MEP was identified if the EMG waveform between ~5-15ms [52, 53] after the TMS pulse was visually larger in amplitude relative to background EMG.

Mapping.

The procedure for mapping has been described in detail previously [54, 55]. The TMS intensity was set at 120% AMT. During active contraction, 3 stimuli were successively delivered at each location around the grid, starting at the hotspot. The number of stimulation sites was pseudorandomly increased until an MEP was no longer observed (i.e., no reliable MEP in all 3 trials at all border sites). Note that on Day 2 and 5, the mapping procedure began immediately after the MVC was determined, with the motor hotspot and threshold procedures skipped. In line with the approach used in previous studies [5, 31, 56], the hotspot location and test stimulus intensity determined on Day 0 was used for subsequent mapping procedures on Day 2 and Day 5. This approach ensures that any changes in motor cortical maps occurring as a result of NGF-injection can be observed. Although we did not re-assess AMT on Day 2 and 5, a previous study showed no change in AMT in the masseter after NGF injection [25].

Intramuscular injection of nerve growth factor (NGF)

The NGF injections were provided at the end of the Day 0 and Day 2 sessions. To prepare the masseter muscle for injection, the surface was cleaned using alcohol wipes. A sterile solution of recombinant human NGF (dose of 5 μg [0.2 ml]) was then administered as a bolus injection into the muscle belly of the right masseter using a 1-ml syringe with a disposable needle (27-G). The needle was inserted perpendicular to the masseter body until bony contact was reached, retracted ~2mm, and NGF injected.

Data processing

Peak Alpha Frequency

EEG data processing was conducted using custom MATLAB (R2020b, The Mathworks, USA) scripts implementing the EEGLAB (eeglab2019_1) [57] and Fieldtrip (v20200215) toolboxes [58].

While our overarching aim was to determine whether PAF was reliable during sustained pain, we also looked at various methodological factors that may influence this reliability. Resultant PAF estimates can be influenced by a number of factors, including pre-processing pipeline [59], choice of frequency window [7], duration of the EEG recording [60] and peak identification method [12]. Assessing the influence of these methodological factors will assist in the development of standardized protocols for obtaining reliable PAF estimates.

To gain an in-depth understanding of the stability of PAF during pain, we investigated how test-retest reliability of PAF was influenced by these methodological factors.

Impact of Pipeline.

Two pre-processing pipelines were compared. These pipelines were labelled the “Standard cleaning” and “No cleaning” pipelines. The standard cleaning pipeline matched those used in previous studies investigating the relationship between PAF and pain: data downsampled to 500Hz, re-referenced to average, and band-pass filtered between 2 and 100Hz using an FIR filter. Channel data was visibly inspected, and overtly noisy channels were removed, and the signal re-referenced. Data were segmented into 5 second epochs, and epochs containing marked muscular artefacts were manually rejected. Principal component analysis was applied to identify and remove components relating to eye blinks and/or saccades. Removed channels were then interpolated using the nearest neighbour method [58]. In the “no cleaning” pipeline, the data was downsampled to 500 Hz, re-referenced to average, and segmented into 5 second epochs. For both pipelines, the power spectral density was derived in .2Hz bins. The 2-50Hz range was extracted using Fast Fourier Transform. A Hanning taper was applied to the data to reduce edge artefacts.

Impact of Recording Length.

Two recording lengths were analysed: the first half of the remaining epochs vs. all the remaining epochs after excluding bad trials.

Impact of Frequency Window.

The peak frequency in the alpha range was calculated across two frequency windows: a wider 8-12 Hz window vs. a narrower 9-11 Hz window.

Impact of Peak Identification Method.

Two methods of calculating PAF were compared. One was the peak picking method [61], which is the single frequency that yielded the largest power within the 8-12 or 9-11 Hz range. The other was a weighted frequency approach, known as the Centre of Gravity method, calculated using the following equation:

$$CoG=\frac{{\sum }_{i=1}^n{f}_i\ast a_i}{{\sum }_{i=1}^n{a}_i}$$

Where f_i is the ith frequency bin within the frequency window of interest, and a_i represents the spectral amplitude at f_i.

Note that analysis of the sources of alpha was not conducted as this was not within the scope of the study. Though source reconstructions methods are available, a multi-method approach (e.g., MRI informed EEG) is usually required to accurately localise alpha generators reliably [62]. Test-retest reliability was assessed by computing the intraclass correlation coefficient (ICC, two-way fixed, single measure) [63] of PAF at each electrode across the three sessions. An ICC ≤ 0.2 indicates poor relative reliability, 0.21 to 0.4 fair, 0.41 to 0.6 moderate, 0.61 to 0.8 good, and ≥ 0.81 excellent relative reliability [64].

Corticomotor excitability

TMS data was processed using a custom MATLAB script. For 15 randomly selected participants, the onsets and offsets of the MEPs were manually determined for each trial. The mean onset and offset across these participants were then used to fix the MEP onsets and offsets across all participants. To confirm if this was reliable, we determined the correlations in the key outcome measure between the fixed MEP window method vs. the manual selection method. The root mean square (RMS) for each MEP window was determined. The RMS of background EMG was determined using a fixed window between 55 and 5ms before the TMS pulse [65]. This was subtracted from the RMS of the MEP window to determine MEP amplitude. The mean MEP amplitude at each stimulation site was determined. Map volume was calculated by summing the amplitudes of sites exhibiting >10% of the maximum MEP amplitude [66].

For each participant, a CME change score was determined by computing the map volume for Day 2 and Day 5 as a proportion of Day 0 [5]. To determine the reliability of these change scores, the ICC between the Day 2 and Day 5 change score was computed.

To determine whether there were two distinct CME responses, the proportion of participants that showed an increase (facilitator) or decrease (depressor) in CME beyond an optimal cut-off point was determined. We constructed a receiver operating characteristic (ROC) curve showing the performance of different cut-off values (mean percentage change in map volume across Days 2/5) for separating participants into depressors and facilitators. Using this approach, we calculated the Youden function over all possible cut off values, and determined the optimal cut-off value for balancing sensitivity and specificity. A chi-squared test was then conducted to determine whether a significant proportion of participants demonstrated an increase or decrease in map volume beyond this optimal cut-off score

Results

Reliability of Peak Alpha Frequency

Ten participants had missing data on Day 0, 2 or 5 due to illness or unwillingness to continue the procedures. This left 75 participants in the final analysis on PAF. For the standard cleaning pipeline, the mean number of channels excluded per recording was 0.5 (SD = 1.09, range = 0-8). The mean number of bad epochs excluded per recording was 1.26 (SD = 2.44, range = 0-18). Given the range of remaining epochs, this suggests that for the standard cleaning pipeline, the remaining full and half recording lengths were on average, close to 5 and 2.5 minutes respectively, with a range of 3.5-5 and 1.75-2.5 minutes respectively. The mean number of eye movement components removed per recording was 1.41 (SD = 0.75, range = 0-2).

Figure 1 shows the individual and group level z-transformed spectral plots (averaged across electrodes) for each day, and for both standard and non-standard cleaning pipelines.

Figure 1.

Individual and mean (n=75, in black) EEG spectral plots (averaged across electrodes) for each day, and for both standard and no cleaning pre-processing pipelines.

Figure 2 shows topographies and scatterplots for ICCs in PAF as a function of pipeline, recording length, frequency window and calculation method. When using the Centre of Gravity Method, ICCs were in the “excellent” range for almost all channels regardless of frequency window, pipeline, or recording length. For the peak picking method, ICCs were mostly in the “good” range, though the occurrence of “moderate” range ICCs was higher when using an 8-12Hz compared to the 9-11Hz frequency window. These findings suggest that while the use of the Centre of Gravity method leads to excellent reliability regardless of methodological considerations, reliability for the peak picking method varies between moderate to good depending on whether a wider or narrower frequency window is used to calculate PAF.

Figure 2.

Topographies and scatterplots for ICCs in PAF as a function of pre-processing pipeline, recording length, frequency window and calculation method. Respectively, green, blue and magenta represent excellent, good and moderate test-retest reliability of PAF between sessions. The scatterplots show PAF on Day 0 plotted against the mean PAF of Day 2 and Day 5 for a representative channel (Cz). PAF calculated with the COG method had excellent reliability, regardless of the pre-processing pipeline or recording length.

Changes in Corticomotor Excitability

Eleven participants had missing TMS data on Day 0, 2 or 5 due to illness or unwillingness to continue the procedures. This left 74 participants in the final analysis on CME. The mean MVC and 20% MVC values were 0.048V (±0.036) and 0.009V (0.007), respectively. The mean AMT and test stimulus intensities were 42.78 (8.50) and 51.38 (10.18), respectively.

Figure 3 shows an exemplar masseter MEP. MEP windows were selected manually for 15 participants (45 sessions in total).

Figure 3.

An exemplar masseter MEP. The window between 5 and 55ms before the TMS pulse represents the background EMG window. The blue highlighted window shows the window between the onset and offset of the MEP.

Figure 4 shows a scatterplot of map volumes when MEP windows were selected manually vs. fixed windows. The R² of .99 suggests both methods led to near-identical estimates of map volume, supporting the reliability of the fixed window method.

Figure 4.

A scatterplot of 45 map volume estimates when MEP windows were selected manually vs. fixed windows.

Shapiro-Wilk tests showed that scores on map volume change (as a proportion of Day 0) violated assumptions of normality for both Day 2 (W = 0.66, p < .001) and Day 5 (W = 0.77, p < .001). As such, these change scores were log-transformed, after which assumptions of normality were satisfied for Day 2 (W = 0.98, p = .29) and Day 5 (W = 0.99, p = 0.69). Figure 5 plots the log-transformed change scores on Day 2 against Day 5. The ICC between these measures was 0.63, suggesting the stability of the CME response to pain was in the good range.

Figure 5.

Corticomotor map volume change scores on Day 2 plotted against change scores for Day 5. The blue plots show participants who demonstrated a reduction in map volume on both Day 2 and 5, while the red plots show the opposite pattern. The orange plots show participants that demonstrate “unstable” changes i.e., a decrease on one day but an increase on the other.

Figure 6 shows the motor cortex maps for those who showed any mean increase in map volume and those who show any mean decrease in map volume. These maps display the mean unadjusted MEP amplitudes at each stimulated location. We constructed a ROC curve representing the performance of different cut off values (of log percentage change in map volume) for separating participants into depressors and facilitators. Using this approach, we calculated the Youden function over all possible cut off values, and found that the optimal cut-off value for balancing sensitivity and specificity was a ~7% change in map volume on the log scale (17.5% on the non-logged scale). The proportion of individuals demonstrating a change beyond the cut-off score was 59/74, which represented a significant proportion of individuals (χ² = 26.17, p < .001). 23/30 participants showed an increase in map volume beyond the cut-off point, which represented a significant proportion (χ² = 8.53, p= .003), while 36/44 showed a decrease in map volume beyond the cut-off point, which also represented a significant proportion (χ² = 17.81, p< .001).

Figure 6.

Motor cortex maps for those who showed a mean increase in map volume across days (n = 30) and those who showed a mean decrease in map volume across days (n = 44). The map shows the mean MEP amplitude at each stimulation site relative to the vertex. The figure on the left shows roughly where TMS was applied over the scalp. The dashed green lines represent the horizontal and vertical lines over the vertex. .

The Relationship between Corticomotor Excitability and Peak Alpha Frequency

In addition, we explored the relationship between PAF and CME to determine whether there was evidence of coupling between these mechanisms. Specifically, we determined whether PAF on Day 0 (averaged across left motor electrodes C1, C3, C5, FC1, FC3 and FC5) was associated with the mean change in CME on Day 2/5 as a proportion of Day 0. There was no evidence of an association, such that individuals with slower PAF on Day 0 were not more likely to show stronger suppression of CME (r = 0.13, p = .27).

Discussion

The aim of the present study was to determine a) whether PAF remains stable over several days of sustained pain, and b) whether individuals show two stable and distinct patterns of CME change during sustained pain. The EEG results showed moderate to excellent reliability in PAF, with excellent reliability observed when using a weighted frequency approach to calculate PAF. We also found that 80% of participants showed facilitation or depression of CME beyond an optimal cut-off point, with the stability of these changes in the good range.

PAF is stable during sustained pain

The present study extends the literature by demonstrating that PAF remains stable even in the presence of experimentally induced pain, though reliability estimates can be influenced by various methodological factors. The most critical factor was the peak identification method, with the centre of gravity method leading to excellent reliability estimates, and the peak picking method leading to moderate to good reliability estimates in PAF. It is conceivable that the use of a single peak estimate is more prone to between sessions fluctuations given the potential presence of multiple peaks, with even more peaks introduced when using a wider frequency window (8-12Hz). In contrast, the use of a weighted peak estimate takes into account the presence of multiple peaks and is thus less susceptible to between-sessions fluctuations [67]. Note that while other frequency windows (e.g. 6-14Hz) may have been considered, a previous study [7] showed that PAF estimations are similar regardless of window size since the central spectral power is still located in the 9-11 Hz. Interestingly, for both the centre of gravity and peak-picking methods, reliability was minimally affected by pre-processing pipeline or recording length. This is consistent with a recent study that showed similar PAF estimates between pre-processing pipelines [68], though this previous study did not investigate a “no pre-processing” pipeline that only involved re-referencing the signal, nor was between-sessions reliability during pain investigated. The lack of influence of pre-processing and recording length is likely due to the nature of the data collection method, wherein participants were instructed to relax their muscles with eyes closed, minimizing the influence of muscle and eye movement artefacts during the recording. The relative ease at which PAF data can be obtained is promising for subsequent stages of biomarker development, including clinical validation and intervention studies.

In summary, we showed that PAF reliability varied between moderate to excellent, suggesting evidence of stability of PAF even in the presence of pain. However, we recommend future research use the centre of gravity method which can achieve excellent reliability even with minimal pre-processing and shorter recording length.

Majority of individuals show facilitation or depression during sustained pain

The findings suggest that experimentally induced musculoskeletal pain induced either corticomotor facilitation or depression in 80% of individuals. A large proportion of individuals showed an increase or decrease in corticomotor excitability beyond a cut-off point, which was identified as a minimum of ~17.5% change in map volume from Day 0. The percentage of the sample showing a mean increase (30/74 i.e. 40%) or decrease (44/74 i.e. 60%) in CME was similar to what was observed in a previous study using the NGF model in the elbow muscle [5]. As such, the results provide support for the generalizability of distinct corticomotor responses to pain in different muscles with markedly different motor functions. Further work is needed to determine whether our findings also generalize as a function of other factors that might influence CME, such as the level of muscle contraction and test stimulus intensity.

It is important to note that no study has determined the proportion of participants showing increases or decreases in CME during pain beyond a cut-off point as determined using an ROC approach, nor has any study determined whether these changes are stable. The reasons for these distinct responses remain unclear, however research suggests that when pain is sustained, individuals are required to balance the risk of increased pain against task completion (e.g., chewing, yawning) resulting in variations in motor strategies i.e., increased vs. decreased CME [5, 31]. Some of these motor strategies are less adaptive and may be a predisposing factor for the transition to chronic pain during acute injuries [69, 70]. As such, our results provide novel characterisation of individualised motor strategies in response to sustained pain, with further work required to determine whether these strategies are indeed involved in the transition to chronic pain.

Limitations and Future Directions

One limitation of the study was the lack of assessment of PAF and CME beyond Day 5, especially given NGF-induced pain can last 2-3 weeks in some individuals [5]. Assessment of PAF stability and corticomotor excitability over a longer period may provide meaningful information on the stability of these biomarkers to pain, including the amount of time that cortical adaptations last, and whether these adaptations persist after the resolution of pain.

Another potential limitation is the absence of a sham NGF injection to confirm that the changes in CME were not random fluctuations. However, studies have shown that measures of M1 organization are stable and reliable over time [71-73]. Moreover, the use of sham injections in both animal and human studies of experimental pain have shown no effects on motor cortex activity measured using TMS [74, 75], functional magnetic resonance imaging [76] or intracortical microstimulation [76, 77]. Another study showed that repeated injections of NGF induced changes in CME that returned to baseline levels on Day 14, suggesting pain specific adaptations in CME. The present study also showed a correlation between the CME responses at Day 2 and Day 5, suggesting the responses relative to Day 0 were likely not random. Taken together, these findings suggest changes in CME observed in our study are likely not natural fluctuations over time.

It is possible that the association between PAF and pain reflect individual differences in the ability to inhibit ascending pain signals, while the CME response reflects adaptations in descending motor pathway signalling. One of our analyses investigated the potential association between these mechanisms, however, we found no evidence that individuals with slower PAF were more likely to show stronger suppression of CME. This might suggest that the mechanisms for PAF and CME are independent, which is plausible given PAF relies on larger scale global brain processes [78], while CME is specific to motor processes. It is also possible that PAF and CME are differentially implicated in broader brain processes and not directly involved in pain perception. For example, top-down attentional mechanisms likely mediate the association between PAF and pain [6, 79, 80], while CME adaptations may relate to anxiety/fear of movement responses in pain [81, 82]. Ultimately, a deeper understanding of the association between PAF and CME may require multimodal imaging (e.g., simultaneous EEG and TMS during various pain states), and in-depth assessment of cognitive/emotional factors.

Conclusions

In summary, the present findings provide further support for the reliability of two prospective cortical pain biomarkers. Specifically, it was shown that PAF was reliable over three recording sessions, even in the presence of sustained pain, and even when considering various methodological factors that could influence PAF. Furthermore, it was shown that experimentally induced sustained pain produced corticomotor facilitation or depression in 80% of individuals. This supports the goals of the ongoing PREDICT trial to validate these biomarkers as predictors of pain sensitivity.

Highlights.

• The reliability of two candidate pain biomarkers was assessed: PAF and CME

• PAF demonstrated excellent reliability over several days of sustained pain

• 80% of participants showed facilitation or depression of CME beyond an optimal cut-off point

• The CME response to pain demonstrated good reliability

• The study demonstrates the reliability of two prospective cortical pain biomarkers