Sleep, Positive Affect and Circulating Interleukin-6 in Women with Temporomandibular Joint Disorder

Abstract

Objective:

Systemic inflammation is commonly observed in idiopathic chronic pain conditions, including temporomandibular joint disorder (TMD). Trait positive affect (PA) is associated with lower inflammation in healthy controls, but those effects may be threatened by poor sleep. The associations between PA with proinflammatory cytokine activity and potential moderation by sleep in chronic pain are not known. We thus investigated the association between PA and circulating interleukin-6 (IL-6) and moderation of that association by sleep in a sample of women with TMD and sleep difficulties.

Methods:

Participants (n = 110) completed the insomnia severity index and provided blood samples at 5 intervals throughout an evoked pain testing session. They then completed a 14-day diary assessing sleep and affect, along with wrist actigraphy.

Results:

There was not a significant main effect of PA on resting or pain-evoked IL-6 (b = 0.04, p = 0.33). Diary total sleep time (TST; b = −0.002, p = .008), sleep efficiency (SE; b = −0.01, p = .005), sleep onset latency (SOL; b = 0.006, p = .010) and wake after sleep onset (WASO; b = 0.003, p = .033) interacted with PA to predict IL-6, such that PA inversely predicted IL-6 at higher levels of TST and SE and at lower levels of SOL and WASO. Surprisingly, when sleep was poor, PA predicted greater IL-6.

Conclusions:

The potential salutary effects of PA on resting IL-6 erode when sleep is poor, underscoring the importance of considering sleep in conceptual and intervention models of TMD.

1. Introduction

Temporomandibular disorders (TMD) are highly prevalent idiopathic orofacial chronic pain conditions impacting approximately 10–15% of the US adult population (1) and affecting twice as many women as men (2). Abnormally elevated, low-grade circulating blood levels of proinflammatory cytokines have been observed in TMD (3,4) and in other idiopathic pain conditions (5,6), with the degree of elevation correlating positively with pain sensitivity (5–7) and central sensitization (8), which is hypothesized to contribute to TMD pain (9). Due to its ties with pain sensitivity and chronification, it is important to identify protective factors that modulate systemic inflammation in idiopathic pain conditions, which include TMD.

Trait-like positive affect (PA) is one such putative protective factor, defined as the tendency to experience pleasurable emotions such as joy, enthusiasm, and contentment (10). PA has been shown to be associated with reduced pain among individuals living with chronic pain conditions (11). PA elevations in daily life promote reduced pain intensity (12,13), independent of negative affect (NA), and PA induced in the laboratory leads to reduced experimental pain sensitivity (14,15). An emerging literature also suggests that PA-promoting interventions may have pain-related benefits for individuals living with chronic pain (16,17). Furthermore, previous research has shown an inverse association between trait-like PA and resting peripheral inflammation in healthy samples (18–22). As such, the main effect of PA on IL-6 in individuals living with TMD demands investigation.

Although PA might play a protective role on IL-6 in TMD, the effect may be threatened by poor sleep, which is common in TMD (23,24). In non-pain samples, poor sleep has been shown to attenuate the health-related benefits of positive psychosocial processes (25,26) and exacerbate the maladaptive effects of stress on health and functioning (27–30). Sleep disturbance predicts elevated inflammation in healthy adults (31,32), individuals with sleep disorders (e.g., insomnia and sleep apnea) (33,34), and those with chronic pain (35). Therefore, it is possible that sleep might interact with PA to predict inflammation; however, this has not been empirically examined. The present study thus investigated the direct association between PA and circulating IL-6, and moderation of that association by sleep (measured both subjectively via self-report and behaviorally via wrist actigraphy) in a sample of women with TMD and self-reported sleep difficulties. We hypothesized that higher PA would be associated with lower resting and pain-evoked IL-6, and that poor sleep would attenuate those associations.

2. Method

Hypotheses were evaluated using baseline data drawn from a parent clinical trial project (R01DE019731) investigating changes in pain modulation, pain-evoked inflammatory activity and clinical pain in response to two behavioral interventions that manipulate TMD pain risk factors. The present study was an ancillary analysis utilizing baseline data obtained from randomized participants. The methods pertaining to the present analysis are described here. All methods were approved by the Johns Hopkins University Institutional Review Board. Participants provided informed consent prior to participation. All data were collected prior to randomization in the parent trial, and data collection occurred from May 2015 through March 2018.

2.1. Participants

Participants were included if they satisfied the inclusion and exclusion criteria for the parent trial (see Supplemental Digital Content), which included reporting at least moderate chronic TMD pain (defined as reporting an average pain severity score over the past week of ≥3 on a numerical rating scale ranging from 0 to 10), at least mild levels of trait catastrophizing (defined as a score > 8 on the Pain Catastrophizing Scale) (36) and sleep continuity disturbance (defined as a score ≥ 8 on the Insomnia Severity Scale) (37). In addition, to be included in the present analysis, participants needed to have provided at least one IL-6 measurement (n = 110). Men were excluded in order to remove the potential moderating effects of sex, reduce variance in study outcomes, and because TMD primarily impacts women. Sample demographics are shown in Table 1. Eligibility procedures and inclusion/exclusion criteria are presented in the Supplemental Digital Content.

Table 1.

Sample demographics

Total n	110
Age, mean (SD)	35.6 (11.1)
BMI, mean (SD)	27.9 (5.8)
Female, n (%)	110	100%
Race n (%)
Asian	4	3.6%
Black or African-American	21	19.1%
More than one race	1	0.9%
Other	4	3.6%
White	80	72.7%
Ethnicity n (%)
Hispanic or Latino	5	4.5%
Not Hispanic or Latino	105	95.5%
Highest Degree Completed n (%)
Postgraduate	30	27.3%
High School	28	25.5%
Some high school	1	0.9%
Technical School	2	1.8%
Undergraduate	48	43.6%
Unknown/Not reported	1	0.9%

2.2. Procedures

2.2.1. Pre-screening, screening and baseline visits

Interested individuals completed a pre-screening visit assessing basic eligibility information, demographics, and medical history. If eligible to continue, participants were scheduled for a screening visit approximately 45 days later. The 45-day period was specified so that participants could discuss discontinuation of pain and sleep medications with their provider as required for the parent study. Medical history, height and weight measurements, and a physical TMD examination were completed during an in person visit by a dental hygienist and dentist trained in the TMD Research Diagnostic Criteria Exam (38). Participants diagnosed with TMD then completed the experimental pain testing protocol involving 5 blood draws (see Measures) and a series of questionnaires (including the PANAS-X assessing PA) at least one week after their diagnostic exam. Participants were asked to refrain from using all analgesics, benzodiazepines/ benzodiazepine receptor agonists, or sedating tricyclic antidepressants for at least 24 hours prior to laboratory pain testing. Participants who expressed being acutely ill on the day of testing were rescheduled.

2.2.2. Ambulatory assessments

Following pain testing, participants were asked to continuously wear a triaxial actigraph watch (Actigraph GT3X+ [ActiGraph LLC, Pensacola, FL, USA]) on their non-dominant wrist and answer twice-daily questionnaires via an interactive voice response system (IVR) over the succeeding two weeks (see Measures). Specifically, participants were given a number to call to answer automated questions about sleep, pain, mood, and pain-related cognitions. Participants completed the morning diary immediately after getting out of bed, and the evening diary shortly before bedtime. Participants self-reported on their prior nights’ sleep during the morning assessment, and their current PA and NA levels during the evening assessment. Completion rates for morning and evening diaries were 70% and 69%, respectively. We probed the potentially systematic nature of missing diary data by computing point-biserial correlations between PA, NA and TST-diary missingness (1 = missing, 0 = not missing) with TST assessed via actigraphy. Actigraphy-based TST was not associated with PA missingness (r_pb = −0.02, p = .54), NA missingness (r_pb = −0.01, p = .56), or TST-diary missingness (r_pb = −0.04, p = .12), suggesting that participants’ tendency to complete the daily diary was not associated with their prior night’s sleep duration. Of the nights that participants were instructed to wear the actigraphy watch, 73.3% of them yielded usable data. Unusable data were due to technical device failure, device taken off wrist, or data determined to be invalid based on actigraphy count patterns.

In-home sleep studies with polysomnography were included in procedures. These data will be reported elsewhere; they are not included in the present analyses because they are only available in a small portion of participants and characterize only one night of sleep.

2.3. Measures

2.3.1. Positive and Negative Affect

Positive affect was assessed with 8 items drawn from the Positive and Negative Affect Schedule-X (PANAS-X) positive affect scale (39), and NA with 5 items from the PANAS-X negative affect scale. PA items were selected based on strong factor loadings for the General Positive Affect (determined, enthusiastic, excited, inspired, interested) and Joviality (happy, joyful, cheerful) subscales, and represent high arousal PA. NA items were selected based on strong factor loadings for the General Negative Affect (afraid, scared, jittery, upset, distressed) subscale, and also represent high arousal NA. Participants were asked in the evening IVR diary assessment to rate their current affect (i.e., “Rate how strongly you are feeling each right now”) on Likert-type scales (0 = “not at all”; 10 = “extremely”). PA and NA items were averaged to create PA and NA mean scores for each evening; these means were then averaged across the dairy period to create trait-like PA and NA scores. The construct validity of the PANAS-X PA and NA scales has been well-established in community samples, and items on each scale have shown strong internal consistency (40). Subsets of PA and NA items were chosen in order to minimize response burden to participants on daily diaries. High internal consistency reliability was evident for the PA (Cronbach’s alpha = .94) and NA (Cronbach’s alpha =.80) items in the present sample.

2.3.2. Resting and Pain-Evoked Interleukin-6

Five blood samples were drawn via IV catheter placed in the non-dominant arm according to American Phlebotomy Association guidelines. Samples were drawn before, during and after a quantitative sensory testing (QST) protocol involving thermal pain testing, pressure pain threshold, pain-60 phasic and tonic (41,42) repeated probe ratings, temporal summation induced by both heat and punctate probes, and combined cold water and pressure pain testing (i.e., conditioned pain modulation). The first sample was drawn 30 minutes prior to QST (between 11:00 and 15:45). The second sample was drawn 15 minutes into the QST session, the third immediately upon completion of the QST, the fourth 90 minutes post QST, and the fifth 150 minutes post QST (see Figure 1). Blood samples were placed on ice and centrifuged at 4°C prior to storing at −80°C. Samples were transferred under temperature-controlled methods prior to laboratory analysis by batch assay. IL-6 was assayed from plasma. Two different laboratories were used for blood processing due to closure of the laboratory initially contracted for processing. Commercially available enzyme-linked immunosorbent assay (ELISA) kits were used for IL-6 assay in both laboratories, and each assay was performed in duplicate. Duplicate assays that demonstrated a coefficient of variation greater than 15% were re-analyzed. The initial laboratory processed the majority of participants’ samples (n = 97; Human IL-6 Quantikine ELISA Kit D6050; intra-assay CV = 12%; inter-assay CV = 3%), and the remaining participants’ samples (n =22; Meso Scale Discovery V Plex Custom Kit Cat # K151A9H-2; intra-assay CV = 6%; inter-assay CV = 10%) were processed in a second laboratory. We adjusted for processing site in statistical models (see section 2.3.4).

Figure 1.

Blood draw timing in relation to QST.

2.3.3. Sleep

Diary total sleep time (TST), sleep onset latency (SOL), and wake after sleep onset (WASO), and sleep efficiency (SE) were derived from individual self-reports on a standard sleep diary (43). Actigraphy indices of sleep continuity (TST, WASO, SOL, and SE) were scored according to guidelines suggested by the Society of Behavioral Sleep Medicine (44). Each participant’s daily sleep periods were first autoscored using the ActiLife software which employs the Cole-Kripke algorithm. The autoscored sleep periods were manually verified by trained study personnel using the daily sleep diary data, when available. Insomnia severity was measured with the Insomnia Severity Index (ISI), a self-report measure of insomnia severity demonstrating good psychometric properties in individuals with sleep complaints (37). Internal consistency reliability of the ISI items in this sample was adequate (Cronbach’s alpha = .79).

2.3.4. Covariates

BMI was calculated using height and weight. Five participants had missing BMI data, and their values were populated with their self-reported BMI obtained at the phone screening. Self-reported race was trichotomized as Black (the reference group), White, and Other (consisting of Asian, more than one race, other race, or race unknown/not reported), as heightened systemic inflammation has been observed in Blacks relative to Whites (45). We observed differences in mean IL-6 levels between the two laboratories. Thus, we adjusted for blood data processing site (dichotomized as 0 and 1) in all models. All models adjusted for NA to account for its overlap with PA (46); specifically, in an identical manner to PA, we entered NA as an interaction term with each sleep metric. Time-of-day (diurnal) and circadian rhythm effects (47) were included as covariates. The time of blood draw 1 was converted to minutes from midnight and included as a covariate adjusting for diurnal effects. Individuals’ average sleep midpoint was included as a covariate adjusting for circadian rhythm effects.

2.3.5. Statistical Approach

Resting IL-6

The effects of sleep and PA on resting IL-6 were tested using aggregated multiple linear regression models. PA, sleep measures and their interactions were included as predictors. Significant interactions were probed by visualizing simple slopes. For each model, we report the model adjusted R². For significant sleep × PA interaction effects, we report the percentage change in R² with the addition of the PA × sleep interaction term in question, relative to a model with PA entered as a main effect only. Alpha was set at .05. To control the false discovery rate, we applied the Benjamini-Hochberg correction procedure to hypothesis tests (i.e., PA × Sleep interaction terms) separately for self-reported sleep indices an actigraphy-based indices, and we report the adjusted p-values in Results. Finally, we probed significant interaction effects by visualizing simple slopes. We supplemented visualizations with statistical significance testing of simple slopes at +/1 SD around the mean, and by quantifying regions of significance with the Johnson-Neyman method (48). Regions of significance obtained through the Johnson-Neyman approach are dependent on sample size; and, this approach is one form of multiple inference testing. As such, consistent with recommendations (48), we focus on the magnitude of the simple slope estimates in conjunction with p-values.

Pain-evoked IL-6

The effects on pain-evoked IL-6 were tested using linear mixed effects models in accordance with recommended procedures in longitudinal analysis (49). In visualizing individual trajectories in IL-6, we observed what appeared to be a “broken-arrow” trajectory in IL-6 over time, with a relatively flatter slope observed from Sample 1 to Sample 3, and a relatively steeper slope from Sample 3 to Sample 5 (see Figure 2). We thus compared the fit indices for unconditional linear, quadratic and linear spline models, allowing for random slopes and intercepts. Linear spline modeling is one way of analyzing broken-arrow time trends in a multilevel analysis framework (50), which allows for slopes to differ before and after a specified event (50). Spline and quadratic models allowing for random slopes both failed to converge, suggestive of poor fit. We thus estimated spline and quadratic models using fixed slopes, and compared them with a linear model allowing for random slopes. The linear model allowing for random slopes fit the data best compared with spline and quadratic models estimating fixed slopes (AIC = 1006.7 _[linear] vs. 1049.2 _[spline]) vs. 1055.3 _[quadratic]) (51).

Figure 2.

Illustration of the change in IL-6 in response to quantitative sensory testing. Panel A shows the individual trajectories in IL-6 with group means superimposed. Panel B shows the group level IL-6 average and standard deviation at each measurement.

In linear growth models, PA, sleep metrics and their interactions with Sample (i.e., samples 1 – 5) were included as predictors. Conceptually, the linear growth curve models test the extent to which the pain-evoked trajectories in IL-6, shown in Figure 2A, are moderated by (i.e., differentiated based upon) PA and sleep characteristics. Given that hypotheses were tested with interaction terms involving a level 2 variable (PA) and level 1 variables (Sample), to adequately control for NA, we similarly entered it as an interaction term with Sample. Models were specified using the lme4 and lmerTest packages in R (52,53) using restricted maximum likelihood estimation. Significance values were computed using Satterthwaite’s method (52). We report the marginal R² (a pseudo R² reflecting the proportion of outcome variance explained by the fixed effects), as well as the conditional R², which indicates the amount explained by both fixed and random effects (54). We estimated the magnitude (i.e. effect size) of significant interaction effects by examining the percentage change in the marginal R² with the addition of the three-way interaction term (Sample × PA × sleep), relative to a model with two-way interaction terms only. We probed significant interactions by visualizing simple slopes. Lastly, in order to characterize the clinical importance of IL-6 in this sample, we correlated baseline IL-6 with pain severity as assessed with the Pain Severity subscale of the Brief Pain Inventory (55,56), which participants completed at baseline (see 2.2.1). To control the false discovery rate, we applied the Benjamini-Hochberg correction procedure to hypothesis tests (i.e., PA × Sleep × Sample interaction terms) separately for self-reported sleep indices and actigraphy-based indices, and we report the adjusted p-values in Results. Finally, we probed significant interaction effects by visualizing simple slopes. We supplemented visualizations with statistical significance testing of simple slopes at +/1 SD around the mean, and by quantifying regions of significance with the Johnson-Neyman method (48).

3. Results

3.1. Descriptive Statistics and Data Preparation

Supporting the clinical relevance of inflammation in TMD, resting IL-6 was positively correlated with clinical pain severity (r = .28, p = .01). PA and NA were not significantly correlated (r = .07, p > .05). Resting IL-6 was positively correlated with BMI (r = .48, p < .001) and with race, such that Whites demonstrated lower IL-6 compared with Blacks and those classified as Other race (r = −.30, p < .01). Correlations among diary and actigraphy-based sleep indices were generally moderate to large. All bivariate correlations between resting IL-6, sleep metrics and affect are shown in the Supplemental Digital Content (Table S1).

The distribution of resting IL-6 scores (i.e., Sample 1) was non-normal (skew = 2.09, kurtosis = 5.07), and was thus natural log transformed. Two outliers were detected using the +/− 1.5 IQR criterion (57) and were removed. The final log-transformed distribution of IL-6 scores approximated normality (kurtosis = −0.30; skewness = −0.22). The distribution of pain-evoked IL-6 scores was also non-normal, demonstrating kurtosis equal to 13.9 and skewness equal to 2.7. As such, pain-evoked IL-6 scores were log transformed. Five outliers were detected in the log-transformed scores using the +/− 1.5 IQR criterion (57) and were removed. The final distribution of pain-evoked IL-6 scores approximated normality (kurtosis = −0.19; skewness = −0.13).

3.2. Resting IL-6

3.2.1. PA and Sleep Main Effects

The main effect of PA on resting IL-6 was not significant (b = 0.04, t(83) = 0.68, p = 0.33; model adjusted R² = .34). There were no significant main effects of diary-based TST (b = −0.0003, t(87) = −0.25, p = .80), SE (b = −0.008, t(87) = −1.07, p = 0.29), WASO (b = 0.002, t(87) = 0.88, p = 0.37) or insomnia severity (b = 0.006, t(87) = 0.32, p = 0.74) on IL-6. Longer diary-based SOL predicted higher IL-6 (b = 0.009, t(87) = 2.56, p = .012). IL-6 was not associated with actigraphy-based TST (b = −0.001, t(75) = −0.58, p = .56), SOL (b = 0.06, t(75) = 1.47, p = .15), WASO (b = 0.006, t(75) = 1.76, p = .083) or SE (b = −0.02, t(75) = −1.71, p = .091), although the latter two effects were marginal.

3.2.2. PA × Sleep Interactions

Diary and Self-Report.

There was a significant diary TST × PA interaction on resting IL-6 (b = −0.002, t(80) = −3.04, p = .008; model adjusted R^{2 =} .41, change in R² = 17%), such that the slope of the association between PA and IL-6 was negative and significant at higher levels of TST (i.e., between 481.7 and 550 minutes; at +1 SD, b = −.14, p = .067), and positive and significant at lower levels (i.e., between 144.4 and 342.2 minutes; at 1 SD above the mean, b = .13, p = .017) of TST (see Figure 3A). There was a SOL × PA interaction on resting IL-6 (b = 0.006, t(80)= 2.39, p = .010; model adjusted R^{2 =} .44, change in R² = 14%), such that the slope of the association between PA and IL-6 was negative but not significant among individuals with shorter SOL (i.e., outside the Johnson-Neyman range of significance, at 1 SD below the mean, b = −.10, p = .155) and positive and significant among individuals with longer SOL (i.e., between 42.9 and 121 minutes; at 1 SD above the mean, b = .19, p = .005); see Figure 3B. Additionally, There was a SE × PA interaction on resting IL-6 (b = −0.01, t(80) = −3.33, p = .005; model adjusted R^{2 =} .42, change in R² = 20%), such that such that the slope of the association between PA and IL-6 was negative and significant at higher levels of SE (i.e., between 93.8 and 95.1 minutes at + 1 SD, b = −.13, p = .062), and positive and significant at lower levels of SE (i.e., between 38.5 and 72.3 minutes; at −1 SD, b = .15, p = .007); see Figure 3C. Finally, the PA × WASO interaction was significant (b = 0.003, t(80) = 2.26, p = .033; model adjusted R² = .38, change in R² = 9%), such that the slope of the association between PA and IL-6 was negative but not significant at lower levels of WASO (i.e., outside the Johnson-Neyman range of significance, and 1 SD below the mean, b = −.08, p = .278) and positive and significant at higher levels of WASO (i.e., between 69.6 and 186.4 minutes; at +1 SD, b = .10, p = .056); see Figure 3D. The PA × ISI interaction (b = 0.02, t(80) = 1.98, p = .050; model adjusted R^{2 =} .36) was marginally significant and showed a similar trend to the diary indices, such that the slope of the association between PA and IL-6 was positive at higher levels of insomnia severity, and negative at lower levels of insomnia severity. Table 2 displays full statistical results for resting IL-6 models involving diary-based sleep indices.

Figure 3.

Visualization of self-reported sleep by positive affect interactions at +/− 1 SD around the mean. As shown, individuals reporting higher positive affect and better self-reported sleep demonstrated the most favorable resting IL-6 profiles, while those reporting both high positive affect and poor sleep demonstrated the highest IL-6 levels.

Table 2:

Self-reported sleep indices and their interaction with PA on resting IL-6

	TST		Insomnia Severity		SOL		SE		WASO
Predictors	Estimates	p	Estimates	p	Estimates	p	Estimates	p	Estimates	p
Intercept	−4.91	0.01	−2.77	0.16	−3.24	0.042	−5.97	0.007	−3.84	0.022
BMI (kg/m2)	0.05	0.002	0.05	0.002	0.05	0.002	0.05	0.002	0.05	0.001
Positive Affect	0.67	0.002	−0.30	0.10	−0.16	0.08	0.90	0.001	−0.08	0.26
Total Sleep Time (min)	0.00	0.24
Negative Affect	−0.64	0.021	0.10	0.59	0.06	0.51	−0.55	0.10	0.03	0.74
Processing Method (1 vs. 0)	−0.56	0.01	−0.55	0.016	−0.49	0.032	−0.50	0.022	−0.57	0.012
Other Race (v. Black)	−0.34	0.37	−0.37	0.34	−0.22	0.55	−0.27	0.479	−0.35	0.36
White Race (v. Black)	−0.72	0.003	−0.72	0.004	−0.70	0.003	−0.66	0.007	−0.67	0.007
Draw Time (min from midnight)	0.00	0.018	0.00	0.029	0.00	0.034	0.01	0.013	0.00	0.017
Sleep Midpoint (min from midnight)	0.00	0.43	0.00	0.47	0.00	0.51	0.00	0.466	0.00	0.34
PA by Total Sleep Time	0.00	0.008
NA by Total Sleep Time	0.00	0.04
Insomnia Severity			−0.05	0.31
PA × Insomnia Severity			0.02	0.05
NA × Insomnia Severity			−0.01	0.33
Sleep Onset Latency (min)					−0.01	0.47
PA by Sleep Onset Latency					0.01	0.010
NA by Sleep Onset Latency					0.00	0.11
Sleep Efficiency (%)							0.03	0.16
PA by Sleep Efficiency							−0.01	0.005
NA by Sleep Efficiency							0.01	0.16
WASO (min)									0.00	0.55
PA by WASO									0.00	0.033
NA by WASO									0.00	0.15
Observations	92		92		92		92		92
R2 / R2 adjusted	0.484 / 0.413		0.440 / 0.363		0.504 / 0.436		0.493 / 0.423		0.459 / 0.384

Note. PA = Positive Affect; NA = Negative Affect; TST = Total Sleep Time; SOL = Sleep Onset Latency; SE = Sleep Efficiency; WASO = Wake after Sleep Onset; min = minutes;

Processing method refers to the two distinct laboratories used for blood processing.

Actigraphy.

There were no significant interactions between the actigraphy-based indices (TST, WASO, SE and SOL) and PA (p’s > .05; full statistics available in the Supplemental Digital Content, Table S2).

3.3. Pain-evoked IL-6

In a model without other variables, Sample positively predicted IL-6 (b = .33, t(95.8) = 12.5, p < .001), indicating that IL-6 increased on average over the QST protocol. The majority of the two and three-way interactions between Sample, PA and sleep parameters were not significant. However, there was a significant three-way interaction in actigraphy-based TST (b = .0008, t(70.1) = 2.65, p = .010), which accounted for a negligible amount of explained variance (marginal R² change = 3%). Inspection of the slopes demonstrated that participants who were high in TST and PA had the lowest resting IL-6 values, and subsequently exhibited a steeper increase in IL-6 from Sample 1 through Sample 5. Full results regarding pain-evoked IL-6 are detailed in the Supplemental Digital Content in Figure S1 and Tables S3 to S11.

4. Discussion

The principal finding of this study was that self-reported sleep modulated the association between trait-like PA and resting IL-6 in women with TMD and insomnia symptoms. In support of our hypothesis, those reporting higher PA and better self-reported sleep (i.e. longer TST, greater SE, shorter SOL and lesser WASO) demonstrated the most favorable IL-6 profiles. Relative to a PA main effect model, the addition of sleep × PA interaction terms led to notable increases in total variance explained (i.e., 14–20%), supporting the potential clinical relevance of sleep as a moderator. These findings are consistent with prior work in TMD identifying trait optimism, another resilience factor, as protective on resting and stress-induced IL-6 (58). They also extend other work in non-pain samples showing lower resting IL-6 among individuals with better sleep and more positive social relationships (59,60), and weaker relations between poor sleep and systemic inflammation among individuals who felt socially integrated as opposed to isolated (61). Central nervous system proinflammatory cytokine activity has been linked with central sensitization (8), which is an enhanced functioning of nociceptive neurons and circuits that amplifies sensory responses elicited by normal inputs, including those that would typically produce non-painful sensations (62). Central sensitization likely plays a role in instigating or perpetuating pain in idiopathic pain conditions (63), including TMD (9). Indeed, basal IL-6 and clinical pain were positively correlated in this sample, supporting the clinical relevance of these findings.

These findings underscore the importance of considering sleep and PA in conceptual and intervention models of TMD. This is particularly important given that elevated systemic inflammation in TMD and other idiopathic pain conditions is associated with increased pain (5–7) and has been identified as a driver of central sensitization (8). Contextualized in the extant literature, the present results hint that poor sleep could potentially contribute to increased inflammation in chronic pain not only through directly dysregulating inflammatory pathways (64), but also by eroding the protective effects of PA. The present study is the first of its kind to document a degradation in the inflammatory protection by PA under conditions of poor sleep, further supporting the importance of improving sleep in the context of idiopathic pain. This could mean that patients who are provided with interventions intended to increase PA may not be as responsive if their sleep is concurrently disturbed. Additional replication studies and longitudinal investigations on the interactions between sleep and PA on inflammatory processes in idiopathic pain are warranted, including studies in the context of clinical trials. Notably, our sample of TMD patients was selected to have at least mild insomnia symptoms. Thus, longer sleep duration and lower insomnia severity within this sample do not necessarily reflect normative or healthy sleep characteristics. This may be a minor concern, given the ubiquity of poor sleep in chronic pain conditions (65).

We can only speculate on the psychobiological underpinnings of the association between combined low PA and disturbed sleep with systemic inflammation. Cautiously, one of many possibilities is that the increased sympathetic outflow characteristic of poor sleep (66) and observed in patients with TMD (67) counteracts PA’s dampening of sympathetic nervous system activity (68), enhancement of parasympathetic activity (69), and restorative influence (70). Sympathetic and parasympathetic components are well known contributors to elevated and lowered peripheral inflammation, respectively (71).

Surprisingly, we also observed that those individuals reporting both high PA and severely disturbed sleep (i.e. shorter TST, longer SOL and WASO, reduced SE) demonstrated the highest levels of circulating IL-6. Although preliminary, these findings suggest that PA could be maladaptive in the context of chronically, severely disturbed sleep, consistent with emerging research suggesting that PA is not beneficial at every level and in every context [44]. According to functional theories of affect, emotions are adaptive, orchestrating behavioral responses to various sets of circumstances [58]. PA functions to promote approach-oriented behavior and a building of personal resources [23,35]; interpersonally, it signals to others that all is well [44]. These sorts of behaviors are likely adaptive when surrounding environmental conditions are safe, personal resources are available to support engagement in novel endeavors [77] and social support is not essential. In contrast, negative emotions arise when resources must be devoted to handling difficulties; for instance, sadness motivates people to conserve resources and energy, and to call for help [30,58]. Conceptualizing chronically, severely disturbed sleep as a context characterized by impaired functioning and depleted personal resources [5,98], PA may be maladaptive to the extent that it prevents engagement in restorative behaviors and the communication of social support needs to others. Empirically, it appears to be a normative response that poor sleep leads to a reduction in next-day positive affective functioning [33,82]. Future research should test the preliminary possibility that PA is not protective, and perhaps even maladaptive, under conditions of severely disturbed sleep in individuals with idiopathic pain. It is important to emphasize however that we make this interpretation with caution, given the paucity of literature on this topic in the context of pain.

Although previous work has shown an association between trait-like PA and resting peripheral inflammation in healthy samples (18–22) this relationship is inconsistent (72–74), and we did not observe a main effect of PA in our sample of female TMD patients with mild insomnia. Various factors associate with resting IL-6 in individuals with chronic pain, including depression severity (75), optimism (58) and poor sleep (35). It may be that PA does not rank among the factors significantly predicting basal IL-6 in TMD. Of note, it is difficult to identify systematic factors that explain the inconsistencies in the non-pain literature regarding the PA-inflammation relationship. In terms of sample characteristics, most of these studies have been done on large, community samples of young and mid to late life adults. Future research should systematically investigate whether PA arousal level can explain these inconsistencies, as most reporting significant relations between PA and IL-6 (18,19,21,22) have used the PANAS-PA subscale to index affect (as was done in the present study), while those reporting null associations used other PA metrics for which it is difficult to define the PA arousal level (e.g., single items assessing happiness (73), the anhedonia subscale of the CES-D (72)).

Since IL-6 responds to experimental pain as a stressor (86–89), we also hypothesized that the inflammatory response to the stress of laboratory pain exposure would be modulated by PA and sleep. We did not find that PA, sleep or their interaction modulated the IL-6 response to laboratory pain. Since previous work has shown that psychosocial factors (i.e., situational and trait-like pain catastrophizing (89,90)) do modulate pain-evoked IL-6 changes in chronic pain samples, PA may be specifically irrelevant to pain-evoked inflammation, and not psychosocial factors in general.

Although the current findings suggest that PA and self-reported sleep interact to predict circulating IL-6, actigraphy-based measures of sleep did not modify associations with PA. Prior studies have also demonstrated discrepancies between subjective and behaviorally-assessed sleep measures (91–93), and differential prediction of outcomes (e.g., fatigue) by method of assessment (93). More broadly, there is a precedence in the literature for relatively weak correlations between self-report and behavioral measures (94). It could be that a mono-method bias partially accounts for the significant interaction between self-reported sleep and PA. Additionally or alternatively, it is possible that in the course of daily life, the perception of sleep continuity disturbances is what matters and has the capacity to erode the protective effects of PA, whereas behavioral indices are less relevant. Indeed, insomnia is characterized by subjective complaints about sleep and daytime symptoms, and actigraphy-based measurements are less valid diagnostic indicators (i.e., they do not always accurately capture patients’ experiences, as they motionless wakefulness as sleep (95)). Future experimental sleep work that manipulates sleep perceptions would be necessary to evaluate the relative importance of subjective versus behavioral sleep indices in this context.

It is important to note that sleep and PA were measured during the two-week period following the inflammation assessments and pain testing procedure; as such, we cannot claim that sleep and PA caused changes in inflammation. Due to our observational design, it also cannot be claimed that inflammation caused changes in sleep and PA; however, this possibility deserves consideration. For example, research participants infected with influenza (96) demonstrated augmented proinflammatory cytokine activity, which in turn predicted next-day reductions in PA. Experimentally-induced inflammation produced altered patterns of brain activation and connectivity tied with depressive symptoms, along with reductions in global mood scores (97). As such, the relationship between PA and inflammation might be bi-directional. Regarding sleep, numerous cytokines are involved in regulating sleep-wake behavior, as discussed in a comprehensive review (98). For instance, injection of IL-6 enhances slow wave sleep and reduces rapid eye movement (REM) sleep (99). Future clinical research examining the impact of augmented PA (via psychological intervention) on inflammation, and moderation by sleep, is needed to address questions of causality.

This study has limitations that need to be considered when interpreting the moderating effects of sleep disturbance on the association between PA and IL-6. First, the extent to which these findings generalize to males is unknown. Since our group has previously shown that women with knee osteoarthritis pain show larger pain-evoked IL-6 responses relative to men (88), sex differences in these IL-6 responses deserve further investigation. Furthermore, we used items assessing high-arousal positive affect, and we therefore cannot speak to the importance of low-arousal PA (e.g., calm, serenity) in this context. A strength of our methodology is the use of daily diaries to measure sleep and affect in the context of daily life. Since no biological variables were measured during this 2-week monitoring period, scores were aggregated over days and used to predict IL-6 levels on a single day of testing. Although the interaction between PA and sleep emerged as significant and clinically relevant, it was less so in comparison with other physical characteristics, namely body mass index, which robustly associated with IL-6 in analyses. This was not surprising given that adipose tissue, the total amount of which correlates with BMI (100), secretes IL-6 (101).

Despite these limitations, we find a novel moderation effect of sleep on the association between PA and IL-6 in women with TMD. This finding suggests extending conceptual and intervention model development in TMD to include positive affect as well as sleep, the latter of which has a strong empirical base (23). In addition to including both behavioral and subjective sleep measures, this investigation used aggregated daily diary assessments of affect, which may provide a more valid index of trait-like affect than commonly used recall measures (102). We encourage replication and extension of these findings through further investigation of the relations between PA, sleep and inflammation in TMD and other idiopathic chronic pain conditions. Relatedly, given that sleep has been shown to degrade next-day PA in laboratory and naturalistic settings (84,85), future researchers might consider testing a mediation model (i.e., disturbed sleep leads to reduced PA, in turn promoting heightened inflammation) using longitudinal data.

List of abbreviations

TST

Total Sleep Time

SOL

Sleep Onset Latency

SE

Sleep Efficiency

WASO

Wake after Sleep Onset

TMD

Temporomandibular disorder

US

United States

PANAS-X

Positive and Negative Affect Schedule-X

QST

Quantitative Sensory Testing

ELISA

enzyme-linked immunosorbent assay

PA

Positive Affect

NA

Negative Affect