Individual differences in pain sensitivity are associated with cognitive network functional connectivity following one night of experimental sleep disruption

2.1 Participants

SMT fMRI data were collected from 22 healthy individuals after uninterrupted and experimentally disrupted sleep. Participants were eligible if they met the following criteria: (a) met Research Diagnostic Criteria for Normal Sleepers, (b) scored <5 on the Pittsburgh Sleep Quality Index and <10 on the Epworth Sleepiness scale, (c) had average total sleep time between 6.5 and 8.5 hr/night and sleep efficiency >85% as reported on 2 weeks of sleep diaries and captured via actigraphy monitoring, (d) showed a stable sleep phase within 21:00 and 10:00, (e) polysomnography-confirmed apnea-hypopnea index <10, (f) denied a lifetime history of pain persisting >6 months or acute pain as measured on the McGill Pain Questionnaire and baseline sleep diaries, (g) reported no significant medical or psychiatric morbidity within 6 months of study participation and obtained T-scores <64 on the Brief Symptom Inventory global scales, (h) no lifetime history of bipolar disorder, psychotic disorder, recurrent major depression, posttraumatic stress disorder, substance abuse, reported history of traumatic brain injury, or seizures, (i) reported no tobacco/nicotine use and low caffeine use (<2 cups per day), (j) had a BMI <35, and (k) no MRI contraindications.

The present analyses were completed on data from 19 individuals (14 women; mean age = 23.95 years (SD = 4.17)) after two participants were excluded for incomplete fMRI data, and one individual was found to be an outlier with influential, extreme values for pain ratings. Seven participants identified as White Non-Hispanic, two individuals identified as White Hispanic, five participants identified as Black Non-Hispanic, one individual identified as Black Hispanic, two participants identified as mixed race, and one individual identified as Other Hispanic. Participant demographics and psychosocial function are listed in Table 1.

2.2 Experimental design and statistical analyses

2.2.3 Pain sensitivity change score

The present study aimed to examine changes in FC among cognitive networks as it related to sleep loss-induced hyperalgesia. In order to examine this aim, we computed a pain sensitivity change score as a predictor variable. For the parent study, participants underwent a quantitative sensory testing session before scanning at both study visits to derive participant-tailored temperatures for the fMRI noxious stimulation task. Temperatures rated between 5 and 8 out of 10 were selected. During fMRI scanning, these tailored temperatures were presented concomitant with music of neutral valence. Pain ratings were made following the delivery of each noxious thermal stimulus, which ended at the same time the music clip ended. The same procedure was employed at each study visit (see Supporting Information for additional details). Due to the aims of the parent study, music stimuli were paired with each noxious thermal stimulus. The parent study also overlaid positive-valence song clips on noxious thermal stimuli in separate blocks, but pain ratings associated with those blocks were not used in the present study because they were associated with pain inhibition (Seminowicz et al., 2019).

A pain sensitivity change score was calculated using pain ratings and tailored temperatures. Temperatures ranged from 40°C to 50°C, so we subtracted 40 from each temperature denominator to make the scales of pain ratings and temperatures equal (0–10). Our approach of subtracting 40 from temperature values was intended for scaling purposes only and does not assume an association between the temperature value and pain rating value.

Some individuals experienced changes in both temperature and pain ratings (e.g., lowered temperatures with higher pain ratings after FA sleep would indicate increased pain sensitivity). As such, we calculated a pain sensitivity change score using the following equation, with higher values indicating greater pain sensitivity following FA:

2.2.5 Neuroimaging data processing

Functional MRI data preprocessing

SMT data were preprocessed in SPM12 v6906. The preprocessing pipeline included realignment (motion correction), coregistration of the T1 anatomical to the mean functional image, segmentation of the anatomical to six standard tissue probability maps in SPM12, normalization of functional images using forward deformation fields from anatomical segmentation, and spatial smoothing using a 6 mm full-width at half maximum (FWHM) kernel. Data were visually inspected for quality control to determine motion artifact. Further, the Artifact Detection Tool (ART; www.nitrc.org/projects/artifact_detect/) was also used to mark potentially confounding outliers in the SMT time series for exclusion (i.e., rotation >.02 rad from prior volume, or translation >.4 mm from the prior volume (Chai, Ofen, Gabrieli, & Whitfield-Gabrieli, 2014). No condition-based differences in outlier detection (p = .4) or motion correction (p = .38) were identified.

We used the CONN toolbox (Whitfield-Gabrieli & Nieto-Castanon, 2012) to complete denoising procedures and FC analyses. Gray matter, white matter, and CSF masks were created via anatomical segmentation and were eroded by one voxel to minimize partial volume effects (Whitfield-Gabrieli & Nieto-Castanon, 2012). Physiological confounds, including respiratory and cardiac activity, were accounted for by using the white matter and CSF masks as nuisance regressors during denoising (Behzadi, Restom, Liau, & Liu, 2007). Denoising including scrubbing of excessive motion and outliers detected via ART and regression of white matter and CSF masks (% volumes removed: mean = 1.9%, range = 0.53–5.34%). A stepwise, rather than simultaneous, regression approach was applied to maintain adequate degrees of freedom of the resting-state data (Patel & Bullmore, 2016). Data were bandpass filtered between .008 and .09 (Aurich, Alves Filho, Marques da Silva, & Franco, 2015).

Hybrid independent component analysis-seed-based functional connectivity

We conducted hybrid independent component analysis (ICA)-seed-based FC analyses on denoised fMRI data in the CONN toolbox. Previous work suggests that hybrid ICA-seed-based FC analyses result in better reproducibility of fMRI results than either ICA or seed-based analyses alone (Kelly et al., 2010). First, group ICA is a data-driven technique that identifies orthogonal sources of variance within the data (Calhoun, Adali, Pearlson, & Pekar, 2001), yielding independent components (ICs). Each IC contains brain regions with activity co-occurring across a unique time course, thought to represent a neural network (Calhoun, Adalı, & Pekar, 2004).

Consistent with steps described by Calhoun et al. (Calhoun et al., 2001), denoised data first underwent participant- and condition-level BOLD signal concatenation. We estimated the optimum number of ICs within the GIFT toolbox using minimum description length criterion (Li, Adali, & Calhoun, 2007), yielding 20 potential ICs. Dimensionality reduction was then conducted at the group-level to identify these 20 dimensions/ICs. To estimate IC maps at the group-level, the fastICA algorithm was used. Group-level IC maps are then back reconstructed on individual-level data via GICA1 backprojection to create individual-level IC spatial maps that can be pooled for second-level GLM statistics; however, only the group-level ICs were used to create seeds for subsequent analyses. Resultant group-level ICs were sorted to freely available spatial templates of RECN, LECN, and the DMN (Shirer, Ryali, Rykhlevskaia, Menon, & Greicius, 2012). The IC with the strongest match to each template (based on the Dice coefficient) was converted into a binarized map used in subsequent seed-based analyses (Table 2; Figure 2).

Table 2. Clusters identified within each independent component that was used as a seed for seed-to-ROI functional connectivity analyses. Independent components represented the right executive control network (RECN), left executive control network (LECN), and default mode network (DMN)

Independent component	Cluster hemisphere	Cluster coordinates	Cluster size	p-value (<.05 FWE)	Region
RECN	Right	34 18 54	5,746	<.001	dlPFC/MFG
	Right	44 −44 54	3,287	<.001	LOC
	Left	−34 −56 52	1,726	<.001	SPL
	Right	64 −32 −14	810	<.001	MTG
	Left	−30 −62 −32	681	<.001	Cerebellum
	Right	6 −26 32	544	<.001	PCC
	Left	−42 50 −10	333	<.001	Frontal pole
	Left	−10 −78 −26	315	<.001	Cerebellum
	Left	−64 −44 −22	309	<.001	ITG
	Right	32 33 −6	243	<.001	Insula cortex
	Right	10 −58 46	233	<.001	Precuneus
	Left	−52 16 40	53	<.001	MFG
	Left	−34 4 30	45	<.001	Ventral premotor
	Left	−38 −28 8	19	.005	Planum temporale
LECN	Left	−46 12 36	7,986	<.001	MFG
	Left	−34 −60 52	3,536	<.001	LOC
	Left	−56 −50 −10	1,461	<.001	ITG
	Right	30 −62 46	284	<.001	LOC
	Right	44 34 8	230	<.001	IFG
	Right	12 −76 −24	157	<.001	Cerebellum
	Right	28 −70 −46	143	<.001	Cerebellum
	Right	28 −62 −30	136	<.001	Cerebellum
	Right	4 8 28	75	<.001	MCC
	Left	−6 −34 42	46	<.001	PCC
	Right	54 −48 −18	41	<.001	ITG
	Right	50 −12 60	31	<.001	Precentral gyrus
	Left	−10 −20 72	25	<.001	Precentral gyrus
DMN	Midline	6 −46 20	8,750	<.001	Precuneus/PCC
	Left	−44 −62 32	2,529	<.001	LOC
	Right	40 −64 44	1,941	<.001	LOC
	Midline	0 52 −12	1,872	<.001	vMPFC/ACC
	Left	−22 30 42	436	<.001	SFG
	Right	22 32 38	357	<.001	SFG
	Right	34 −80 −6	204	<.001	LOC
	Left	−58 −22 −18	69	<.001	MTG
	Left	−8 50 −42	54	<.001	Cerebellum
	Right	54 −10 −18	34	<.001	MTG
	Left	−24 −88 −8	23	.003	Fusiform gyrus
	Left	−18 −96 4	20	.005	Occipital pole
	Right	10 −46 −46	19	.01	Cerebellum

• Abbreviations: ACC, anterior cingulate cortex; IFG, inferior frontal gyrus; ITG, inferior temporal gyrus; LOC, lateral occipital cortex; MFG, middle frontal gyrus; MTG, middle temporal gyrus; OFC, orbitofrontal cortex; PCC, posterior cingulate cortex; SFG, superior frontal gyrus; SPL, superior parietal lobule.

Binarized IC masks were added to the CONN toolbox as a priori regions of interest (ROIs). Further, the Power 264-region atlas was used to define 10 mm ROI spheres across the whole brain (Power et al., 2011). This whole-brain atlas describes subgraphs of functionally integrated nodes as defined by graph theory analyses on resting-state fMRI data, rather than structurally defined. First-level analyses were completed in CONN to extract time series data within each ROI and compute Fisher's r-to-Z transformed bivariate correlations among these ROIs.

GLM analyses

Contrast images were entered into a second-level seed-to-ROI analysis to determine FC differences between US and FA across the sample. The pain sensitivity change score described above was inserted as a regressor of interest, and RECN, LECN, and DMN masks were defined as seeds. In total, there were six primary sets of GLM tests, including three seed-to-ROI tests (i.e., RECN, LECN, and DMN as seeds) and three tests examining associations among RECN, LECN, and DMN masks (i.e., RECN-LECN, LECN-DMN, RECN-DMN). ROIs in the seed-to-ROI analyses included 264 spheres from the Power 264 atlas. A conservative analysis-level threshold across seeds was applied as a correction for multiple comparisons (pFDR < .05). We used a two-tailed test given the nature of our competing directional hypotheses.

3.1 Sleep manipulation check

To examine the efficacy of our sleep disruption manipulation we examined condition-based differences in several sleep parameters. As expected, there was a significant decrease in minutes of TST, stage N2, SWS, and REM sleep phases, and a significant increase in stage N1 sleep. Participants reported significantly greater sleepiness during the FA compared to US condition. Table 3 provides means, SD values, and paired-samples t test results for these variables. Combined, these results indicate the efficacy of the experimental manipulation.

3.2 Pain ratings, thermal stimuli temperatures, and SMT performance

Table 3 lists statistical information about condition-based differences in pain ratings, thermal stimuli temperatures, and SMT performance. Neither pain ratings nor thermal stimuli temperatures significantly differed between the US and FA conditions. On the SMT, participants’ accuracy in identifying arrow directions was better during US compared to FA (74.8% and 63.3% accuracy, respectively). The low accuracy scores were due to missed trials or incorrect responses.

3.3 Left and right ECN and DMN functional connectivity

3.3.1 RECN as a seed

Several ROIs from the Power atlas demonstrated significant FC changes with the RECN seed associated with pain sensitivity change scores (omnibus model: F_(4,14) = 25.28, pFDR = .05; individual ROI statistics reported in Table 4; Figure 4). Specifically, there was a positive association between increased pain sensitivity and increased RECN FC to left dlPFC (part of the LECN), left medial and inferior frontal gyri (MFG, IFG, respectively; part of the DMN), and the right paracingulate gyrus (part of the DMN). There was also a positive association between increased pain sensitivity and decreased RECN FC to the right lateral occipital cortex (LOC; part of the DMN). Table 4 provides additional information about these ROIs, and Table 5 provides descriptive statistics for FC values based on condition.

Table 4. Significant power atlas ROIs yielded from seed-to-ROI analyses examining differences in cognitive network FC between the uninterrupted sleep and forced awakenings sleep conditions associated with pain sensitivity change scores

Seed	Power 264 atlas ROI	Associated network	Region	Coordinates			t-stat	pFDR
				x	y	z
RECN	188	LECN	l-dlPFC	−42	38	21	4.21	.03
	201	LECN	l-dlPFC	−42	25	30	4.24	.03
	100	DMN	l-MFG	−35	20	51	4.35	.03
	137	DMN	l-IFG	−46	31	−13	4.23	.03
	108	DMN	r-paracingulate gyrus	9	54	3	3.97	.04
	161	Visual	r-LOC	42	−66	−8	4.28	.03

• Abbreviations: dlPFC, dorsolateral prefrontal cortex; DMN, default mode network; FDR, false discovery rate; l-, left; LECN, left executive control network; IFG, inferior frontal gyrus; MFG, medial frontal gyrus; LOC, lateral occipital cortex; r-, right; RECN, right executive control network; ROI, region of interest.

Table 5. Descriptive information for functional connectivity values based on sleep condition

Seed	Power 264 atlas ROI	Associated network	Region	FC mean (SD)		Correlation with pain sensitivity changes (r, p)
				US	FA
RECN	188	LECN	l-dlPFC	0.54 (0.27)	0.58 (0.26)	.72, .001
	201	LECN	l-dlPFC	0.60 (0.23)	0.64 (0.28)	.72, .001
	100	DMN	l-MFG	0.42 (0.18)	0.45 (0.22)	.73, .000
	137	DMN	l-IFG	0.14 (0.16)	0.26 (0.23)	.69, .001
	108	DMN	r-paracingulate gyrus	0.03 (0.21)	0.05 (0.24)	.72, .001
	161	Visual	r-LOC	0.28 (0.2)	0.27 (0.28)	.51, .03

• Abbreviations: dlPFC, dorsolateral prefrontal cortex; DMN, default mode network; FDR, false discovery rate; l-, left; LECN, left executive control network; IFG, inferior frontal gyrus; MFG, medial frontal gyrus; LOC, lateral occipital cortex; r-, right; RECN, right executive control network; ROI, region of interest.

3.4 Validation analyses

We tested whether clinical pain ratings moderated the relationship between PSQI scores and FC values from the six significant RECN-ROI pairs on a separate sample of migraine patients who reported poor sleep quality. Among these six ROIs, there was only one significant moderation effect of pain on the relationship between RECN-right paracingulate gyrus (F_(1,29) = 6.4, p = .02). The other RECN-ROI moderation analyses were not significant (ps range = .37–.91). Because this was not an ideal validation sample, however, we do not further interpret results.

4.1 Pain sensitivity following forced awakenings

When combining pain ratings and thermal stimuli temperatures, some individuals displayed hyperalgesia while others showed decreased pain sensitivity. Even across pain ratings, which were obtained following tonic noxious thermal stimuli that were paired with neutral-valenced music, there were no overall mean changes in pain sensitivity across the sample after a single night of FA sleep. Although these findings are somewhat inconsistent with previous research demonstrating hyperalgesia following sleep loss, it should be noted that most of those studies were conducted over multiple nights of accumulated sleep loss. In fact, this study's parent project recently found that two nights of sleep disruption via FA-induced hyperalgesia in a larger sample (n = 79), reducing both heat pain threshold and cold pain tolerance. The present study's design using one night of FA did not induce a ubiquitous hyperalgesic response; however, results cautiously show the early changes in RECN FC contributes to sleep disruption-induced hyperalgesia.

Further, differences in quantitative sensory testing across studies make comparisons across studies challenging. In the present study, we analyzed pain ratings obtained following the application of tonic noxious thermal stimuli that were paired with neutral-valenced music, as the parent study design did not include noxious thermal stimuli presented alone (see Seminowicz et al., 2019). As with many quantitative sensory testing studies, the uniqueness of our study design may limit the generalizability of the present results. We encourage future replication attempts. Cautiously, these results suggest varied responses to acute sleep loss in healthy adults, with some individuals evidencing greater pain vulnerability to sleep loss, which may have more significant clinical implications (Lautenbacher, Kundermann, & Krieg, 2006).

4.2 RECN functional connectivity changes following forced awakenings

Our study specifically examined FC of bilateral ECNs and DMN to functionally defined ROIs throughout the brain (Power et al., 2011), and FC among bilateral ECN and DMN masks. The ECN is a group of brain regions associated with top-down modulation of sensory stimuli via goal-directed attentional amplification of relevant information, as well as goal-directed suppression of irrelevant information (Awh et al., 2006; Gazzaley & Nobre, 2012; Smith et al., 2018). Importantly, the ECN is engaged during cognitive pain control in individuals with and without chronic pain (Ceko et al., 2015; Ceko et al., 2015; Kong et al., 2013; Rainville, 2002; Sevel, Letzen, Staud, & Robinson, 2016; Villemure & Bushnell, 2002). The DMN, on the other hand, is a group of brain regions associated with self-referential thought and focus on internal milieu (Andrews-Hanna et al., 2010; Lorenzi et al., 2011; Raichle, 2015). Its activity is typically anticorrelated with ECN activity during cognitive tasks so that better suppression of the DMN is associated with better cognitive performance (Fox et al., 2005; Seminowicz & Davis, 2007; Smith, Sip, & Delgado, 2015). Because evidence links ECN and DMN interactions with cognitive control of sensory inputs that are susceptible to perturbations from sleep disruption, the combined previous findings were suggestive of ECN-DMN FC as a putative mechanism for sleep loss-induced hyperalgesia.

Among tested networks in the present study, however, changes in pain sensitivity were only associated with changes in RECN FC, and we did not observe significant changes in the association between pain sensitivity and ECN-DMN FC. Instead, there was a positive association between increased pain sensitivity following sleep deprivation and increased RECN FC to specific regions within the DMN and LECN as well as decreased RECN FC to a primary visual network region as a function of changes in pain sensitivity. These regions included the left dlPFC (LECN), left MFG/IFG and right paracingulate gyrus (DMN), and right LOC (primary visual network).

The dlPFC is among the brain regions commonly associated with pain vigilance, awareness, and attention (Lorenz, Minoshima, & Casey, 2003; Seminowicz et al., 2013; Seminowicz & Davis, 2007; Seminowicz & Moayedi, 2017; Valet et al., 2004). Findings from experimental sleep research suggest that individual differences in emotional responses following sleep loss are associated with changes in attention (Alfarra, Fins, Chayo, & Tartar, 2015). Attentional processes are well documented in the context of experimental pain so that pain vigilance contributes to exacerbated pain intensity and distraction is associated with lower pain report (for a review, see Wiech, Ploner, & Tracey, 2008). Individuals sleeping less than 6.5 hr per night have also been found to have impaired ability to derive analgesic benefits from distraction techniques (playing video games) during an experimental pain task. Further, because stronger anticorrelation between DMN-ECN activity is associated with better segregation of internally and externally directed awareness (Buckner, 2008; Horovitz et al., 2009; Larson-Prior et al., 2009), increased RECN-DMN FC—as identified in this study—potentially suggests poorer ability to modulate attention. The combined findings of increased intrinsic, interhemispheric dlPFC FC and RECN FC to regions within the DMN might represent heightened attention to sensory stimuli, contributing to greater pain sensitivity.

It is not fully clear why we observed lateralization effects of significantly altered RECN, but not LECN, FC. Although debated, there is some evidence suggesting a greater propensity of right hemisphere activity in relation to pain processing (Coghill, Gilron, & Iadarola, 2001; Ji & Neugebauer, 2009; Symonds, Gordon, Bixby, & Mande, 2006). Additionally, a previous study examining resting-state attentional networks in individuals with insomnia found altered right, but not left, hemisphere FC (Li et al., 2018). Although it is possible that RECN FC is one mechanism contributing to sleep loss-induced hyperalgesia, the fact that this was the only seed to demonstrate significant findings across all six a priori tests tempers the strength of inference. Future research is encouraged to measure whether other neural networks contribute to sleep loss-induced hyperalgesia.

4.3 Limitations

The present results should be interpreted in the context of several important limitations. First, the small sample size used in the present study, coupled with the high number of a priori tests, weakens the strength of inference. Second, previous studies using the FA protocol over two nights showed it maximally induced hyperalgesia (Iacovides, George, Kamerman, & Baker, 2017) and diminished pain inhibition (Finan et al., 2017; Smith et al., 2007). In the present study, we used a smaller time window (one night) than previously reported, which might account for our lack of robust hyperalgesia findings. Our previous data suggest that more than one night of FA sleep is beneficial for examining sleep-pain responses. Third, although the FA paradigm is thought to have face validity for the type of sleep disturbance profile commonly observed in patients with chronic pain (i.e., multiple prolonged awakenings with significant concomitant reduction in total sleep time) (Rosseland et al., 2018), it may not generalize fully to sleep disruption linked with chronic insomnia, which may be due to a variety of clinical factors that influence pain in addition to the multiple awakenings themselves. Our results should be extended to include patient populations that suffer from sleep loss. Fourth, as mentioned earlier, pain ratings were collected from a paradigm using neutral music overlay, so we could not rule out the possible influence of distraction on imaging findings. The primary findings should be replicated in a design that includes thermal stimuli presented alone. Fifth, we did not assess individuals’ cognitive strategies used to control pain. Given that there are individual differences in cognitive pain control strategies (Seminowicz, Mikulis, & Davis, 2004), future studies should measure how ECN-DMN FC patterns relate to strategy use following sleep disturbance.