Subjective–Objective Sleep Discrepancy Is Associated With Alterations in Regional Glucose Metabolism in Patients With Insomnia and Good Sleeper Controls

Abstract

Objectives

Sleep discrepancies are common in primary insomnia (PI) and include reports of longer sleep onset latency (SOL) than measured by polysomnography (PSG) or “negative SOL discrepancy.” We hypothesized that negative SOL discrepancy in PI would be associated with higher relative glucose metabolism during nonrapid eye movement (NREM) sleep in brain networks involved in conscious awareness, including the salience, left executive control, and default mode networks.

Methods

PI (n = 32) and good sleeper controls (GS; n = 30) completed [¹⁸F]fluorodeoxyglucose positron emission tomography (FDG-PET) scans during NREM sleep, and relative regional cerebral metabolic rate for glucose (rCMR_glc) was measured. Sleep discrepancy was calculated by subtracting PSG-measured SOL on the PET night from corresponding self-report values the following morning. We tested for interactions between group (PI vs. GS) and SOL discrepancy for rCMR_glc during NREM sleep using both a region of interest mask and exploratory whole-brain analyses.

Results

Significant group by SOL discrepancy interactions for rCMR_glc were observed in several brain regions (p_corrected < .05 for all clusters). In the PI group, more negative SOL discrepancy (self-reported > PSG-measured SOL) was associated with significantly higher relative rCMR_glc in the right anterior insula and middle/posterior cingulate during NREM sleep. In GS, more positive SOL discrepancy (self-reported < PSG-measured SOL) was associated with significantly higher relative rCMR_glc in the right anterior insula, left anterior cingulate cortex, and middle/posterior cingulate cortex.

Conclusions

Although preliminary, these findings suggest regions of the brain previously shown to be involved in conscious awareness, and the perception of PSG-defined states may also be involved in the phenomena of SOL discrepancy.

fluorodeoxyglucose, positron emission tomography, neuroimaging, primary insomnia, sleep discrepancy

Statement of Significance

Patients with insomnia often report greater difficulty going to sleep and staying asleep than is documented using objective measures, such as polysomnography. This subjective–objective sleep discrepancy is a well-known phenomenon to sleep experts. Clinicians and sleep researchers are interested in understanding the source of these discrepancies, because identifying their source may lead to interventions capable of treating a core mechanism of insomnia. The findings from this study suggest that sleep discrepancy in patients with insomnia may be associated with altered brain activity during non-rapid eye movement sleep. Brain activity in the right anterior insula, left anterior cingulate cortex, and middle/posterior cingulate cortex may be involved in the perception and retrospective diary reports of sleep onset latency.

INTRODUCTION

Current diagnostic classifications^1,2 define insomnia based on self-reported sleep difficulties, including prolonged sleep onset latency (SOL) and wake after sleep onset (WASO). Subjective–objective sleep discrepancy, the time difference between self-reported and objectively measured sleep features, is common in insomnia.^3–5 Although good sleeper controls (GS) tend to report SOL values that are similar to or shorter than those recorded with polysomnography (PSG), patients with insomnia tend to report longer SOL values than PSG SOL.^6–8 Patients with insomnia are also more likely than GS to report having been awake when researchers experimentally rouse them from sleep.⁹ These negative sleep discrepancies—when self-reports suggest greater sleep difficulty than are measured by objective measures—have contributed to conceptualizations of insomnia as a psychological problem characterized by patient misperception, bias, or error regarding their sleep difficulties.^10,11 Although a subset of patients experience a consistent pattern of negative discrepancy, sleep discrepancy varies in magnitude, direction, and night-to-night frequency in most patients with insomnia. Although the magnitude of the discrepancy tends to be greater in the direction of negative sleep discrepancy, patients with insomnia also under-estimate SOL relative to PSG on many nights.^7,12,13 Moreover, among older individuals with insomnia, we have found greater within-individual variability in SOL discrepancy from night-to-night than is found between individuals on average.^14,15 Because averaging across nights may obscure results,¹⁶ we propose that understanding the pathophysiology of negative SOL discrepancy in patients with insomnia may be best achieved by studying it on nights that it occurs.

Several observations contribute to the hypothesis that sleep discrepancy has neurophysiological origins. Increased cognitive or central nervous system “hyperarousal,” impaired inhibitory processes in the brain during sleep, or insufficient local sleep, have been proposed as potential mechanisms of negative discrepancy.^17–21 Providing support for the hyperarousal hypothesis, more negative sleep discrepancy in patients with insomnia has been linked to heightened self-reported cognitive activation in the form of intrusive thoughts, sleep threat monitoring, rumination, and worry during sleep onset. Negative sleep discrepancy in patients with insomnia has also been linked to heightened central nervous system activation, as assessed by increased electroencephalography (EEG) activity in the β range (14–35 Hz) during sleep.^8,22 Perlis and colleagues noted that the duration of heightened β EEG activity after PSG-defined sleep onset approximates the amount of negative SOL discrepancy (i.e., 10–45 minutes) typically observed in patients with insomnia.²³ Another study assessed cyclic alternating pattern (CAP) rates in patients with insomnia who had extreme negative sleep discrepancy and found higher CAP rates—an EEG marker of enhanced brain activation and unstable sleep—during the sleep interval between objective and subjective sleep onset.²⁴ Several other studies have experimentally induced negative sleep discrepancy using caffeine or behavioral paradigms that increase wake drive in GS.^25–29 An event-related potential (ERP) study found that, compared with patients with insomnia who did not have negative sleep discrepancy on two consecutive nights, patients with insomnia who had negative sleep discrepancy on those nights showed greater attentional processes (greater N1 and P2 amplitudes) during sleep onset—suggesting impaired inhibition or hyperarousal—but that they also had greater shifts in attentional processes from wakefulness to sleep—suggesting that stronger inhibitory processes may be required for these patients to achieve and maintain nonrapid eye movement (NREM) sleep.¹⁹ Collectively, these observations suggest that negative SOL discrepancy may be associated with altered brain activity, involving hyperarousal or impaired inhibitory processes during the process of sleep onset and sleep.

Local sleep processes may also help us to understand sleep discrepancy.²⁰ Although sleep has traditionally been conceptualized as system-wide, whole brain phenomena, neuroimaging methods provide evidence for the notion of regional variations in sleep intensity at the level of brain regions/networks.^30,31 Local sleep processes occur within neuronal groups, which are further organized in brain regions and networks. These regions and networks show use-dependent patterns in sleep intensity during NREM sleep.³² In humans, frontal and parietal heteromodal cortex, posterior cingulate, and thalamus show the greatest reductions in relative glucose metabolism during NREM sleep relative to the wake state.³³ In GS, greater amplitude of slow wave activity corresponds with lower cerebral blood flow in the ventromedial prefrontal cortex, anterior cingulate, anterior insula, basal forebrain, putamen, and precuneus/posterior cingulate during NREM sleep.³⁴ Alterations in sleep intensity at the brain region and network levels of measurement have also been documented in human sleep disturbances once thought to be whole-brain phenomena, including insomnia³³ and the first night effect.³⁵ Alterations in regional sleep intensity could plausibly correspond with reports of sleep difficulties that conflict with PSG measures. For instance, relatively greater activity during NREM sleep in brain regions/networks involved in self-referential thinking or awareness could constitute a neural correlate of negative sleep discrepancy.

Each of these models—hyperarousal, impaired inhibitory processes, and reduced local sleep processes—predicts that negative sleep discrepancy in patients with insomnia may be associated with heightened brain activity during PSG-defined sleep. Among adults with insomnia, we were particularly interested in determining whether discrepancy in SOL would be associated with heightened brain activity during NREM sleep in regions associated with the perception of being awake. Levels of conscious awareness for exteroceptive stimuli (e.g., sensory/environmental), interoceptive stimuli (e.g., bodily), or self (e.g., thoughts, emotions, memories) may contribute to an individual’s perception of being awake. The anterior cingulate, insula, and dorsolateral prefrontal cortex may be central nodes of brain networks involved in the perception of being awake during NREM sleep.³⁶ The insula is a major node of the salience network that responds to environmental stimuli perceived to be important and may be involved in conscious awareness of feelings;^37–39 a white-matter region next to and possibly part of the insula—the claustrum—has been proposed as another region supporting conscious awareness.⁴⁰ The anterior cingulate cortex is also a central node of the salience network that has been implicated in levels of conscious awareness and attention processes.^39,41 The frontal cortex is involved in higher-order thought processes including working memory. The left middle frontal gyrus, a major node of the left executive control network, is of particular interest to this study as it has been linked to insomnia in several prior studies.^42,43 In addition to those previously proposed regions, the precuneus/posterior cingulate cortex represents a central node of the default mode network that has been implicated as being involved in levels of conscious awareness of the self.⁴⁴ Importantly, these regions have also been linked to insomnia in several neuroimaging studies.³³

In this study, we investigated the associations between SOL discrepancy and relative regional cerebral metabolic rate for glucose (rCMR_glc)—a measure of regional brain activity—during NREM sleep in a sample of patients with primary insomnia (PI) and GS. Because PSG was only collected directly before and during the first NREM sleep period on the night of the [¹⁸F]fluorodeoxyglucose positron emission tomography (FDG-PET) scan (approximately 40 minutes), discrepancy was examined only for SOL, and not for other measures such as WASO or total sleep time. We posited that PI would have more negative SOL discrepancy than GS and that within PI greater negative SOL discrepancy (i.e., self-report SOL greater than PSG SOL) on the PET night would be associated with higher relative glucose metabolism in brain networks involved in conscious awareness, including the salience, left executive control, and default mode networks. In contrast, we anticipated that self-reported SOL in GS would be more commensurate with PSG than in patients with PI and that the variance in SOL discrepancy in GS would not be associated with systematic differences in glucose metabolism in the brain during NREM sleep.

METHODS

Participants

The methods of this study have been described in detail elsewhere⁴³ and will be summarized here. Participants included patients with PI, determined by a structured clinical interview based on the Diagnostic and Statistical Manual of Mental Disorders, 4th ed. (DSM-IV) criteria⁴⁵ and GS, who had no insomnia, based on the same interview and a Pittsburgh Sleep Quality Index (PSQI) score < 5. Participants were drawn from eight protocols conducted at the University of Pittsburgh between 1998 and 2012 that used the FDG-PET method⁴⁶ for studying regional brain metabolism in PSG-defined sleep-wake states. The current study constitutes a secondary analysis of data from these protocols. The Institutional Review Board and the Human Use Subcommittee of the Radiation Safety Committee at the University of Pittsburgh approved all protocols. Participants gave written informed consent and were compensated for participation.

These protocols provided a sample pool of 122 participants. Participants for the current analyses met the following criteria: (1) age 18–60 years, (2) no self-reported sleep disorders (other than insomnia for the PI group), (3) caffeine < 400 mg per day on average, (4) ability to abstain from tobacco and alcohol during the study, (5) ability to abstain from drugs known to affect sleep for at least 2 weeks before participation (4 weeks for fluoxetine), (6) negative pregnancy test for women, and (7) no significant current medical or psychiatric condition. Exclusion criteria for the present analyses and the number of individuals excluded for each criteria were as follows: (1) self-reported left-handedness (n = 1), (2) PSQI > 5 for GS (n = 1), (3) apnea–hypopnea index (AHI) ≥ 15 (n = 4), and (4) periodic limb movements with awakening (PLMs) ≥ 20 per hour (n = 12), technical problems with the PET scan (i.e., scan was scheduled but not completed by participant, coregistration problems, limited field of view, and failed FDG injection; n = 13), having REM sleep during the uptake period (n = 3) and missing sleep diary data needed to compute SOL discrepancy on the NREM PET scan night (n = 26). The final sample included 32 PI and 30 GS. Demographic and clinical characteristics are presented in Table 1. In brief, participants were young to middle-aged adults (M = 37 years, range 19–60), 58% female, and 81% white. Groups were well matched for age, gender, and race.

Table 1

Characteristics of Patients With PI and GS Controls.

Characteristics	PI (n = 32)	GS (n = 30)	t/χ2/Z	Df	p
Age, y	37(9)	39(11)	t = −0.8	60	.450
Sex, female	13(41%)	19(63%)	χ2 = 3.2	1	.074
Race, white	27(84%)	26(87%)	χ2 = 0.1	1	.798
Mood symptomsa
Mild–moderate depressive symptoms	12(38%)	2(7%)	χ2 = 8.4	1	.004**
Mild–moderate anxiety symptoms	7(22%)	1(3%)	χ2 = 4.7	1	.030*
Baseline laboratory night
Diary SOL, min	30[15, 60]	15[10, 20]	Z = −3.5		<.001***
PSG SOL, min	21[10, 34]	15[6, 22]	Z = −2.1		.038*
SOL discrepancy, min	9[0, 24]	0[−4, 7]	Z = −2.5		.012*
NREM PET scan night
Diary SOL, min	30[15, 45]	15[10, 20]	Z = −2.6		.009**
PSG SOL, min	21[13, 42]	18[9, 38]	Z = −1.0		.342
SOL discrepancy, min	3(16)	−5(12)	t = 2.3	60	.028*

*p < .05, **p < .01, ***p < .001.

^aParticipants did not meet criteria for a current anxiety or depression diagnosis.

SOL = sleep onset latency; NREM = nonrapid eye movement sleep; PSG = polysomnography; PET = position emission tomography; PI = primary insomnia; GS = good sleeper. Categorical values are reported as n (%). Variables with normal distributions within both the PI and GS groups, determined using Kolmogorov–Smirnov Test for Normality (p > .05), are reported as mean (standard deviation) values and variables with non-normal distributions in at least one group (PI or GS) are reported as median [interquartile range] values.

Procedures

Participant demographics, clinical characteristics, and sleep quality were assessed with validated self-report questionnaires. To rule out sleep, psychiatric, and medical disorders, all participants were also assessed with validated clinician-administered questionnaires and interviews, overnight PSG, and medical history/physical examination.

Clinical Interview

Current and past history of psychiatric disorders was assessed by self-reported history and the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-IV).⁴⁷ SCID-IV data from six participants (one PI and five GS) were missing in the dataset used in these analyses. Participants in both groups were free of current psychiatric disorders. Participants were not excluded from this study for prior psychiatric history. Among PI, prior history included depression (n = 7) and substance use (n = 1) disorders. Among GS, one participant had a prior psychiatric history of depression, anxiety, and substance use disorders.

Mood Measures

Instruments used to measure the severity of anxiety and depression symptoms differed across protocols. For self-reported anxiety, the Beck Anxiety Inventory,⁴⁸ a standard Visual Analog Scale for Anxiety, or the State-Trait Anxiety Index—Form Y-1⁴⁹ was used. For self-reported depression, the Beck Depression Inventory,⁵⁰ Inventory of Depressive Symptomatology,^51,52 or Profile of Mood States-depression score⁵³ was used. The total scores for each measure (minus sleep items) were converted to standard ordinal scales delineating minimal from mild-moderate levels of anxiety or depression severity.^54–58 Although no participants had a current psychiatric diagnosis, 49% of the PI group had mild-moderate symptoms of anxiety or depression compared with only 6% of the GS group (Table 1).

Sleep Measures

Questionnaire.

 The PSQI was used to assess sleep quality over the past month.^59,60 The PSQI is a well-validated, 18-item self-report measure. Scores range from 0–21, with higher scores denoting worse sleep quality. A score greater than five indicates clinically significant levels of sleep disturbance.⁶¹

Sleep diaries.

 On the PSG nights, participants completed a sleep diary each morning to assess self-reported sleep features, including SOL for the prior night. This questionnaire includes the standard sleep diary questions and several additional questions related to self-reported sleep quality and mood.

Polysomnography.

 Participants completed at least three overnight sleep studies, including screening/adaptation, baseline, and NREM PET scan nights using standard procedures described in detail previously.⁴⁶ Sleep apnea and periodic limb movement disorder were screened during the initial screening/adaptation PSG night according to standard methods.⁴³ Sleep features other than AHI and PLMs were not compared across groups on the screening night due to well-characterized effects of adaptation on sleep.⁶² Standard PSG-assessed sleep features including SOL and WASO (described below) were assessed on a subsequent uninterrupted baseline night. An overnight NREM PET scan night occurred on a separate night during which PSG monitoring was limited to only the first part of the night (SOL and FDG-PET uptake periods). Nightly urine screens were conducted to confirm that participants were free of alcohol and recreational drugs and that female participants were not pregnant.

The PSG montage used during overnight studies included the C4/A1-A2 EEG channel, bilateral electrooculogram referenced to A1-A2, and submentalis electromyogram. These PSG signals were digitized and visually scored for staging in 20-second epochs according to validated procedures by sleep technicians who were blinded to participant group. All studies were scored according to Rechtschaffen and Kales criteria.⁶³

The criterion for PSG-assessed SOL was the number of minutes from lights out until sleep onset, defined as the start of the first stage 2 NREM sleep epoch followed by at least 10 minutes of sleep (NREM stages 2–4 or REM sleep), and interrupted by no more than 2 minutes of wake or stage 1. This definition of SOL was the standard used by the Sleep and Chronobiology Center at the University of Pittsburgh for PET studies. This stringent criterion had been used in many other studies of sleep, depression, and insomnia at the University of Pittsburgh and was chosen to maintain consistency with these previous studies. This criterion ensured that the PET scans captured a stable state of NREM sleep. SOL was calculated for both the baseline night and the NREM PET scan night. To confirm that results were not affected by the definition of sleep onset, we computed two additional criteria for PSG-defined sleep onset for the PET night data: (1) the first epoch of any sleep stage and (2) the first epoch of stage 2 sleep; the analyses and results of these different criteria for PSG SOL are reported in Supplementary Material.

Sleep discrepancy variables.

  Because PSG was only collected directly before and during the uptake period on the NREM PET scan night (approximately 40 minutes), only SOL could be assessed for that night. Thus, we focused on subjective–objective sleep discrepancy relevant to participants’ perception of the amount of time spent awake in bed as measured with SOL. Sleep discrepancy was calculated for SOL by subtracting PSG-measured SOL from self-reported values from the sleep diary on each night.

Neuroimaging Protocol

The FDG-PET method used in this study has been described in detail in previous publications.⁴³ In brief, participants received a structural magnetic resonance imaging (MRI) scan. These images were anterior commissure–posterior commissure (AC-PC) aligned and then normalized to the ICBM 152 template (Montreal Neurological Institute) via the unified segmentation technique⁶⁴ in SPM8 (http://www.fil.ion.ucl.ac.uk/spm/software/spm8/, accessed October 11, 2017).

Each participant had an FDG-PET study during NREM sleep. The FDG injections for NREM PET imaging were conducted at the Neuroscience Clinical and Translational Research Center during the first NREM sleep cycle. Intravenous (IV) catheters were placed in each arm of participants: one for the FDG injections and another for venous blood sampling. The FDG injections for the NREM sleep PET scans occurred after 20 minutes of continuous sleep (NREM stage 2–4) in bed. Participants were left undisturbed for 20 minutes before being awakened and transported by a wheelchair to the PET center for the NREM sleep PET scan. PSG-defined sleep-wake states before, during, and 20 minutes after FDG injections were confirmed using EEG. Sixty minutes after FDG injection, participants underwent PET imaging using one of the two Siemens/CTI ECAT HR+ tomograph scanners (CTI PET Systems, Knoxville, TN) in 3D mode (63 transaxial planes, field of view [FOV] = 15.2 cm, slice width = 2.4 mm). Subjects were positioned in the scanner to maximize full brain coverage. A 30-minute emission scan (six sequential 5-minute scans) was acquired over the 60–90 min post-injection period while participants lay with eyes closed. Venous blood was sampled (1 mL each) at six time points (45, 55, 65, 75, 85, and 90 minutes post-injection), for the determination of FDG radioactivity (all samples) and glucose (first and last samples) plasma concentrations. A windowed transmission scan (10–15 minutes) was acquired before emission imaging and used for PET attenuation correction. Other corrections included scanner normalization, dead time, scatter, random coincidences, and radioactive decay. The PET data were reconstructed by filtered back-projection. The final in-plane spatial resolution was 6.0 mm. Attenuation-corrected, decay-corrected, FDG-PET data were motion-corrected (if needed) and averaged over all frames (60–90 minutes post-injection) using AIR 3.0 software (http://air.bmap.ucla.edu/AIR3/index.html, accessed October 11, 2017).⁶⁵ Each subject’s averaged FDG-PET data were coregistered to their AC-PC aligned structural T1-weighted MRI scan, normalized using the previously obtained transformation parameters, and smoothed with a 10.0-mm full width at half maximum Gaussian filter. Data were quality control checked for FOV positioning, motion, and co-registration problems. Relative regional cerebral metabolic rate for glucose (rCMR_glc) was calculated at each voxel by dividing by the global FDG-PET intensity across all brain voxels for each scan and then multiplying by 50. This calculation accounts for global nuisance effects and puts these relative data into an intuitively accessible scale.

FDG and glucose plasma concentrations were used to quantify the absolute metabolic rate of deoxyglucose (MRD_glc) based on a modified version of a simplified kinetic method,⁶⁶ validated, and routinely applied in our laboratory.⁶⁷ Because the brain metabolizes a large portion of the total glucose utilized in the body, MRD_glc provides an indirect, semi-quantitative measure of absolute brain glucose metabolism.

Analyses

Assumptions that data were normally distributed were checked for each variable within each group (PI and GS) using Kolmogorov-Smirnov Test for Normality. Demographic, clinical, and sleep features were compared across groups using student’s t, Mann–Whitney U, or chi-square tests (Table 1). The sleep features of the sample pool from which these participants were drawn have been well characterized in our previous paper.⁴³ Here, we present the sleep features most relevant to the present analysis. Spearman’s coefficient was used to assess the correlation between SOL discrepancy on the baseline night and SOL discrepancy on the PET night. These analyses were conducted in IBM SPSS 22 (IBM Corp., Armonk, NY, USA) or SAS version 9.3 (SAS Institute, Cary, NC).

Region of Interest Mask Analysis

Relative rCMR_glc during NREM sleep was analyzed using Statistical Parametric Mapping 12 (SPM 12) (http://www.fil.ion.ucl.ac.uk/spm/software/spm12/, accessed October 11, 2017). Regions of interest (ROI) included the bilateral insula, bilateral anterior cingulate cortex, left middle frontal gyrus, and bilateral precuneus/posterior cingulate, all defined from the Anatomical Automatic Labeling atlas.⁶⁸ These regions were combined into a large bilateral ROI mask for the ROI mask analysis. Cluster-wise extent thresholds were computed using the most current version of 3dClustSim (https://afni.nimh.nih.gov/pub/dist/doc/program_help/3dClustSim.html, accessed October 11, 2017). This version of 3dClusterSim corrected for the previously identified bugs in the code. An uncorrected voxel-wise height threshold of p < .005 was entered into 3dClustSim, as well as the full-width at half maximum computed by 3dFWHMx of the residuals of each statistical analysis. The algorithm then computed the minimum cluster size threshold needed to correct for multiple comparisons, which controls the family-wise error rate at α < 0.05.

Regression analyses examined concurrent associations between SOL discrepancy on the NREM PET scan night and relative rCMR_glc during NREM sleep on the same night in PI and GS. Regression models included the main effect for group (PI vs. GS), main effect for SOL discrepancy, and a group by SOL discrepancy interaction for rCMR_glc limited to the combined ROI inclusive mask. The group and SOL discrepancy variables were mean centered before being multiplied to create the interaction term. The interaction term determined whether the slope of the correlations between our two continuous variables differed by group (PI vs. GS). For each significant cluster, a custom MatLab script was used to extract the mean rCMR_glc values for each participant. Extracted values were exported to SPSS for follow-up analyses.

Exploratory Whole-Brain Analysis

We conducted an exploratory whole-brain (voxel-wise) analysis, paralleling the main ROI mask analysis, to explore whether SOL discrepancy was associated with relative rCMR_glc outside of our hypothesized regions. For the significant cluster that emerged in this analysis, we used a custom MatLab script to extract the mean rCMR_glc value for each participant and moved those values to SPSS for follow-up analyses.

Follow-up Analyses

Anxiety and Depression Symptoms and Group Slopes.

 To assess associations between glucose metabolism and SOL discrepancy, moderation analyses were run using the PROCESS plugin for SPSS (Model 1) for each mean rCMR_glc value extracted from the clusters identified as significant in the ROI mask and exploratory whole-brain analyses. Each served as the dependent variable in their respective model. Group (PI vs. GS) and SOL discrepancy were entered as the independent and M variables, respectively. We included symptoms of anxiety or depression (none vs. mild/moderate symptoms) as a covariate.

Indirect Whole-Brain Glucose Metabolism Analysis.

 To ensure that the results of the ROI mask and whole-brain exploratory analyses on regional rCMR_glc were not driven by an association between SOL discrepancy and whole-brain glucose metabolism, regression analyses were conducted to test for main effect of group (PI vs. GS), main effect of SOL discrepancy, and a group (PI vs. GS) by SOL discrepancy interactions for MRD_glc during NREM sleep. This analysis was conducted in SPSS.

Sensitivity Analyses.

 We conducted several sensitivity analyses to confirm that our results were not due to potential confounds including how PSG SOL was defined, which PET scanner was used, AHI or PLMs, or potential outliers. The details of these analyses and their results are reported in Supplementary Material.

RESULTS

Table 1 shows demographic, psychological, and sleep characteristics of PI and GS. Groups did not differ significantly in demographic characteristics. Patients with PI had significantly more symptoms of anxiety and depression than GS. On the baseline night, PI had greater SOL than GS, based on both sleep diary and PSG measures. On the NREM PET scan night, PI reported significantly greater SOL than GS, but PSG-assessed SOL was not significantly different. SOL discrepancy on the baseline night was not significantly correlated with SOL discrepancy on the PET night (r_s = 0.18, p = .17). The PI group had significantly higher SOL discrepancy values on the baseline and the NREM PET scan nights than GS. By design, the NREM sleep period captured with the PET scan occurred at least 20 minutes after PSG defined sleep onset (M = 24.0 ± 11.2 minutes before FDG injection). Perceived sleep onset (based on sleep diary SOL) also occurred before the NREM sleep period captured with the PET scan (M = 23.7 ± 17.7 minutes before the FDG injection), in all but three participants. As was true in the larger sample,⁴³ the groups did not differ significantly in the number of epochs scored as wake or stages 1–4 during the uptake period.

ROI Mask Analysis

Figure 1 shows the results of the analysis within the large ROI mask that investigated SOL discrepancy on the NREM PET scan night. A significant group (PI vs. GS) by SOL discrepancy interaction indicated that the association between SOL discrepancy and relative rCMR_glc during NREM sleep in the right anterior insula [t_max(58) = 3.5, k = 248, x = 38, y = 20, z = 2] and left anterior cingulate cortex [t_max(58) = 3.7, k = 140, x = −8, y = 52, z = 0] differed by group status (p_corrected < .05). See Supplementary Figure 1 to view the full extent of the significant clusters.

Figure 1

Regions of Interest Mask Analysis. We investigated relative regional cerebral metabolic rate for glucose (rCMR_glc) in patients with primary insomnia (PI, n = 32) and good sleeper controls (GS, n = 30) during nonrapid eye movement sleep in relation to sleep onset latency (SOL) discrepancy on that night. SOL discrepancy was calculated as self-reported SOL minus polysomnography measured SOL. Regions of interest including the bilateral insula, bilateral anterior cingulate cortex, left middle frontal gyrus, and bilateral precuneus/posterior cingulate were combined in an inclusive mask and used in a regression analysis conducted in SPM 12. There was a significant group (PI vs. GS) by SOL discrepancy interaction for relative rCMR_glc in the (a) right anterior insula and (b) left anterior cingulate cortex. Cluster sizes (k) > 118 were significant at height threshold p < .005, cluster threshold p_corrected < .05. We extracted the voxel-wise data from the significant clusters and plotted the data to the right. In the scatter plot, the closed circles represent mean rCMR_glc extracted from the respective significant cluster for PI participants and the open circles represent GS. The solid line represents the line of best fit for the PI, and the dashed line represents the line of best fit for the GS.

As expected based on results from our previous study,⁴³ there was a significant main effect for group such that PI had significantly lower relative rCMR_glc than GS in clusters within the ROI mask. These main effects are not repeated in this manuscript. In the total sample, there was no significant main effect of SOL discrepancy on relative rCMR_glc, when investigated voxel-wise within the ROI mask.

Whole-Brain Exploratory Analyses

Figure 2 shows the results of the whole-brain exploratory analysis for SOL discrepancy on the NREM PET scan night. This analysis revealed a significant interaction for a cluster in the middle/posterior cingulate cortex [t_max(58) = 4.3, k = 784, x = −8, y = −26, z = 42]. See Supplemental Figure 2 to view the full extent of the significant cluster.

Figure 2

Exploratory whole-brain analysis (voxel-wise). We investigated relative regional cerebral metabolic rate for glucose (rCMR_glc) in patients with primary insomnia (PI, n = 32) and good sleeper controls (GS, n = 30) during NREM sleep in relation to sleep onset latency (SOL) discrepancy on that night. SOL discrepancy was calculated as self-reported SOL minus polysomnography measured SOL. In the whole-brain exploratory (voxel-wise) regression analysis, there was a significant group (PI vs. GS) by SOL discrepancy interaction for relative rCMR_glc in the middle/posterior cingulate cortex. Cluster sizes (k) > 705 were significant at height threshold p < .005, cluster threshold p_corrected < .05. We extracted the voxel-wise data from the significant cluster and plotted the data to the right. In the scatter plot, the closed circles represent mean rCMR_glc values extracted from the respective significant cluster for PI participants and the open circles represent GS participants. The solid line represents the line of best fit for the PI group and the dashed line represents the line of best fit for the GS group.

As expected based on results from our previous study in a larger sample,⁴³ there was a significant main effect for group such that PI had significantly lower relative rCMR_glc in several clusters across the brain. These main effects are not repeated in this manuscript. In the total sample, there were no main effects of SOL discrepancy on relative rCMR_glc when investigated voxel-wise in the entire brain.

Follow-up Analyses

Anxiety and Depression Symptoms and Group Slopes

Table 2 reports the results of follow-up analyses investigating the associations among group, SOL discrepancy, anxiety or depressive symptoms, and mean rCMR_glc extracted from clusters identified in the ROI mask and whole-brain exploratory analyses. The interactions in each cluster remained significant while adjusting for the presence of anxiety or depression symptoms. Symptoms of anxiety or depression symptoms, present primarily in patients with insomnia, were associated with higher mean rCMR_glc extracted from the cluster from the group by sleep discrepancy interaction in the right anterior insula.

Table 2

Follow-up Analyses Investigating Having Anxiety or Depression Symptoms (Anx/Dep), the Main Effect for Group (PI vs. GS), the Main Effect for SOLd, the Group by SOLd Interaction for Mean rCMR_glc Extracted From the Clusters Identified as Significant in the Main Analyses.

Model	F-statistic	Variable	Standardized β	t-Statistic	p
Right anterior insula	F(5,57) = 6.05, p < .001***	Group (PI vs. GS)	0.97	−1.59	.117
		SOLd	0.03	−0.73	.47
		Group × SOLd	0.06	4.55	<.001***
		Anx/Dep symptoms	0.97	2.01	.049*
Left anterior cingulate	F(5,57) = 6.52, p < .001***	Group (PI vs. GS)	0.7	−2.91	.005**
		SOLd	0.02	−1.6	.115
		Group × SOLd	0.05	3.91	<.001***
		Anx/Dep symptoms	0.98	1.73	.088
Middle/posterior cingulate	F(5,57) = 13.77, p < .001***	Group (PI vs. GS)	0.75	−2.06	.044*
		SOLd	0.02	−1.9	.06
		Group × SOLd	0.05	4.77	<.001***
		Anx/Dep symptoms	0.76	0.65	.521

*p < .05, **p < .01, ***p < .001.

PI = primary insomnia; GS = good sleeper controls; SOLd = sleep onset latency discrepancy; rCMR_glc = regional cerebral metabolic rate for glucose.

This analysis also showed that more negative SOL discrepancy—having diary SOL estimates that exceeded PSG SOL—in PI was associated with significantly higher mean rCMR_glc extracted from the cluster identified in the right anterior insula and the middle/posterior cingulate cortex (p < .01 for both). Having diary SOL estimates that were less than PSG SOL in GS was associated with higher mean rCMR_glc in each of the clusters identified in the main analyses (p < .001 for all).

Indirect whole-brain glucose metabolism analysis

There was no main effect for group (PI vs. GS) or for SOL discrepancy, or group by SOL discrepancy interaction for MRD_glc, the semi-quantitative (indirect) measure of whole-brain glucose metabolism, during NREM sleep (p > .10).

DISCUSSION

Merica and colleagues first proposed that negative sleep discrepancy in patients with insomnia may reflect a pattern of heightened brain activity in sub-portions of the brain during PSG-defined sleep.^20,69 This study identified specific brain regions that may be involved including the right anterior insula, left anterior cingulate cortex, and middle/posterior cingulate. Each cluster identified in this study was in major nodes of the anterior salience or dorsal default mode networks, whose functions include salience detection and conscious awareness of exteroceptive stimuli, interoceptive processes, and self.^37,40,70 They are also brain regions that have been linked to greater slow wave activity, a measure of sleep depth.³⁴ The middle/posterior cingulate cluster that emerged in the whole-brain exploratory analysis was more anterior than the precuneus/PCC region we had predicted. Nevertheless, the identified cluster converges on a “hotspot” region in the posterior cortex identified as being involved in conscious experience during NREM sleep in a high-density EEG study.⁷¹ Consideration of the associations between SOL discrepancy and glucose metabolism during NREM sleep in PI and GS may provide insights into whether SOL discrepancy is consistent with the predictions of the hyperarousal, impaired inhibition, or local sleep models of sleep discrepancy.

Our hypothesis that negative SOL discrepancy in PI would be associated with higher relative glucose metabolism in brain regions associated with conscious awareness was supported. In PI, reporting greater SOL than was measured with PSG was associated with significantly higher relative glucose metabolism during NREM sleep in the right anterior insula and middle/posterior cingulate cortex. This is consistent with all three sleep discrepancy models—hyperarousal, impaired inhibition, and insufficient local sleep.

Contrary to our hypothesis, we found that SOL discrepancy was also associated with regional glucose metabolism in GS, and in an opposite direction to that observed in the PI group. In GS, more positive SOL discrepancy—reporting less SOL than was measured with PSG—was associated with higher relative glucose metabolism during NREM sleep in the right anterior insula and middle/posterior cingulate cortex as well as in the left anterior cingulate cortex.

The cause of this group-by-SOL discrepancy interaction with regional glucose metabolism during NREM sleep is not immediately obvious. If higher glucose metabolism in these brain regions during NREM sleep was due to hyperarousal, impaired inhibition, or reduced local sleep in PI, we would not expect positive SOL discrepancy to be associated with higher glucose metabolism in GS during NREM sleep; GS do not have hyperarousal, impaired inhibition, or insufficient local sleep.

Although speculative, it is possible that regional glucose metabolism reflected different types of neuronal activity among individuals in PI and GS groups. For example, regional metabolism measured by FDG-PET does not distinguish between inhibitory and excitatory neuronal activity. Different patterns of regional glucose metabolism relative to SOL discrepancy could be due to different levels of inhibitory and excitatory neuronal activity across groups. For instance, positive SOL discrepancy may be associated with higher glucose metabolism due to heightened inhibitory processes in GS; negative SOL discrepancy may be associated with higher glucose metabolism due to heightened excitatory processes in PI. A similar explanation has been proposed to explain why α sleep is both positively and negatively correlated with sleep discrepancy in different patient samples;⁷² the measured physiology may appear the same but the underlying type of brain activity may be different. Conversely, it may be possible that GS who had positive sleep discrepancy on the PET night perceived going to sleep before PSG-defined sleep due to greater inhibitory processes in brain regions involved in conscious awareness, whereas patients with insomnia who had negative sleep discrepancy, those who perceived going to sleep after PSG-defined sleep on the PET night, required greater inhibitory processes in those brain regions to reach the stable NREM state captured with our PET scan. A similar explanation has been proposed to explain ERP differences in patients with insomnia with and without negative sleep discrepancy.¹⁹ Again, we acknowledge that such explanations are speculative.

Clinical Considerations

The perception of disturbed sleep, particularly when it exceeds PSG measures, may lead to increased worry and maladaptive sleep behaviors; thus, negative sleep discrepancy represents a precipitating, exacerbating, and perpetuating risk factor for insomnia.⁷ Cognitive behavioral therapy for insomnia (CBTI) reduces the tendency of patients to report greater SOL than is measured with PSG.^15,73 Benzodiazepines reduce the likelihood that patients with insomnia will report PSG-defined sleep as wakefulness.^74–76 Thus, CBTI and insomnia medications may target mechanisms of negative sleep discrepancy. Future studies could test whether these treatments are associated with altered metabolism or other measures of activation in brain regions involved in conscious awareness including the right anterior insula, left anterior cingulate, and middle/posterior cingulate. Such studies may lead to novel insomnia treatments.

Strengths/Limitations

This was a secondary analysis of several studies that were not designed specifically to assess conscious awareness or sleep discrepancy. Self-reported sleep features were assessed retrospectively, in the morning following final awakening. Ideally, self-reported SOL would have been collected upon initial awakening directly before the PET scan. Questions concerning conscious awareness during the uptake period would have also been useful. Nevertheless, most people make judgments about their sleep upon final awakening; thus, the methods used in this study have high ecological validity to the study of sleep discrepancy. Analyses were restricted to SOL on the PET night because EEG recording ended following the 20-minute uptake period, making it impossible to investigate WASO discrepancy on the PET night. Although the mean group difference in SOL discrepancy on the PET night was relatively small (8 minutes), it was significantly different between groups. Importantly, there was suitable power to replicate our previous findings of group differences in relative glucose metabolism during NREM sleep, and there was enough inter-individual variability in our samples to adequately test our hypothesis.

We report on individuals with and without insomnia who were able to fall asleep with instrumentation for PSG as well as two IV catheter lines and the knowledge that they would be awakened for a PET scan. This could have biased our sample toward individuals with less difficulty sleeping under stressful situations. That group (PI vs. GS) by SOL discrepancy interactions were identified suggests that, despite such potential bias, individuals with and without insomnia have different biological characteristics. Our finding that the association between SOL discrepancy and regional brain metabolism was opposite in direction among individuals with PI and GS could be interpreted as evidence supporting the validity of insomnia. In other words, good sleepers and individuals with insomnia cannot be viewed as existing on a single continuum, at least with regard to SOL discrepancy.

We did not find a correlation between SOL discrepancy on a baseline night and the PET night. Other studies have failed to find correlations between SOL discrepancies across nights.^14,15 Thus, SOL discrepancy may be an event that happens by degree (magnitude, direction, and frequency) in patients with insomnia, rather than a fixed trait in most patients. A strength of our study was that SOL discrepancy was computed on the same night as the PET scan, allowing us to investigate the pathophysiology of SOL discrepancy on the night that it occurred. Nevertheless, follow-up studies that recruit patients with paradoxical insomnia are needed to determine whether the results of this study apply to those patients.

Previous studies have linked sleep discrepancy to daytime sleepiness,^77,78 depression,^79,80 and poorer cognitive function.^81,82 The associations between SOL discrepancy and regional glucose metabolism in the PI group were robust even when adjusting for the presence of mild–moderate anxiety and depression symptoms. The finding that anxiety and depression symptoms were associated with greater anterior insula glucose metabolism during NREM sleep may suggest that a common pathophysiological pathway in the brain exists among negative SOL discrepancy, right anterior insula activity during NREM sleep, and psychiatric symptoms in patients with insomnia.

CONCLUSION

We report significant correlations between SOL discrepancy and glucose metabolism in brain regions that have been previously linked to sensory processing and conscious experience during NREM sleep. A significant group interaction demonstrated that individuals with PI and GS had opposite patterns of glucose metabolism in relation to SOL discrepancy. Future somnoimaging studies—studies that use neuroimaging techniques to investigate sleep states—in combination with interventions may further clarify the complex biology and clinical significance of SOL discrepancy.

SUPPLEMENTARY MATERIAL

Supplementary material is available at SLEEP online.

FUNDING

One of the protocols included in these analyses was supported by Sepracor, Inc. (PSE00001). That protocol contributed six participants to this study. The remaining protocols used in this study were supported by federal grants including the National Heart, Lung, and Blood Institute (HL65112) and National Institute of Mental Health (MH24652, MH61566).

WORK PERFORMED

Data were collected at the University of Pittsburgh and analyses were conducted at the University of Pittsburgh and Brigham Young University.

DISCLOSURE STATEMENT

EAN is on the Board of Directors and is Chief Medical Officer for Cerêve, Inc. DJB is a paid consultant to Cerêve, Inc., Emmi Solutions, and Merck. DBK was supported by T32HL082610 (PI: Buysse), AMS was supported by T32MH018269 (PI: Pilkonis), and BPH was supported by K01DA032557. The other authors have indicated no conflicts of interest.

ACKNOWLEDGMENTS

The authors thank Jean Miewald and Mary Fletcher for database management. The authors acknowledge with gratitude the dedicated work and technical skills provided by University of Pittsburgh staff at the Neuroscience Clinical and Translational Research Center, the Positron Emission Tomography Center, the Sleep and Chronobiology Faculty, the General Clinical Research Center, Sleep Imaging Research Program, and the Clinical Neuroscience Research Center. The authors thank many polysomnographic technologists who conducted the overnight sleep studies for the protocols reported in this paper including Dori Adams, Linda Bankson, Denise Duryea, Rachel Huff, Denny Knorr, Dan Limpert, Nancy Lutheran, Eric Miller, Jim Monahan, Kristen Page, Nicole Patton, Karen Quigley, Michael Quigley, Sarah Rankin, Steve Swanger, John Thase, Sarah Thase, Anne Vaniea, F. Jay Ver, and Monica Winkelman. We thank Jonathan Trout for data management and editorial comments.