Delayed fear extinction in individuals with insomnia disorder

Abstract

Study Objectives

Insomnia increases the risk for anxiety disorders that are also associated with fear-extinction deficits. We compared activation of fear and extinction networks between insomnia disorder (ID) without comorbidity and good sleepers (GS).

Methods

Twenty-three ID participants age- and sex-matched to 23 GS participants completed 14 days of actigraphy and diaries, three nights of ambulatory polysomnography and a 2-day fear conditioning and extinction paradigm. Fear conditioning and extinction learning occurred on the first day, followed 24 hours later by extinction recall. Blood-oxygen-level-dependent functional magnetic resonance imaging (fMRI) signal and skin conductance responses (SCR) were recorded. Nineteen participants per group produced usable fMRI data. Beta weights from areas where activation differed between groups were regressed against sleep and psychophysiological measures. SCR was compared between groups at various stages of the paradigm.

Results

During fear conditioning, both ID (N = 19) and GS (N = 19) activated fear-related structures. Across extinction learning, ID (N = 19) demonstrated little change, whereas GS (N = 16) activated both fear and extinction-related areas, including the hippocampus, insula, dorsal anterior cingulate (dACC), and ventromedial prefrontal (vmPFC) cortices. During extinction recall, while GS (N = 17) demonstrated limited activation, ID (N = 16) activated regions similar to those previously activated in GS (vmPFC, dACC, insula). Sleep quality was predictive of activations seen at various stages of the paradigm. SCR data suggested ID were more physiologically reactive than GS.

Conclusions

Across extinction learning, GS but not ID activated both fear and extinction-related networks. At extinction recall, ID engaged similar regions whereas GS no longer did so. Individuals with ID may show a delayed acquisition of fear extinction memories.

Statement of Significance

Insomnia and anxiety-related disorders each have been shown to increase the risk for the other. Abnormalities of fear extinction learning and memory and their neural substrates have been shown in anxiety, trauma, obsessive-compulsive, and related disorders. If similar abnormalities are present in persons with insomnia disorder (ID) alone, they may represent one mechanism whereby preexisting insomnia can increase the risk for developing pathological anxiety. Sleep disturbance might impact neural structures responsible for encoding, consolidating, or retrieving inhibitory learning. Such a putative pathway would highlight the importance of effectively treating ID and other forms of sleep disruption to reduce the risk of incident anxiety-related disorders and to optimize treatments, such as exposure therapy, that rely on therapeutic fear extinction.

Introduction

Insomnia Disorder (ID) is a common sleep disorder (with most estimates suggesting prevalence rates of 6–10% [1]) characterized by patient-reported difficulties in falling or staying asleep, frequent nighttime awakenings, chronic feelings of fatigue while awake [2–4], as well as significant distress and impaired daytime functioning (e.g. fatigue, mood disturbances, and cognitive impairments) [5]. Individuals with DSM-IV-TR Primary Insomnia [1] report the above sleep issues not attributable to another physical or mental condition. Primary Insomnia is equivalent to DSM-5 [6] ID without comorbidity and to the International Classification of Sleep Disorders (ICSD)-3 Chronic ID [7] without mental or physical comorbidity.

While healthy sleep promotes normal emotional regulation as well as the consolidation of emotional memories [8, 9], acute sleep disturbances and insomnia are pervasive following a variety of stressors and traumatic events [10, 11]. Further, epidemiological evidence suggests that preexisting insomnia may be a predisposing factor for the development of incident anxiety and traumatic stress disorders [10, 12–14]. Similarly, insomnia and rapid eye movement (REM) sleep disruption following a traumatic event may predict subsequent development of post-traumatic stress disorder (PTSD) symptoms [10, 15–18]. We have suggested that risk for pathological anxiety and disturbed sleep is bidirectional, that preexisting sleep disturbance may disrupt emotion regulatory circuits, and that the mutually reinforcing effects of sleep disruption and emotion-regulatory impairment can increase the probability that pathological anxiety will develop [10].

Experimental sleep deprivation and restriction in humans have been shown to interfere with both emotional regulation and the consolidation of emotional memory [8, 19]. The extinction of conditioned fear, i.e. learning that something once dangerous no longer need be feared, is a phylogenetically ancient emotion regulatory process that is thought to be impaired in individuals with certain anxiety and/or traumatic stress disorders [20, 21]. Rather than erase fear, extinction is a new memory that inhibits fear [22] and thus, like other memories, must be consolidated in order to persist [20]. In humans, sleep deprivation prior to extinction learning [19] and REM sleep deprivation after extinction learning [23] both impair extinction recall following a delay. Further, normal nocturnal sleep intervening between extinction learning and recall has been shown to promote generalization of extinction memory [24, 25], while a nap containing REM sleep, but not one without REM, has been shown to enhance extinction recall [26]. Nocturnal sleep, as well as naps, have been shown to promote extinction memory in exposure protocols designed to reduce phobic fear [24, 27]. Other sleep-related factors affecting extinction learning and memory include time-of-day [28] and morningness–eveningness [29] (for a review of sleep effects on fear and extinction see ref. [30]).

In humans, the amygdala, the dorsal anterior cingulate cortex (dACC), and parts of the insular cortex are regions known to show activation in response to presentations of fear-conditioned stimuli (CS) [31–35]. Conversely, fear extinction learning and extinction recall processes are associated with activations in the ventromedial prefrontal cortex (vmPFC) and hippocampus [20, 36, 37]. Anxiety and traumatic stress disorders have been associated with functional irregularities within these networks such that neural regions responsible for generating fear responses are overactive, and the activity of structures involved in inhibiting conditioned fear may be suppressed [20, 32, 33, 38], especially during retrieval of an extinguished fear memory [39–41].

There are growing numbers of reports on abnormalities in resting-state functional connectivity (rsFC) in insomnia (reviewed in refs. [42] and [43]). For example, a recent rsFC study demonstrated that ID, compared to good sleepers (GS), showed decreased rsFC between the left amygdala and bilateral regions of the rostral anterior cingulate cortex (rACC) [43]. This connectivity in ID was intermediate between GS and individuals with generalized anxiety disorder (GAD), suggesting a potential mechanism by which preexisting insomnia may increase vulnerability to GAD [43]. However, further understanding of the biology that might underpin abnormalities of fear learning and extinction in insomnia require task-based functional neuroimaging studies in which such stimuli are presented.

To the best of our knowledge, there are, to date, no task-based functional neuroimaging studies examining fear conditioning and extinction in persons with ID. Accordingly, this study sought 1) to compare regional activation during fear conditioning and extinction between GS and ID, and 2) to examine, in these groups, relationships between sleep measures and activation in fear- and extinction-related networks. We hypothesized that ID compared to GS might demonstrate enhanced activation of the neural circuitry involved in fear expression and lesser activation of the neural circuitry involved in fear extinction. Additionally, we expected that such fear and extinction-related differences in neural activation might be associated with objective or self-reported sleep quality deficits.

Methods

Overview

In a 2-week study, persons with DSM-5 ID without comorbidity and GS were compared with regard to their sleep as well as their psychophysiological and neural responses in a fear conditioning and extinction paradigm. Longitudinal objective (actigraphy) and self-reported (diary) sleep characteristics, three nights of ambulatory polysomnography (PSG), and psychological questionnaire responses were obtained from all participants. After a period of sleep monitoring, participants underwent a validated 2-day fear conditioning and extinction paradigm in an MRI scanner with simultaneous skin conductance level (SCL) recording [36]. Supplementary Figure S1 (Supplementary Methods) shows the study design.

Participants

Forty-six right-handed participants with ages ranging from 18 to 65 years (mean = 29.30, SD = 12.25, see Table 1) were recruited from the Boston area. Prospective participants completed a phone screening to preliminarily rule out exclusion criteria such as: history of neurological, major medical, psychiatric, or sleep disorders (other than insomnia for the ID group); current use of psychoactive or sleep medications, recreational drugs or tobacco; alcohol > 10 drinks/week or caffeine > 5 beverages/day; shift work; and MRI contraindications. All study procedures accorded with the Declaration of Helsinki and were approved by the Partners Healthcare Institutional Review Board. All participants provided written informed consent and were paid for their participation.

Table 1.

Group means and standard deviations of demographic and sleep characteristics (Total Sleep Time, Sleep Efficiency, Sleep Onset Latency, and Sleep Midpoint) among total sample of ID (N = 23) and GS (N = 23)

	ID	GS	P	Effect size (Cohen’s d)
Demographics
Number of participants	23	23
Number of males	6	5
Age (years)	28.96 ± 11.95	29.65 ± 12.81	0.85	−0.06
Shock tolerance (mA)	2.14 ± 0.99	2.09 ± 0.98	0.88	0.05
Sleep questionnaires
PSQI	10.43 ± 2.41	1.48 ± 1.10	<0.0001*	4.80
ISI	18.95 ± 4.01	1.04 ± 1.33	<0.0001*	6.00
ESS	9.09 ± 5.70	4.17 ± 3.08	0.001*	1.07
MEQ	50.04 ± 9.34	52.57 ± 7.70	0.32	−0.29
NEO PI-R: Neuroticism	76.61 ± 18.74	67.73 ± 19.10	0.12	0.47
Actigraph data
Total sleep time (min)	383.63 ± 60.04	392.62 ± 37.88	0.56	−0.18
Sleep efficiency (%)	82.35 ± 4.41	84.96 ± 5.54	0.09	−0.52
Sleep onset latency (min)	22.37 ± 16.32	13.46 ± 9.78	0.06	0.66
Sleep midpoint (min)	232.61 ± 58.24	235.12 ± 11.70	0.33	−0.03
Sleep diary data
Total sleep time (min)	402.92 ± 84.50	451.11 ± 43.87	0.037*	−0.72
Sleep efficiency (%)	85.18 ± 8.29	96.11 ± 2.15	<0.0001*	−1.80
Sleep onset latency (min)	34.86 ± 17.00	11.69 ± 6.36	<0.0001*	1.80
Sleep midpoint (min)	237.03 ± 59.35	221.03 ± 57.79	0.37	0.27
PSG data†
Total sleep time (min)	389.00 ± 63.67	377.50 ± 75.31	0.58	0.16
Sleep efficiency (%)	86.56 ± 7.52	88.68 ± 7.03	0.38	−0.29
Sleep onset latency (min)	28.87 ± 29.49	17.36 ± 14.79	0.15	0.49

ID, individuals with insomnia disorder without comorbidity specifier; GS, good sleepers; PSQI, Pittsburgh Sleep Quality Index; ISI, Insomnia Severity Index; ESS, Epworth Sleepiness Scale; MEQ, Morningness-Eveningness Questionnaire; NEO PI-R, NEO Personality Inventory-Revised; PSG, polysomnography.

*p < 0.05, independent t-test.

^†PSG data averaged from baseline and consolidation nights of sleep recording. Reported for subsets of ID (N = 16) and GS (N = 15) individuals.

Baseline assessments and sample size

Participants passing a preliminary telephone screening attended an in-person psychiatric evaluation using the Structured Clinical Interview for DSM-IV for Non-Patients (SCID 1/NP) [44] and a sleep disorders interview using the Pittsburgh Structured Clinical Interview for Sleep Disorders (unpublished, in-house instrument). Individuals assigned to the ID group met criteria for DSM-IV-TR Primary Insomnia, DSM-5 ID without comorbidity, and ICSD-3 Chronic ID. ID scored ≥13 (and all but two were ≥15) on the insomnia severity index (ISI) [45] and >5 on the Pittsburgh sleep quality index (PSQI) [46], while GS scored ≤4 on the ISI and ≤3 on the PSQI. All accepted participants were right-handed scoring ≥70 on the modified Edinburgh Handedness Inventory [47] and showed negative results on a urine toxicology screen testing for drugs of abuse as well as tricyclic antidepressants, amphetamines, barbiturates, and benzodiazepines. Forty-six participants met study eligibility criteria, with 23 individuals (17 females) assigned to the ID group and 23 (18 females) to the GS group. After neuroimaging data were acquired and assessed for image quality (detailed below), 19 (14 females) in the ID group and 19 (14 females) in the GS group remained for analyses.

Sleep monitoring period

Qualified participants completed a 14-day sleep-assessment period using wrist actigraphy and sleep diaries. During this time, they completed the first (acclimation/diagnostic) of three nights of ambulatory PSG. The second PSG (baseline) took place on the night before their first MRI session, and the third PSG (consolidation) took place on the night between their first and second MRI session. During the sleep monitoring period, participants also completed the Morningness–Eveningness Questionnaire (MEQ) [48], the Epworth Sleepiness Scale (ESS) [49], and the Revised NEO Personality Inventory (NEO-PI-R) [50]. Consumption of alcohol and recreational drugs was prohibited throughout the study period, and caffeine and daytime naps were additionally prohibited on the day prior to MRI session 1 as well as during the 2 days on which scans occurred.

Sleep monitoring: actigraphy and diary

Participants wore the Actiwatch-2 (Phillips Respironics, Bend, OR) continuously during the sleep evaluation period. Participants were instructed to press an event-marker when beginning to attempt sleep and when waking for the day. These time stamps were used to demarcate time-in-bed. Within this period, the default algorithm of Actiware 5.61 software determined total sleep time (TST), sleep onset latency (SOL), and sleep efficiency (SE: TST in minutes divided by total time in bed; Table 1). Sleep midpoint was computed as the midpoint between sleep onset and final awakening (expressed as minutes past midnight). When button presses were missed, times recorded on diaries were substituted.

The Evening–Morning Sleep Questionnaire (EMSQ) [51, 52] served as a sleep diary throughout the sleep evaluation period. The EMSQ queries self-reported TST and SOL as well as sleep quality and specific sleep difficulties. Diary-based SE and sleep midpoint were calculated in the manner described for actigraph above.

Sleep monitoring: ambulatory PSG

Ambulatory PSG used the Somte-PSG monitor (Compumedics USA, Charlotte, NC). A standard montage was used, and additional details are provided in Supplementary Methods. Sleep architecture, including sleep-stage percentages, REM latency, REM density, and REM fragmentation, were computed from scored records (see Supplementary Methods for computation details).

Fear conditioning and extinction paradigm

After the sleep monitoring period, participants completed a validated 2-day/2-session fear conditioning and extinction paradigm [36, 53] in one of two Siemens 3T MRI scanners. Both scanning sessions took place between 1700 and 2200 with session 2 taking place approximately 24 hours after session 1. This paradigm consisted of four phases—habituation, fear conditioning, and extinction learning (session 1), and extinction recall (session 2). SCL was recorded continuously throughout all phases.

Digital images of three colored lamps (red, blue, or yellow) presented in two different rooms (“contexts”) were projected onto a viewing screen inside of the MRI scanner. The unconditioned stimulus (US) consisted of a mild electric shock delivered to the index and middle fingers of a subject’s right hand. Prior to task initiation, participants were exposed to increasing intensities of mild electric finger shocks and selected a level that was “highly annoying but not painful” [54].

During habituation, participants were exposed to all six possible CS color and context combinations. During fear conditioning, 2 of 3 colored lamps (CS+) were presented eight times each in one of the contexts (Conditioning Context) and were followed by the US on a partial reinforcement schedule (5 of 8 presentations of each CS+; 63%), while the third colored lamp that was never followed by a shock (CS-) was interspersed among the CS+ presentations. Fear conditioning was immediately followed by extinction learning, during which one of the two CS+s (extinguished CS+ [CS+E]) was presented 16 times without reinforcement along with 16 interspersed presentations of the CS- in the second context (Extinction Context). The second CS+, therefore, remained conditioned but unextinguished (CS+U). The following day, during the extinction recall phase, all three CSs (CS+E, CS-, CS+U) were presented without reinforcement in the Extinction Context. At the conclusion of each phase, each participant was asked to rate, on a scale of 1 to 5, the degree to which they expected to be shocked at the first two and last two presentations of each differently colored CS. Details of this protocol have been widely published [31, 36, 37, 39–41, 55, 56] and also appear in Supplementary Methods.

MRI image acquisition

Whole brain images were acquired using a 32-channel head coil in either a 3T Siemens MAGNETOM Trio or Prisma scanner (Siemens Medical Systems, Iselin, NJ). An automated scout image was acquired, and shimming procedures were conducted to optimize field homogeneity. High-resolution 3D T1 MEMPRAGE sequences (TR/TE1,2,3,4/flip angle = 2530 ms/1.64,3.5,5.36,7.22 ms/7° for Trio and TR/TE1,2,3,4/flip angle = 2530 ms/1.69,3.55,5.41,7.27 ms/7° for Prisma) were collected with an in-plane resolution and slice thickness of 1.0 mm. Following these anatomical scans, functional magnetic resonance imaging (fMRI) blood oxygenation level dependent (BOLD) images were acquired with an interleaved T2*-weighted EPI sequence. Forty-eight slices per TR for the Trio scanner and 46 slices per TR for the Prisma scanner were obtained with the following parameters: 2.5 mm slice thickness, 3 × 3 mm in-plane resolution, TR = 2560 ms, TE = 30 ms, flip angle = 90°. The same parameters were used throughout both days of scanning.

fMRI data analysis

MATLAB v2014a (The Mathworks Inc, Natick, MA) and Statistical Parametric Mapping (SPM8; Wellcome Trust Centre for Neuroimaging, www.fil.ion.ucl.ac.uk) were used to process BOLD images and perform statistical analyses on fMRI data. Functional images underwent slice timing, realignment, co-registration with the structural images, normalization into MNI space, and smoothing with an 8-mm full-width half-maximum Gaussian kernel to increase the signal-to-noise ratio and account for between-subject anatomical variation. Low-frequency signal drift was removed with a high-pass temporal filter with a 128-second cutoff. Serial correlations from aliased biorhythms in the fMRI time series were accounted for with an autoregressive AR (1) model. The Artifact Detection Toolbox (ART, http://gablab.mit.edu) detected motion spikes in the functional temporal data. These motion artifact data were used as regressors in the first-level analysis with movement parameters from the realignment process (x, y, z, roll, pitch, and yaw). Within each phase, participants’ scans exhibiting movement greater than 3 mm translation or 3° rotation were excluded from further analyses. For remaining participants, the ART tool generated motion regressors that were applied to all first level analyses in SPM with conditions of movement = 1.0 mm and rotation = 0.087 radian. After exclusions were applied, there remained totals of 38 participants (19 ID and 19 GS) for the fear conditioning phase, 35 participants (19 ID and 16 GS) for the extinction learning phase, and 33 participants (17 ID and 16 GS) for the extinction recall phase that were suitable for further fMRI analyses.

Condition-events were modeled using canonical hemodynamic response function to the reinforced (shocked) CS (CS+) and the unreinforced CS (CS-). Following the extinction learning phase, the CS+ were further differentiated into CS+E and CS+U. Within each experimental phase, “Early” refers to the average neural responses to the first four presentations of a given stimulus (CS+, CS-, CS+E, CS+U) and “Late” refers to the average neural responses to the last four presentations of a given stimulus. The “>” symbol in a given contrast (e.g. CS+ > CS-) indicates the subtraction (in SPM8) of neural activation to the latter stimulus (CS-) from neural activation to the former stimulus (CS+). Thus “Early CS+ > CS-” describes the subtraction of average neural activation seen in response to the first four CS- presented (whose order corresponds to the first four CS+) from the average neural activation seen in response to the first four CS+ presented. Such differences (Early CS+ > CS-) represent differential neural responses associated with the initial fear acquisition that can then be compared between groups.

To examine fear conditioning-induced activations, we looked at the neural activation at various stages of fear conditioning. Specifically, the Early CS+ > CS- contrast, as noted, showed regions associated with initial fear acquisition. Similarly, the Late CS+ > CS- contrast identified those regions associated with greater response to (fear of) the reinforced (shocked) vs. un-reinforced CS once differential conditioning was established. The Early CS+ > Late CS+ contrast, in turn, revealed those regions that reflected the increase in fear of the 2 CS+ as fear conditioning to these stimuli was established by their pairing with shocks.

To examine neural activations during extinction learning, we selected the following contrasts: Early CS+E > CS-, Late CS+E > CS-, and Late CS+E > Early CS+E. The Early CS+E > CS- contrast identified regions that continued to reflect greater response to (fear of) the previously reinforced (CS+) than the unreinforced CS (CS-) at the beginning of fear extinction learning. Similarly, the Late CS+E > CS- contrast identified regions that continued to be more activated by the reinforced (CS+) than the unreinforced (CS-) after extinction learning had taken place. The Late CS+E > Early CS+E contrast, in turn, revealed areas which, over the course of extinction learning, increased their response to the reinforced CS (CS+E).

Finally, to assess activational changes involved in extinction recall, we examined the following contrasts: Early CS+E > CS-, Early CS+U > CS-, and Early CS+E > CS+U. The Early CS+E > CS- contrast identified regions that continued to respond more to the reinforced CS that, 24 hours previously, was extinguished following fear conditioning (CS+E) in comparison to the CS that was never reinforced (CS-). Similarly the Early CS+U > CS- contrast identified regions that continued to respond more to the reinforced CS that, 24 hours previously, remained un-extinguished following fear conditioning (CS+U), again in comparison to the CS that was never reinforced (CS-). The CS+E > CS+U contrast, in turn, revealed regions more responsive to the CS that was both reinforced and subsequently extinguished 24 hours previously (CS+E) compared with the CS that was previously only reinforced (fear conditioned) and left un-extinguished (CS+U). Thus, this contrast differentiates regions more activated by the recall of both conditioning and extinction than recall of fear conditioning alone. Note that only Early responses were analyzed because, in the absence of reinforcement during extinction recall, additional (CS+E) and new (CS+U) extinction learning took place as this phase proceeded.

The aforementioned contrasts were specified during first-level processing and were modeled at the second-level using a general linear model and random effect analysis. Our main goal was to investigate differences in the activation of the fear expression and fear extinction networks between ID and GS. To accomplish this, second-level analyses were first carried out for each of the above contrasts within each group. The within-group results for each contrast were then compared between the two groups. Both within-group and between-group analyses used age, sex, and scanner type (Trio or Prisma) as covariates to account for any potential confounds. We then conducted a region-of-interest (ROI) analysis, focusing primarily upon structures that have been implicated in fear conditioning and extinction by prior studies: the amygdala, the anterior cingulate cortex (rACC and dACC), the vmPFC, the hippocampus, and the insular cortex. ROIs for the amygdala, hippocampal gyrus, and insular cortex were created using the Automated Anatomical Labeling atlas (AAL, http://qnl.bu.edu/obart/explore/AAL/) [57] through the interface of WFU PickAtlas software (http://fmri.wfubmc.edu/software/PickAtlas) [58]. AAL also offered anatomical definitions for Brodmann areas (BA) 11, 24, 25, 32, and 35. However, these defined areas did not fully encompass the vmPFC and dACC. Therefore, based on a meta-analysis of 298 fear-related studies included in the NeuroSynth database (www.neurosynth.org) [59], two vmPFC ROIs with coordinates at [6, 40, −20] and [0, 16, −18] were identified and defined with 6 mm spheres. ROIs for the dACC ([0, 24, 22] and [0, 14, 28]) were identified and created in a similar manner. Accordingly, a composite anterior limbic/paralimbic ROI mask was constructed using AAL-defined regions for the amygdala, hippocampus, insular cortex, and BA 11, 24, 25, 32, 35 as well as spheres for the vmPFC and dACC identified via NeuroSynth.

In our analyses, we specified a cluster-determining threshold of p_uncorrected < 0.005 and a cluster-size threshold of ≥10 contiguous voxels. Clusters detected within these regions that survived small volume family-wise error (FWE) correction (p_{cluster_FWE} < 0.05) were included in further analysis [60]. Beta weights (representative of BOLD effect) from surviving clusters that showed significant between-group differences were extracted using the REX (Region of Interest Extraction) toolbox (http://web.mit.edu/swg/software.htm).

SCL recording

SCL was recorded using the MP150 system (BIOPAC Systems, Inc., Goleta, CA) with MR-compatible transducer, cables, and electrodes. SCL was recorded at 37.5 Hz using two disposable 11-mm, Ag/AgCl sensors filled with isotonic paste attached 14 mm apart on the hypothenar surface of the left hand. Stimuli were presented using SuperLab 4.5 (Cedrus Corp., San Pedro, CA) which produced event marks allowing precise synchronization of each stimulus onset with the ongoing SCL recording. For each CS presentation, skin conductance response (SCR) was computed as mean SCL in microSiemens (μS) during the last 2 seconds of context presentation subtracted from the maximum SCL generated during CS presentation. SCRs were square-root-transformed and, if the untransformed SCR was negative, the negative sign was retained after calculating the square root of the SCR’s absolute value [54]. A participant was deemed a “non-conditioner” and was excluded from further analyses using SCR if they did not exhibit two or more non-square-root-transformed SCR responses to a CS+ that were equal to or greater than 0.05 µS during fear conditioning.

Contrast-based SCR comparisons and shock expectancy

After the exclusion of non-conditioners from SCR analyses, group sizes were as follows: fear conditioning (ID = 15, GS = 16), extinction learning (ID = 15, GS = 16), and extinction recall (ID = 14, GS = 13). Comparisons of square-root-transformed SCRs were conducted using mixed model analysis of variance (ANOVA) with one between subject factor: ‘group’ (GS, ID) and either ‘CS type’ (CS+ vs. CS-) or ‘timing’ (Early vs. Late) as a within-subject factor. These comparisons were conducted in a manner mirroring contrasts used in analysis of neuroimaging data. As done in fMRI analyses, for SCR within a particular phase, “Early” was defined as mean SCR to the first four presentations (or second to fourth presentation during early fear conditioning because US pairing had not yet occurred for the first) of a particular type of CS while “Late” refers to mean SCR to the last four presentations. And as done in fMRI contrasts, to examine fear conditioning-induced SCR between groups, we looked at the SCR in response to: Early CS+ vs. CS-, Late CS+ vs. CS-, and Early CS+ vs. Late CS+. To examine SCR during extinction learning between groups, we selected the following comparisons: Early CS+E vs. CS-, Late CS+E vs. CS-, and Late CS+E vs. Early CS+E. Finally, to examine SCR group differences in extinction recall, we selected the following comparisons: Early CS+E vs. CS+U, Early CS+E vs. CS-, and Early CS+U vs. CS-. Significance was set at alpha = 0.05, and the Greenhouse-Geisser correction was applied to all within-subject effects. Effect sizes of ANOVA model effects are partial η² (pη²) with small, medium and large effect sizes of 0.0099, 0.0588, and 0.1379 respectively [61].

For shock expectancy, mixed ANOVAs for each phase included the between-subject factor ‘group’ as well as two within-subject factors: ‘order’ (first two, last two) and ‘CS type’. For fear conditioning, CS type included CS- and average ratings for CS+1 and CS+2, for extinction-learning CS type included CS- and CS+E ratings, and for extinction recall this within-subject factor included CS-, CS+E, and CS+U.

Objective (SCR) and self-reported (shock expectancy) indices of conditioning, extinction learning, and extinction recall

In order to examine relationships between neural activation and processes of fear conditioning, extinction learning, and extinction memory using regression (see Statistical analyses [fMRI BOLD, sleep factors, PSG, SCR, and shock expectancy]), unitary indices of the magnitude of each process were computed from objective (SCR) and self-reported (shock expectancy) data. The methods of computing each of these indices are described in Supplementary Methods. Objective and self-reported measures for fear conditioning were designated conditionability index (CONDidx) and subjective conditionability index (subCONDidx), respectively; for extinction learning—extinction learning index (EXTidx) and subjective extinction learning index (subEXTidx); and, for extinction recall—extinction retention index (ERI) [36] and subjective extinction retention index (subERI). Self-reported indices for each phase utilized participants’ ratings of shock expectancy for the first and last two presentations of the CS+s at each experimental phase.

Sleep characteristics—approach and analysis

Independent sample Student’s t-tests were conducted to measure the difference between ID and GS with regard to sleep, psychophysiological, and self-rated variables. Due to a high degree of overlap between variables (e.g. diary-based TST and actigraph-based TST), principal components analysis (PCA; SPSS version 24 for Macintosh) was used to minimize collinearity of sleep characteristics (Actigraph TST, SE, SOL, and sleep midpoint and Diary TST, SE, and SOL) and to reduce the number of variables used for subsequent analysis. Three factors emerged—Objective Sleep Difficulties (factor 1), Self-reported Sleep Success (factor 2), and Sleep Duration (factor 3). Additional details on PCA methodology are provided in Supplementary Methods.

Statistical analyses (fMRI BOLD, sleep factors, PSG, SCR, and shock expectancy)

These analyses were performed using SPSS version 24 for Macintosh. Stepwise regressions were conducted to predict activation in the extracted beta weights using sleep factors identified using PCA. Stepwise multiple linear regressions were conducted to predict extracted beta weights from between-group differences using Objective Sleep Difficulties, Self-reported Sleep Success, Sleep Duration, and PSG-based REM latency, REM fragmentation, REM density, Stage N1%, N2%, N3%, and REM% (values were averaged across baseline and consolidation nights; see Supplementary Table S1). The stepwise methodology used p = 0.05 to enter a variable into each step and p = 0.1 to remove. Results from stepwise regressions underwent Bonferroni corrections.

Simple regressions examined associations between beta weights from between-group contrasts and scores on the MEQ, ESS, and NEO-PI Neuroticism. In the ID group alone, simple regressions examined associations between these same beta weights and the ISI and PSQI. Simple regressions also examined associations between beta weights of between-group contrasts and objective (SCR) and self-reported (shock expectancy) indices of fear conditioning, extinction learning, and extinction memory.

Results

General characteristics

Demographic, shock tolerance, sleep quality (from actigraphy, sleep diaries, and PSG), and questionnaire results are compared between ID and GS in Table 1. ID and GS participants did not significantly differ in age, sex ratio, or in the level of shock they tolerated as “highly annoying but not painful.” As expected, ID participants scored significantly higher than GS on the PSQI, ISI, and ESS and reported lower diary TST and SE, and greater diary SOL (Table 1). Actigraph SOL and SE also showed trends in the predicted direction (Table 1). However, ID and GS did not significantly differ in MEQ, NEO-PI Neuroticism, Actigraph TST and midpoint, or PSG sleep quality (TST, SE, SOL; Table 1) or architecture (Supplementary Table S1). Additionally, there were no significant between-group differences in objective or self-reported indices of fear conditioning, extinction learning, or extinction recall (Supplementary Table S2).

Fear conditioning

Contrast-based SCR and shock expectancy

SCR data from 15 ID and 16 GS participants were analyzed. For ANOVA comparing CS+ and CS- at early fear conditioning (Figure 1A), there was no significant main effect of group (ID vs. GS; p = 0.14), a significant main effect of CS type [F(1,29) = 35.36, p < 0.0001, pη² = 0.549], and no significant group × CS type interaction (p = 0.56). However, the group main effect showed a medium effect size [F(1,29) = 2.34, p = 0.14, pη² = 0.075] suggesting that, with a larger sample, overall larger SCR in ID (Figure 1A) might become significant. For ANOVA comparing CS+ and CS- at late fear conditioning (Figure 1B), there was again no significant group main effect (p = 0.11), a significant main effect of CS type [F(1,29) = 10.79, p = 0.003, pη² = 0.271] and no group × CS type interaction (p = 0.64). However, again, the group main effect showed a medium effect size (F(1,29) = 2.788, p = 0.11, pη² = 0.087). For the comparison of Early CS+ vs. Late CS+, there was no significant main effect of group (p = 0.21), a significant timing main effect [F(1,29) = 24.32, p < 0.0001 pη² = 0.456] and no significant group × timing interaction (p = 0.68).

Figure 1.

Mean contrast-based SCR by group and stimulus type. (A) SCR (CS+ vs. CS-) at Early Conditioning. (B) SCR (CS+ vs. CS-) at Late Conditioning. (C) SCR (CS+ vs. CS-) at Early Extinction. (D) SCR (CS+ vs. CS-) at Late Extinction. (E) SCR (CS+E) at Early and Late Extinction. (F) SCR (CS+E vs. CS+U) at early extinction recall. Error bars represent standard error of the mean. μS^1/2: square-root-transformed SCR in microSiemens.

For shock expectancy, there was no main effect of group (p = 0.41) or interactions of group with CS type (p = 0.11) or with CS type × order (p = 0.65). However, a significant group × CS type interaction [F(1,42) = 4.27, p = 0.045, pη² = 0.092] reflected the fact that, collapsing across CS type, shock expectancy increased across fear conditioning in ID but decreased in GS. Notably, a highly significant CS type × order interaction [F(1,42) = 47.48, p < 0.0001, pη² = 0.531] indicated successful fear conditioning. This interaction was also seen in the 13 individuals who failed to show SCR evidence of conditioning [F(1,11) = 7.97, p = 0.017, pη² = 0.420].

Neuroimaging

Neuroimaging data from 19 ID and 19 GS participants were analyzed (whole brain analyses are reported in Supplementary Results Table S3). For the Early CS+ > CS- contrast during fear conditioning, both ID and GS showed activation of neural nodes of the fear network (dACC, bilateral insula) as well as the left hippocampus. Between-group comparisons for this same contrast revealed that left anterior insular cortex activation was greater in ID compared to GS, whereas left middle/posterior insular cortex activation was lesser in ID compared to GS (Table 2; Figure 2, A and B).

Table 2.

Fear conditioning: brain regions of activation from within-group and between-group analyses (ID: N = 19, GS: N = 19) for contrasts at p < 0.05 and with small volume FWE corrections

		Brain regions	Side	Coordinates (mm)			Cluster size	T value	p value FWE
				x	y	z
Early CS+ > Early CS-	ID	dACC	C	3	6	32	38	3.13	0.018
		rACC	C	−2	46	2	37	3.2	0.018
		Anterior insular cortex	L	−34	18	−6	428	5.81	0.0001
			R	36	18	−8	129	3.58	0.01
		Hippocampal gyrus	L	−34	−16	−14	19	3.85	0.004
			R	26	−12	−22	36	4.09	0.001
	GS	dACC	C	−2	20	26	82	3.71	0.025
		Insular cortex	L	−30	20	8	740	5.67	0.0001
			R	40	20	6	62	3.27	0.042
		Hippocampal gyrus	L	−18	−26	−10	29	3.53	0.005
	ID > GS	Anterior insular cortex	L	−34	10	−14	16	3.19	0.022
	GS > ID	Middle/posterior insular cortex	L	−36	−2	14	11	2.99	0.015
Late CS+ > Late CS-	ID	Anterior insular cortex	L	−28	24	8	94	3.93	0.007
		Middle insular cortex	L	−40	16	6	64	3.22	0.024
	GS	dACC	C	−8	12	34	18	2.97	0.017
	ID > GS	None significant
	GS > ID	None significant
Early CS+ > Late CS+	ID	Middle cingulate cortex	R	10	−20	42	18	3.74	0.006
		Anterior insular cortex	L	−32	14	−12	59	3.01	0.02
			R	34	16	−18	154	4.15	0.005
		Hippocampal gyrus	L	−22	−22	−12	53	3.43	0.005
			R	22	−6	−20	94	3.09	0.02
		Amygdala	R	30	−2	−18	36	3.41	0.016
	GS	None significant
	ID > GS	None significant
	GS > ID	None significant

ID, individuals insomnia disorder without comorbidity; GS, good sleepers; dACC, dorsal anterior cingulate cortex; rACC, rostral anterior cingulate cortex; C, center; L, left; R, right. Montreal Neurological Institute (MNI) coordinates, cluster size, and T value are also reported for each area of interest.

Figure 2.

Fear conditioning. Group differences in brain activation evoked by the contrast CS+ > CS- during early fear conditioning. Persons with ID showed greater activation in the left anterior insula cortex (A) than GS, while GS showed greater activation in the left middle insular cortex than ID (B).

For the CS+ > CS- contrast during late fear conditioning, ID showed brain activation in only the left insular cortex, and GS showed significant neural activation in only the dACC. However, for this contrast, there were no significant between-group differences.

For the Early CS+ > Late CS+ contrast, ID showed activation of the right middle cingulate cortex (mCC), bilateral anterior insular cortex, bilateral hippocampal gyrus, and right amygdala; whereas for GS there were no significant activations. However, for this same contrast, there were no between-group differences (Table 2).

Stepwise regression, across both groups, showed only one association between PCA sleep characteristics and beta weights for clusters reflecting group differences. Decreased Self-reported Sleep Success was predictive of the increased activation in the left anterior insular cortex for the Early CS+ > CS- (ID > GS) contrast (ß = −0.657, p = 0.025, R² = 0.125; Figure 3A). However, this association did not survive a Bonferroni correction and was not seen within ID alone (ß = 0.721, p = 0.214, R² = 0.036) or GS alone (ß = 0.789, p = 0.422, R² = −0.017). Simple regressions showed no associations between group-difference beta weights and scores on the ESS, MEQ, and Neuroticism across both groups, between these beta weights and ISI and PSQI scores among ID alone, or with CONDidx and subCONDidx across both groups.

Figure 3.

Stepwise regressions using sleep characteristics to predict beta weights between ID and GS. (A) Early fear conditioning. Self-reported sleep success predicted decreased beta weights for a region in the left anterior insular cortex where the Early CS+ > CS- contrast showed greater activation in ID than GS. (B) Late extinction learning. Self-reported sleep success predicted increased beta weights for a region in the right hippocampal gyrus where the CS+E > CS- contrast showed lesser activation in ID than GS. (C) Across extinction learning. Objective sleep difficulties predicted decreased beta weights for a region in the right middle insular cortex where the Late CS+E > Early CS+E contrast showed lesser activation in ID than GS. (D) Early extinction recall. Objective sleep difficulties predicted increased beta weights for a region in the left ventral anterior insular cortex where the Early CS+E > Early CS- contrast showed greater activation in ID than GS. (E) Early extinction recall. Self-reported sleep success predicted decreased beta weights for a region in the left middle insular cortex where the Early CS+E > Early CS- contrast showed greater activation in ID than GS. *p < 0.05. **p < 0.01.

Extinction learning

Contrast-based SCR and shock expectancy

SCR data from 15 ID and 16 GS participants were analyzed. For ANOVA comparing CS+E and CS- at early extinction learning (Figure 1C), there was no significant main effect of group (p = 0.15), a main effect trend for CS type [F(1,29) = 3.27, p = 0.081, pη² = 0.101], and no significant group × CS type interaction (p = 0.53). Note, however, that the group main effect showed a medium effect size [F(1,29) = 2.16, p = 0.15, pη² = 0.069] suggesting that, with a larger sample, overall larger SCR in ID (Figure 1C) might become significant. For ANOVA comparing CS+E and CS- at late extinction learning (Figure 1D), there was no significant group main effect (p = 0.47), a main effect trend for CS type [F(1,29) = 3.03, p = 0.093, pη² = 0.095], and no group × CS type interaction (p = 0.16). For the comparison of Early CS+E vs. Late CS+E (Figure 1E), there was no significant main effect of group (p = 0.14), a significant timing main effect [F(1,29) = 4.70, p = 0.038, pη² = 0.139], and no significant group × timing interaction (p = 0.19). Note, however, that the group main effect showed a medium effect size [F(1,29) = 2.87, p = 0.14, pη² = 0.073] suggesting that, with a larger sample, overall larger SCR in ID might become significant.

For shock expectancy, there was no main effect of group (p = 0.93) or interactions of group with CS type, order or CS type × order (all ps > 0.41). A significant order × CS type interaction [F(1,42) = 19.22, p < 0.0001, pη² = 0.314] illustrated that successful extinction of the CS+E had taken place.

Neuroimaging

Neuroimaging data from 19 ID and 16 GS participants were analyzed (whole brain analyses are reported in Supplementary Results Table S4). For the CS+E > CS- contrast during early extinction learning, ID showed no significant activations, whereas GS showed increased activation in vmPFC (gyrus rectus); however, there were no significant between-group differences (Table 3). For this same contrast during late extinction learning, there were no significant activations within either group. However, comparing groups for this contrast, ID showed lesser brain activation in the right hippocampal gyrus (Table 3, Figure 4A).

Table 3.

Extinction learning: brain regions of activation from within-group and between-group analyses (ID: N = 19, GS: N = 16) during the contrasts at p < 0.05 and with small volume FWE corrections

		Brain regions	Side	Coordinates (mm)			Cluster size	T value	p value FWE
				x	y	z
Early CS+E > Early CS-	ID	None significant
	GS	vmPFC (gyrus rectus)	R	12	34	−22	12	4.06	0.006
	ID > GS	None significant
	GS > ID	None significant
Late CS+E > Late CS-	ID	None significant
	GS	None significant
	ID > GS	None significant
	GS > ID	Hippocampal gyrus	R	38	−24	−8	14	2.86	0.01
Late CS+E > Early CS+E	ID	None significant
	GS	dACC	C	6	36	26	26	3.38	0.031
		rACC	C	−2	42	10	80	3.67	0.017
		Middle insular cortex	R	50	4	−2	106	3.9	0.002
		Orbitofrontal cortex	R	26	44	−8	30	3.82	0.012
	ID > GS	None significant
	GS > ID	dACC	C	−6	24	34	16	2.94	0.013
		Middle insular cortex	R	48	12	−10	13	3.07	0.015

ID, individuals with insomnia disorder without comorbidity; GS, good sleepers; vmPFC, ventro-medial prefrontal cortex; dACC, dorsal anterior cingulate cortex; rACC, rostral anterior cingulate cortex; C, center; L, left; R, right. Montreal Neurological Institute (MNI) coordinates, cluster size, and T value are also reported for each area of interest.

Figure 4.

Extinction learning. (A) Group differences in brain activation evoked by the CS+E > CS- contrast during late extinction learning. Persons with ID showed lesser activation in the right hippocampal gyrus. (B) Group differences illustrated in brain activation evoked by the Late CS+E > Early CS+E contrast. Persons with ID showed lesser activation in the dACC. (C) Group differences illustrated in brain activation evoked by the Late CS+E > Early CS+E contrast. Persons with ID showed lesser activation in the right middle insular cortex.

For the contrast of the Late CS+E > Early CS+E, activations were seen in GS for the rACC, dACC, right middle insular cortex, and right orbitofrontal cortex (OFC). However, no significant clusters appeared for this same contrast in ID (Table 3). When ID and GS were compared for this same contrast, ID showed lesser activation in the dACC (Figure 4B) and right middle insular cortex (Figure 4C; Table 3).

Stepwise regression models demonstrated that an increase in the Self-reported Sleep Success factor predicted increased activation in the right hippocampal gyrus for the Late CS+E > CS- (GS > ID) contrast (ß = 0.707, p = 0.002, R² = 0.245; Figure 3B), although this association was not observed within ID (ß = 0.648, p = 0.177, R² = 0.063) or GS (ß = −0.056, p = 0.928, R² = −0.076) alone. Additionally, Objective Sleep Difficulties predicted decreased activation in the right middle insular cortex for the Late CS+E > Early CS+E (GS > ID) contrast (ß = −3.48, p = 0.001, R² = 0.294; Figure 3C). Simple regression showed that this association remained within GS (ß = −3.116, p = 0.044, R² = 0.220) but not ID (ß = −1.982, p = 0.222, R² = 0.044). Simple regressions showed no associations between group-difference beta weights and scores on the ESS, MEQ, and Neuroticism across both groups, between these beta weights and ISI and PSQI scores among ID alone, or with EXTidx, and subEXTidx across both groups.

Extinction recall

Contrast-based SCR and shock expectancy

SCR data from 14 ID and 13 GS participants were analyzed. For ANOVA comparing CS+E and CS- at early extinction recall, there was no significant main effect of group (p = 0.50), a significant main effect for CS type [F(1,26) = 6.67, p = 0.016, pη² = 0.204], and no significant group × CS type interaction (p = 0.41). For ANOVA comparing CS+U and CS- at early extinction recall, there was no significant main effect of group (p = 0.59), a significant main effect for CS type [F(1,26) = 4.40, p = 0.046, pη² = 0.145], and no significant group × CS type interaction (p = 0.62). For ANOVA comparing CS+E and CS+U at early extinction recall, there was no significant main effect of group (p = 0.65) or CS+ type (p = 0.37) and no significant group × CS+ type interaction (p = 0.58; Figure 1F).

Expectancies were analyzed only for the “first two” level of order and showed no main effect of group (p = 0.29) or group × CS type interaction (p = 0.31). A significant main effect of CS type [F(2,76) = 8.84, p < 0.0001, pη² = 0.331] was seen with post-hoc means contrasts showing CS+E > CS- (p = 0.0005), CS+U > CS- (p < 0.0001), and CS+U > CS+E (p = 0.032), illustrating the maintenance of differential conditioning across the extinction retention interval for both CS+s. In addition, expectancies showed that extinction learning differentially reduced shock expectancy for the CS+E vs. CS+U.

Neuroimaging

Neuroimaging data from 16 ID and 17 GS participants were analyzed (whole brain analyses are reported in Supplementary Results Table S5). In ID, early recall of a previously extinguished conditioned stimulus (CS+E) compared to an unextinguished conditioned stimulus (CS+U) revealed increased activation of the vmPFC and right OFC (Table 4). This same contrast showed no significant activations in GS. In addition, comparison between groups for this same contrast showed greater vmPFC activation in ID compared to GS (Table 4, Figure 5A). In ID, recall of Early CS+E compared to the corresponding unreinforced stimulus (CS-) revealed increased activation of bilateral anterior and posterior insula as well as the left amygdala and left hippocampal gyrus (Table 4). This same contrast showed no significant activations in GS. When Early CS+E > CS- was compared between groups, ID showed greater activation of dACC (Figure 5B), left amygdala and left middle insula (Figure 5C), bilateral posterior insula (Figure 5D), and left anterior insula (Figure 5E; Table 4). Few within- or between-group activations were seen for the Early CS+U > CS- contrast (Table 4).

Table 4.

Early extinction recall: brain regions of activation from within-group and between-group analyses (ID: N = 16, GS: N = 17) during the CS+E > CS+U, CS+E > CS-, and CS+U > CS- contrasts at p < 0.05 and with small volume FWE corrections

		Brain regions	Side	Coordinates (mm)			Cluster size	T value	p value FWE
				x	y	z
CS+E > CS+U	ID	vmPFC	C	−4	36	−14	41	3.65	0.014
		Orbitofrontal cortex	R	26	36	−24	10	4.36	0.009
	GS	None significant
	ID > GS	vmPFC	C	−2	36	−14	15	2.78	0.024
	GS > ID	None significant
CS+E > CS-	ID	Posterior insular cortex	L	−40	4	6	105	3.6	0.009
			R	36	−22	10	14	3.16	0.027
		Anterior insular cortex	L	−30	14	−16	44	4.19	0.01
			R	34	20	12	33	3.68	0.016
		Amygdala	L	−24	−8	−16	31	3.76	0.016
		Hippocampal gyrus	L	−26	−10	−12	13	3.85	0.008
	GS	None significant
	ID > GS	dACC	C	−2	24	30	23	3.07	0.025
		Middle insular cortex	L	−36	2	8	73	4.51	0.001
		Posterior insular cortex	L	−38	−18	14	21	3.33	0.015
			R	34	−20	14	13	3.45	0.003
		Anterior insular cortex	L	−32	14	−18	17	2.93	0.015
		Amygdala	L	−22	0	−24	12	3.2	0.019
	GS > ID	None significant
CS+U > CS-	ID	None significant
	GS	Middle insular cortex	L	−38	14	8	24	3.92	0.013
	ID > GS	None significant
	GS > ID	None significant

ID, individuals with insomnia disorder without comorbidity specifier; GS, good sleepers; vmPFC, ventro-medial prefrontal cortex; dACC, dorsal anterior cingulate cortex; C, center; L, left; R, right. Montreal Neurological Institute (MNI) coordinates, cluster size, and T value are also reported for each area of interest.

Figure 5.

Extinction recall. (A) Group differences in brain activation evoked by the CS+E > CS+U contrast during early extinction recall. Persons with ID showed greater activation in the ventromedial prefrontal cortex (vmPFC). (B) Group differences in brain activation evoked by the CS+E > CS- contrast during early extinction recall. Persons with ID showed greater activation in the dACC. (C) Group differences in brain activation evoked by the CS+E > CS- contrast during early extinction recall. Persons with ID showed greater activation in the left amygdala and left middle insular cortex. (D) Group differences in brain activation evoked by the CS+E > CS- contrast during early extinction recall. Persons with ID showed greater activation in bilateral posterior insular cortex. (E) Group differences in brain activation evoked by the CS+E > CS- contrast during early extinction recall. Persons with ID showed greater activation in the left anterior insular cortex.

Increased Objective Sleep Difficulties predicted increased neural activation for the Early CS+E > CS- contrast in the left ventral anterior insular cortex (ß = 0.720, p = 0.03, R² = 0.127; Figure 3D). This association did not survive Bonferroni correction, although it remained within GS (ß = 1.278, p = 0.029, R² = 0.265) but not ID (ß = 0.790, p = 0.378, R² = −0.014). For this same contrast, activation in the left middle insular cortex was also predicted by a decrease in the Self-reported Sleep Success factor (ß = −1.16, p = 0.0001, R² = 0.337; Figure 3E), although this association was not observed within ID (ß = 0.301, p = 0.733, R² = −0.108) or GS (ß = 0.334, p = 0.558, R² = −0.048) alone. Within ID alone, higher scores on the ISI predicted increased activity in the vmPFC for the Early CS+E > CS+U (ID > GS) contrast (R = 0.59, p = 0.016). Greater daytime sleepiness (higher ESS scores) was associated with greater activity in the vmPFC for the Early CS+E > CS+U (ID > GS) contrast across all participants (R = 0.449, p = 0.009), but not among ID (R = 0.486, p = 0.057) or GS (R = −0.071, p = 0.787) alone. Higher ESS scores were also associated with increased activity in the left middle insular cortex for the Early CS+E > CS- (ID > GS) contrast across all participants (R = 0.376, p = 0.034), but not among ID (R = 0.082, p = 0.77) or GS (R = 0.177, p = 0.497) alone. Furthermore, in this same contrast, higher scores on the MEQ were associated with decreased activation in the left anterior insular cortex among all participants (R = −0.422, p = 0.016), but not among ID (R = −0.441, p = 0.1) or GS (R = −0.202, p = 0.437) alone. Greater ERI was associated with lesser activation of the left anterior insula for the Early CS+E > CS- contrast across all participants (ß = −0.025, p = 0.045, R² = 0.195), though this relationship was not observed within ID (ß = −0.013, p = 0.263, R² = 0.032) or GS (ß = −0.003, p = 0.664, R² = −0.053) alone. Neither Neuroticism nor SubERI were associated with group-difference beta weights.

Discussion

We investigated differences among the neural correlates of fear learning, extinction learning, and extinction recall between persons with ID without comorbidity and GS. We then examined the relationship of these functional differences with sleep characteristics and psychophysiological indices of fear conditioning and extinction. Psychophysiological variables (SCR) were analyzed to examine the peripheral expression of conditioned fear and its extinction. Shock expectancies were also analyzed to examine declarative mnemonic awareness of the conditioning and extinction contingencies. Our hypotheses—(1) that ID would demonstrate enhanced activation of fear circuitry and lesser activation of extinction circuitry; and (2) that these differences would correlate with sleep quality deficits—were partially but not entirely supported. During fear conditioning, ID and GS demonstrated activity in neural regions commonly associated with the expression of conditioned fear. In the extinction learning phase, ID demonstrated virtually no significant neural activity in any selected contrast, whereas GS activated neural regions implicated in both fear expression and extinction. The following day during early extinction recall, ID exhibited significant neural activation in both fear expression and extinction networks, similar to activations displayed by GS during late extinction learning, whereas GS showed little activation to selected contrasts. Regional activations (beta weights) were correlated with sleep characteristics in a manner that suggested that poorer sleep quality was associated with greater activation of areas involved in fear expression. SCR data suggested that overall psychophysiological reactivity to CSs might be greater in ID than GS whereas shock expectancy data suggested similar awareness of and memory for contingencies across phases between the groups.

Fear conditioning

During early fear conditioning, both ID and GS demonstrated activation of regions involved in the expression of conditioned fear—namely, the insula and dACC. The activation of these regions in the Early CS+ > CS- contrast is consistent with previous studies [31, 62] and suggests that both ID and GS were learning to associate the CS+ with the US. Between-group comparisons for this contrast revealed that, in early fear conditioning, GS were activating more of the middle/posterior left insula, whereas ID were activating more of the anterior left insula. Poorer self-reported sleep quality (Self-reported Sleep Success factor) was associated with increased activity in the left anterior insula. In late fear conditioning, both ID and GS continued to activate certain nodes of the fear expression network in the CS+ > CS- contrast (left anterior and middle insula, and dACC, respectively). Notably, from early to late fear conditioning (Early CS+ > Late CS+), ID exhibited a reduction in activation of fear-related structures (anterior insula and amygdala) whereas no such reduction was seen in GS. However, because groups showed no significant difference in activation for this contrast (Early CS+ > Late CS+), activation may have been higher for ID during early fear conditioning, diminishing over time to resemble GS.

The posterior insula is known to play a role in receiving and integrating interoceptive information [63–65] and is associated with the coordination of sensory aspects of pain [66]. Multimodal interoceptive information is then re-represented in the middle insula, where it is assimilated with emotionally salient environmental stimuli [64, 67]. From there, the anterior insula—which has reciprocal connections to limbic regions such as the ACC, vmPFC, and amygdala—integrates autonomic and visceral information into emotionally-salient “self-reported feeling states” [63]. Hyperactivation of the anterior insular cortex is a well-established finding in neuroimaging studies of individuals with anxiety disorders in fear-related experimental contexts [32, 33, 68, 69] and is postulated to be associated with anticipated interoceptive feeling [56]. Findings from a recent high-density EEG study on ID and interoception suggest that insomnia is associated with altered cortical responses to interoceptive signals, potentially implicating excessive cognitive processing of, or hypersensitivity to, interoceptive signals in ID [70]. Based on these and other studies linking the insula to fear conditioning [34, 35, 67], we theorize that ID may be demonstrating enhanced anticipation of an aversive event and recasting this awareness as an emotional state via the anterior insula, whereas GS may be associating interoceptive information with fear conditioning cues, but without this engendering an anticipatory emotional feeling state. This possibility is supported by group comparisons of psychophysiological data. The medium effect size for the main effect of group (i.e. to CS+ and CS- combined) on SCR during early and late fear conditioning suggests that ID may be overall more reactive than GS. Interestingly, shock expectancy data also showed overall expectancy (i.e. to CS+ and CS- combined) increased over fear conditioning in ID but decreased in GS.

Extinction learning

During early extinction learning, GS showed increased activation in the vmPFC for the CS+E > CS- contrast. The vmPFC is thought to be involved in fear extinction learning and consolidation of fear extinction memory [20, 21, 36, 37, 71, 72]. Although this activation did not significantly differ between groups, GS may have begun to inhibit the conditioned fear response earlier during the extinction learning phase. In late extinction learning, ID compared to GS demonstrated lesser neural activation in the right hippocampal gyrus to the CS+E > CS- contrast. The hippocampus acts as a contextual modulator of fear and extinction memory and plays a pivotal role in identifying safe and “dangerous” contexts [73, 74]. Sleep has been shown to be especially important to hippocampus-dependent learning and memory [75], including initial memory encoding [76]. Therefore, the more poorly sleeping ID individuals may have been impaired in their ability to detect changes in context, potentially resulting in poorer extinction learning than GS. This conclusion is supported by regression analyses suggesting that, in areas less active in ID than GS, a decrease in the Self-reported Sleep Success factor predicted decreasing activation in the right hippocampal gyrus to the Late CS+E > CS- contrast (Figure 3B). Thus, individuals reporting worse sleep may have been impaired in their ability to facilitate context-dependent fear extinction learning.

Across extinction learning (Late CS+E > Early CS+E), GS exhibited simultaneous increases in activation of fear expression network nodes (e.g. insula, dACC [32]) as well as in frontal areas associated with top–down emotion regulation (e.g. rACC, medial superior frontal cortex [77–79]). In contrast to GS, this dual activation of fear expression and regulatory networks across extinction learning was absent in ID. The synchronous activity in GS of fear expression and emotion (fear) regulation networks across extinction learning may indicate dynamic competition between the fear-conditioning memory and the newly learned fear-extinction memory. Increased Objective Sleep Difficulties predicted decreased activation of the right middle insula across all participants and maintained this trend within GS participants alone, providing further evidence for the role of healthy sleep in extinction learning. The fact that overall SCR (i.e. to CS+ and CS- combined) during Early Extinction was greater in ID (Figure 1C), albeit nonsignificantly but with a medium effect size, suggests group differences in reactivity were also expressed peripherally.

Extinction recall

In early extinction recall, for the CS+E > CS+U contrast, ID demonstrated greater neural activation in the vmPFC than GS. As noted earlier, the vmPFC is a region consistently implicated in fear extinction learning and recall [20, 36, 37]. Additionally, for this same contrast, ID but not GS showed enhanced activation in regions associated with fear expression (e.g. amygdala, dACC, and insula). Thus, the hypothesized competition between fear expression and fear extinction that occurred in GS during extinction learning may have been delayed in ID until extinction recall. For the CS+E > CS- contrast during early extinction recall, increased Objective Sleep Difficulties predicted increased activation in the left anterior insular cortex (a “limbic sensory area” [63] responsive to anticipated threat) [56]. Additionally, increased activation in the left middle insula was predicted by a decrease in the Self-reported Sleep Success factor. Thus, although the above results may at first seem paradoxical, what may be occurring is delayed onset of fear extinction processes in poorly sleeping individuals.

Sleep has been hypothesized to enhance consolidation of fear extinction [30] as well as provide a moderating effect on negative emotion [8]. Impairment of sleep’s moderating effect on negative emotion might allow greater consolidation of conditioned fear and thus greater resistance to extinction [30]. Indeed, recent evidence suggests insomnia may interfere with emotion regulation. For example, a recent population-based study suggests a longitudinal association between declining emotion regulation and insomnia [80]. Additionally, Wassing and colleagues [81] reported that individuals with more disrupted REM sleep (estimated using a self-report proxy measure) showed poorer overnight resolution of emotional distress and that both disrupted REM and poorer emotional resolution were associated with severity of insomnia symptoms. Furthermore, in healthy participants, REM sleep has been hypothesized to “depotentiate” previously salient, negatively valenced stimuli [9]. In a recent study, it was shown that gamma power in REM, a putative proxy for elevated levels of the stress- and fear-related neuromodulator norepinephrine, varied inversely with the degree of overnight reduction in amygdala activation to such salient stimuli [82]. Interestingly, the present study did not observe the association of neural activation with REM sleep parameters (REM percent, latency, density, and fragmentation) that have been reported in studies of healthy individuals [26, 83–85]. This may result from reduced statistical power due to the limited sample size and the relatively large number of voxels compared between groups in our anterior limbic/paralimbic mask. However, regions from which beta weights were extracted for correlation with sleep parameters were those that differentially activated between GS and ID to CSs, a criterion differing from those used in selecting ROIs in healthy controls.

It is notable that the majority of significant associations with neural activations we report during the experimental phases were seen for the Self-reported Sleep Success factor rather than the Objective Sleep Difficulties factor (with the exception of extinction recall) or polysomnographic sleep-architecture measures. This may reflect the well-known discrepancy between objective and self-reported sleep quality in ID [2–4]. Interestingly, this may also reflect the association between insomnia and anxiety-related disorders (reviewed in ref. [10]), wherein characteristic cognitive distortions of both conditions (e.g. catastrophic thinking) may exacerbate symptoms. Thus, results from the current study indicate that at least self-reported, and possibly also objectively poor, sleep quality may be associated with better preservation of conditioned fear and impaired consolidation of extinction learning [10, 30], perhaps reflecting a deficiency in sleep’s moderating effect on negative emotion [8, 9].

Limitations

The small sample size increases the probability for type II error, as does the relatively large number of contrasts and correlations for type I error. Although factor analysis reduced the number of predictors in stepwise regressions, partially mitigating these concerns, the sample size remained less than ideal. Similarly, the well-known phenomenon of certain participants failing to condition as measured by SCR [86] further reduced the numbers available to correlate with psychophysiological parameters. The latter was partially mitigated by having parallel analyses using shock expectancies that demonstrated conditioning even in those who failed to show conditioning using SCR. Analyses comparing SCR conditioners (N = 23) and SCR non-conditioners (N = 15) in fear conditioning demonstrated that there were no group differences in neural responses, with the exception of SCR conditioners exhibiting greater activation of a small region of the dACC in Early CS+ > CS- compared to non-conditioners (Supplementary Table S7). One possible explanation for this sole difference is the fact that the dACC is a region that activates in concert with SCRs generated by a variety of causes [87] and, thus, may be a nonspecific correlate of SCR production. As the first study to examine neural responses to fear conditioning and extinction in ID, we report associations of the full complement of psychophysiological, sleep, and self-report variables with these responses with the awareness that further studies are required to replicate findings. In addition, predominance of female participants limits generalizability of findings, since fear extinction learning and extinction recall are thought to be influenced by gonadal hormones such as estrogen [62]. Future studies should document gonadal hormone levels and menstrual cycle phase to accurately account for the potential influence of sex differences. Because of a facility upgrade, two different scanners were used at different times in the experiment. Although we controlled for differences between scanners by adding the scanner type as a covariate in our analyses, there remains the possibility that scanner type had an effect on the observed results. The extraction of beta weights only from regions where ID and GS differed for correlation with sleep variables poses the risk that observed relationships reflect the grouping variable (diagnosis) rather than true correlation. However, the observation of similar within-group correlations with these same beta weights indicates that such a confounding factor was not always present (Supplementary Table S6).

Conclusions

It may be the case that healthy individuals extinguish fear at a “normal” rate, whereas individuals with insomnia extinguish at a slower rate. Alternatively, stronger fear conditioning may account for resistance to and hence a delay of extinction. In either case, individuals experiencing anxiety, mood, and trauma-related symptoms (which may include sleep difficulties) may extinguish at an even slower rate, accounting for evidence showing greater activation of fear expression networks during extinction learning and recall in such disorders [20, 21, 39, 40]. Thus, as in the case of resting state studies [43], these task-based findings suggest a mechanism (viz. delayed extinction of conditioned fear) whereby, over time, ID may predispose an individual to develop pathological anxiety. Further research is required to determine how sleep may impact fear extinction across a range of severities of mood, anxiety, and trauma-related symptoms.

Supplementary material

Supplementary material is available at SLEEP online.

Funding

This project was supported by National Institutes of Health (NIH) grants MH101567 and MH 109638 to E.F.P-S. and the 2016 Harvard Mind/Brain/Behavior Interfaculty Initiative (E.F.P-S., Co-PI).

Notes

Conflict of interest statement. None declared.