Functional imaging of working memory in obstructive sleep-disordered breathing

Robert J. Thomas, Bruce R. Rosen, Chantal E. Stern, J. Woodrow Weiss, Kenneth K. Kwong


Functional magnetic resonance imaging was used to map cerebral activation in 16 patients with obstructive sleep-disordered breathing (OSDB) and 16 healthy subjects, during the performance of a 2-back verbal working memory task. Six patients with OSDB were reimaged after a minimum period of 8 wk of treatment with positive airway pressure. Working memory speed in OSDB was significantly slower than in healthy subjects, and a group average map showed absence of dorsolateral prefrontal activation, regardless of nocturnal hypoxia. After treatment, resolution of subjective sleepiness contrasted with no significant change in behavioral performance, persistent lack of prefrontal activation, and partial recovery of posterior parietal activation. These findings suggest that working memory may be impaired in OSDB and that this impairment is associated with disproportionate impairment of function in the dorsolateral prefrontal cortex. Nocturnal hypoxia may not be a necessary determinant of cognitive dysfunction, and sleep fragmentation may be sufficient. There may be dissociations between respiratory vs. cortical recovery and objective vs. subjective recovery. Hypofrontality may provide a plausible biological mechanism for a clinical overlap with disorders of mood and attention.

  • apnea
  • executive functions
  • imaging

executive functions are cognitive control processes that include flexibility in problem solving, planning, response inhibition, allocation of attention, maintenance and manipulation of information over time, and self-regulation of goal-directed behavior (18, 25). The prefrontal cortex is a critical although not exclusive site in the neural system that controls executive processing, which also includes the posterior parietal cortex and anterior cingulate cortex (17, 18). Subcortical mechanisms are also important in the regulation of executive processes; these include thalamic arousal influences and the caudate nucleus (40, 42).

Working memory is an important executive process used for temporary storage, active monitoring, updating, and manipulation of information (1). Impairments in executive functions, including working memory have been described in several disorders including attention deficit hyperactivity disorder (65), depression (54), and schizophrenia (10). After sleep deprivation, subjects demonstrate impaired executive control, such as an increased rate of forgetting, slow responses to simple mathematical calculations, and false responses during a vigilance task (68). It is thus possible that impaired working memory is an important abnormality in pathological sleepy states including obstructive sleep-disordered breathing (OSDB) or narcolepsy.

The focus of our neuroimaging investigation in patients with OSDB was the executive network that includes the dorsolateral prefrontal cortex, posterior parietal cortex, and anterior cingulate gyrus, because there is evidence of its vulnerability in states of inadequate or disrupted sleep (22, 29, 45). Sleep deprivation in young adults induces executive dysfunction (29). Patients with severe OSDB have longer P300 latencies than normal individuals (64). Postdeprivation recovery sleep is associated with a greater increase in frontal compared with nonfrontal delta power (9). Evaluation of regional cerebral blood flow after sleep deprivation shows prefrontal and posterior parietal hypometabolism (5). A positron emission tomographic study of sleep deprivation demonstrated reduced relative regional cerebral metabolic rate of glucose, predominantly in the thalamus and prefrontal and posterior parietal cortices (66). Thus tasks requiring activation in the executive network may reasonably be used as a probe of cognitive function in the sleepy state.

Normal dopamine neurotransmission is critical for maintaining wakefulness. Support for this conclusion includes the fact that most clinically active stimulants are associated with an increase in dopaminergic neurotransmission (52), and excessive sleepiness is common in Parkinson’s disease (26). Dopamine is a critical modulator of prefrontal and working memory function (7). We thus hypothesized that we would find disproportionate prefrontal rather than global impairments in activation in OSDB patients. This could reflect reduced activity and an inability to exhibit compensatory responses in the mesocortical dopamine system, a consequence of hundreds of nightly repetitive transient activations associated with individual arousals from sleep. We further hypothesized that patients with sleep fragmentation alone would demonstrate similar patterns of reduced activation as patients with hypoxic sleep-disordered breathing, because either clinical syndrome can demonstrate equivalent degrees of sleepiness (19).

A commonly used cognitive probe for functional imaging studies of working memory is the n-back task (6). In this task, the subject is told to respond whenever an item matches an item shown n items ago. We used a block design verbal 2-back working memory task to examine the prefrontal cortex in control subjects and patients with OSDB. Reasons for choosing this task at the time the study was designed included the following. 1) Extensive use of this and similar tasks in the neuroimaging literature provides considerable data regarding expected patterns of activation in health and disease. 2) Our own experience using this task in healthy subjects showed sensitivity to sleepiness effects. 3) Both attentional and executive task responses could be obtained during the same experimental runs. The blood oxygen-level dependent (BOLD) signal (37) was used to obtain and compare brain activation maps between patients and controls. Posttreatment evaluations were performed in a subset of subjects who fulfilled rigid criteria for successful positive airway pressure therapy.


The patients in this study were adults with OSDB with symptoms of at least 5-yr duration, aged 21–50 yr, 15 male and 1 female, free of medications or any other medical, neurological, or psychiatric disease, recruited from the Sleep Disorders Center at the Beth Israel Deaconess Medical Center. The minimum respiratory disturbance index (RDI, apneic + nonapneic respiratory obstructions) was 40 events/h of sleep, and all patients complained of debilitating excessive daytime sleepiness and/or fatigue. Sixteen (5 female) medication-free, nonsmoking subjects recruited for a separate protocol with no clinical evidence of sleep, medical, or psychiatric disease as assessed by an interview, physical examination, and a sleep-disorders evaluation template that included the Epworth sleepiness scale, formed a normative contrast group; the task used had some differences, as described below. Patients scanned for a minimum of 8 wk of positive airway pressure treatment also served as their own controls. All polysomnograms were performed in the Clinical Sleep Laboratory at the Beth Israel Deaconess Medical Center, Boston, and scanning was performed at the Athinoula A. Martinos Center for Biomedical Imaging, Charlestown, MA; the protocol was approved by both Institutional Review Boards. Written, informed consent was obtained from the subjects.

Polysomnography and scoring of respiratory events.

All patients underwent full sleep studies: electroencephalogram, electrooculogram, chin electromyogram, airflow (nasal pressure + thermistor), respiratory effort by plethysmography or piezo bands, intercostal electromyogram, bilateral anterior tibialis electromyogram, and oximetry. An apnea was scored when the airflow signal (both thermistor and nasal pressure) was ≤10% of a stable baseline for ≥10 s. Nonapneic respiratory events were scored when there was 1) any clearly visible reduction in either airflow signal (nasal pressure or thermistor), with an abrupt recovery of flow in association with an arousal or oxygen desaturation (≥2–3%), or 2) a flattened inspiratory flow profile (flow limitation) and respiratory event termination characterized by an abrupt return to a sinusoidal pattern of flow. A separate scoring of hypopneas associated with 4% desaturation was done. The RDI combined apneic and all nonapneic events. This included events characterized only by flow limitation and recovery breaths, capturing the entire burden of obstructive respiratory disease. The apnea-hypopnea index (AHI) combined apneic and nonapneic events associated with 4% desaturation, per hour of sleep. Respiratory effort was used primarily to exclude those with predominantly central sleep apnea. We considered two categories of arousals: 1) All alpha intrusions of ≥3 s were scored, according to standard guidelines, generating an arousal index regardless of associated respiratory abnormality. 2) When cued by respiratory flow reversals, time-linked EEG transients that included alpha durations of <3 s and bursts of K complexes or delta waves were also considered arousal equivalents. We have shown previously that the spectrum of EEG transients that are associated with respiratory event termination go beyond the standard 3-s rule (67).

Continuous and/or bilevel positive airway pressure titration and use.

The pressure setting that resulted in the best quality of sleep, a nonlimited airflow profile, and oxygen saturation >90% during the titration night was determined by attended polysomnography. Respiratory disturbance index on optimal pressure was <5 respiratory arousals per hour of sleep. Compliance (time on pressure) to therapy was monitored, and adequate use was defined as use of at least 7 h per night or 100% of total sleep time. Six of the nine patients agreed to participate in posttherapy scanning. The other treatment outcomes included tonsillectomy (1), maxillomandibular advancement (1), consideration of an oral appliance (1), inadequate use of positive airway pressure (3), and no treatment (patient’s choice) (1).

Cognitive testing (working memory) protocol.

A stream of random alphabets was presented in the center of the visual field every 4 s, 15 for each 60-s block; each stimulus lasted 500 ms. The first and second stimuli could only be nontargets but had to be retained in working memory to determine the nature of the third and subsequent stimuli. Thus the load on working memory was the ordering, retention, updating, and manipulation of two alphabets and consideration of its relationship with the third (new) presented alphabet, which could be a target (e.g., a m-f-m sequence) or a nontarget (e.g., a y-t-b sequence). One-third of the presented alphabets were targets. Although the memory component of the task was of identical difficulty and duration (2-back, 6 cycles of 1 min each), the baseline periods (1 min for patients, 30 s for controls) and total task duration (9.5 min for the healthy individuals, 13.5 min for patients) were different. Long-duration scans were initially chosen to try to demonstrate time-on-task decrements. The absence of this phenomenon in the scanning environment (perhaps secondary to noise, some discomfort with the confinement, and task switching during the block design) and the loss of data from movement motivated the evaluation of the shorter task duration, first in parallel with the longer duration in six patients. Because there was no difference in individual maps with this change and no loss from excessive motion, the shorter baseline was used for all subsequent testing. Only subjects who demonstrated high accuracy (>90%) during a pretesting scanning session were scanned, and subjects were encouraged to be as accurate as possible. During the “off” periods, vigilance was monitored and reaction times were determined to the discrimination of randomly occurring symbols + and *. During this task, either symbol was present continuously, and the requirement was to note a change to the other, which occurred at randomly occurring intervals but an average of every 5 s (twice in a 10-s period). The working memory component of the task required significant “online” maintenance and manipulation of the alphabet stream and sustained attention, whereas the baseline task required sustained attention. Early attempts to use a fixation baseline resulted in subjects falling asleep. A 3-back version of the task was tried and quickly discarded because no patient could perform it adequately pretreatment, unlike sleep-deprived healthy subjects. Each healthy subject performed the task twice, and patients two to four times; a total of 55 runs were obtained in the patients, and results of the two runs with the least motion were selected and averaged. Reaction times (for two runs 180 responses/subject for the 2-back task and 168 and 84 responses/patient or healthy subject, respectively, for the baseline attention task) were obtained by use of a scanner-compatible button box.

Functional imaging and data analysis.

Imaging was performed on a 3.0-T Siemens scanner. Oxygen saturations were recorded continuously during the scanning session both as a safety measure and to remove a confound of hypoxia-related changes in BOLD signal activity. Behavioral performance responses were monitored continuously to ensure patient safety and to discard runs in which the patients may have transiently fallen asleep. Structural images and echoplanar functional images were acquired over a 90- to 120-min session. The three-dimensional structural anatomical image had a 1-mm resolution and a 256-mm field of view. A gradient echo T2*-weighted sequence (repetition time: 2,000 ms, echo time: 30 ms, flip angle: 90°) was used to obtain BOLD contrast data. Functional data sets had 16 slices, each 5 mm with a 1-mm gap; 285 (30 s baseline for control subjects) or 380 (60 s baseline for sleep apnea patients) time points; a 200-mm field of view; and a 64 × 64 matrix. Inplane voxel x/y resolution was 3.125 mm. An identical-slice high-resolution T1-weighted scan was used as an intermediate step for functional overlays before transformation into the three-dimensional space, reducing the effect of echoplanar anatomical distortion. Slices were obtained approximately parallel to the anterior commissure-posterior commissure line. Preprocessing included motion correction, spatial smoothing using a 6-mm Gaussian filter for individual subject analysis and 10 mm for group averaging, linear trend removal, and temporal high-pass filtering using the Brain Voyager 2000 (Brain Innovation, Maastricht, The Netherlands) software package. Statistical correlation maps were generated by using a general linear model (24), as implemented within the software, and incorporated into the three-dimensional anatomic data set through interpolation of the functional voxels to the same resolution as the anatomic voxels (1 × 1 × 1 mm) (32). In the design matrix, the reference function was defined by convolving a boxcar with a gamma function. During the initial developmental phase of the analysis protocol, we evaluated a decrementing function that predicted a reduction of the BOLD signal to 50 or 25% of the starting values (time-on-task decrement of BOLD response); this modification did not result in an improved fit and was thus discarded. Stereotaxic transformation of data into a standardized space (Talairach) was performed for neuroanatomical localization of activated foci. After estimation of whole brain activation corrected for multiple comparisons (Bonferroni P ≤ 0.05), we tabulated volume of activation and mean percent BOLD signal (the signal time course over all activated voxels in the area was extracted and a mean value of percentage signal change computed by the software) from the following three areas: 1) dorsolateral prefrontal cortex [Brodmann area (BA) 46/9], 2) anterior cingulate cortex (BA 24), and 3) inferior parietal lobule in the posterior parietal cortex (BA 40). Given the normal variability in cortical anatomy, this localization is only approximate.

Further precautions were taken to minimize false positives. Data sets that had movement-correlated signals mimicking true activation, nonphysiological BOLD signal change amplitudes (≥5% increase), and extracerebral activation were discarded. The minimum activation cluster size should be set large enough to make it unlikely that a cluster of that size would occur by chance, assuming that falsely activated voxels should be randomly distributed. This relies on the assumption that areas of true neural signal activity tend to stimulate signal changes over contiguous voxels, and clusters of ≥200 voxels (0.2 cm3) were identified (23). Group analysis using random effects modeling was performed for both patients and controls (55). A contrast of hypoxic and nonhypoxic groups using fixed effects analysis was performed. Because the tasks had some difference (baseline duration) between the patients and controls, direct patient-control subtraction was not done. Activation maps had a pseudocolor code of correlations: red: r = 0.4, yellow: r = 0.8. The radiological convention was used, with the right side of picture = left side of brain.

Statistical methods.

Significance of differences in performance speed between baseline and performance blocks and between patients and healthy subjects were estimated using t-tests (STATA, Stata version 1). In an exploratory analysis, multiple regression with stepwise backward elimination was used to estimate correlations between task performance and RDI, body mass index (BMI), age, and desaturations. Linear relationships between functional activation (as assessed by percent BOLD signal change and number of activated voxels) and performance were estimated.


Subject characteristics.

Sixteen patients participated in pretreatment scans; 15 were male. The mean age was 40.3 ± 7.3 yr, the mean RDI 58.4 ± 15.7, and the mean Epworth score 12 ± 4. Mean BMI was 26.21 ± 1.83 kg/m2. Mean nocturnal minimum oxygen saturation was 84.7 ± 9.8%. Clinical craniofacial dysmorphism (retrognathia, high-arched palate, right-left cranial asymmetries), common in OSDB populations and associated subtle brain asymmetries, were noted in 10 of 16 patients. The mean age of the 16 healthy subjects was 37.6 ± 6.3 yr, and mean BMI was 23.9 ± 0.86. BMIs and ages (marginally) of patients were significantly higher relative to healthy subjects (t-test: P < 0.001 and 0.04). Fifteen of 55 task runs in patients were discarded because of excessive motion artifacts. A further eight runs were not used because there were more than two consecutive missed responses noted during the postscan review of performance, suggesting microsleep episodes. To examine the effects of nocturnal hypoxia, patients were also categorized into two groups, by using a 90% minimum desaturation cutoff, into nonhypoxic and hypoxic groups. These two groups did not differ significantly in mean Epworth scores (11.1 ± 4.5 vs. 13.25 ± 3.4), BMI (26.5 ± 2 vs. 25.9 ± 1.73 kg/m2), age (38.6 ± 8.7 vs. 42 ± 5.7 yr), or RDI (53.9 ± 16.7 vs. 62.9 ± 14.3), but on the basis of the group selection criteria they differed on the AHI and lowest nocturnal saturation (Table 1). Nine of 16 patients with obstructive sleep apnea successfully utilized positive airway pressure therapy, consistent with our strict criteria (7 h minimum or 100% use).

View this table:
Table 1.

Disease severity, functional activation, and performance: hypoxic vs. nonhypoxic disease

Pretreatment behavioral performance.

Healthy subjects and patients did well on the task, but patients had a significantly reduced percent correct, 94 ± 1.3 vs. 85.8 ± 4.1 (t-test: P < 0.001). As a group, mean performance speed in OSDB patients was significantly slower than controls on the 2-back task (patients: 908 ± 377 ms vs. controls: 596 ± 117 ms, t-test, P = 0.02), but not on the baseline symbol discrimination task (patients: 566 ± 203 vs. controls: 461 ± 62 ms, t-test, P = 0.13). Hypoxic and nonhypoxic groups did not differ (Table 1). There was no significant linear relationship between performance and age in patients or controls. In the exploratory multiple regression analysis, age, arousal index, sleep stages, RDI, AHI, BMI, percent BOLD signal increase in the posterior parietal cortex or anterior cingulate, and hypoxia severity did not significantly predict baseline or 2-back task performance.

Pretreatment activation maps.

All healthy subjects had activation of lateral and medial prefrontal and posterior parietal cortex; in patients, individual and group average map showed no dorsolateral prefrontal activation (Figs. 1 and 2; see Table 3). Posterior parietal activation was seen in all patients and healthy subjects, but the averaged spatial extent was reduced in patients. Medial wall activation in the region of the anterior cingulate activation occurred in 15 patients and all healthy subjects. Anterior cingulate and posterior parietal percent signal change and anterior cingulate activation volumes were not significantly different between hypoxic and nonhypoxic groups (Table 1) or in patients relative to healthy subjects (Table 2). There was no significant linear relationship between volumes of activation and baseline or 2-back performance. A group contrast map between hypoxic and nonhypoxic patients showed greater posterior parietal activation (Fig. 3, Table 3) in the latter group.

Fig. 1.

Individual activation maps of 3 sleep apnea patients and 3 age-matched (within 5 yr) healthy subjects. The well-documented individual differences in activation maps during functional imaging experiments are seen, but common themes can also be identified. Pretreatment (Pre) activation is seen in the posterior components of the executive network. Posttreatment (Post) activation in the anterior components of the network remains minimal. In contrast, the healthy subjects show activation in the lateral prefrontal and posterior parietal cortices. All activations are Bonferroni corrected for multiple comparisons, P < 0.05. Talairach space z-coordinate is 37. Statistical color scale: red, r = 0.4; yellow, r = 0.8.

Fig. 2.

Effects of sleep apnea on brain activation during a working memory task. Pretreatment random effects group activation map of 16 patients with obstructive sleep apnea, contrasted with 10 healthy subjects performing a similar working memory task. Activation in healthy subjects is more extensive, is bilateral, and involves areas well described in the literature: posterior parietal cortex, medial wall in the region of the anterior cingulate cortex, and dorsolateral prefrontal cortex. Sleep apnea patients do not show significant lateral prefrontal activation but do so in left posterior parietal and the medial wall. R, right; L, left. All activations are Bonferroni corrected for multiple comparisons, P < 0.05. Talairach space z-coordinate is 30. Statistical color scale: red, r = 0.4; yellow, r = 0.8.

Fig. 3.

Effects of hypoxia. Fixed-effects group map of 8 subjects with (H) and without (NH) nocturnal hypoxia demonstrates similar lack of prefrontal activation in both groups. A group difference (D) contrast map shows that nonhypoxic patients show greater posterior parietal activation. All activations are Bonferroni corrected for multiple comparisons, P < 0.05. Talairach space z-coordinate is 37. Statistical color scale: red, r = 0.4; yellow, r = 0.8.

View this table:
Table 2.

Functional activation and performance: disease vs. healthy

View this table:
Table 3.

Activated regions and relevant contrasts

Posttreatment clinical response, behavioral performance, and activation maps.

The clinical response to positive airway pressure treatment was complete in the studied subjects: daytime fatigue, sleepiness, and subjective sleep quality. Mean Epworth Sleepiness Scale scores decreased from 14.5 to 5 (P ≤ 0.001), and all symptoms attributable to poor sleep were resolved. There was no statistically significant difference in percent correct: 83.7 ± 84.6 (t-test: P = 0.41), 2-back performance (1,096 ± 320 ms pretreatment vs. 863 ± 288 ms posttreatment; P = 0.17) or baseline response speed (640 ± 178 ms pretreatment vs. 606 ± 146 ms posttreatment). A fixed-effects group activation map of the six subjects posttreatment showed no return of lateral prefrontal activation, with some new activation in posterior cortical areas including the parietal cortex (Fig. 4, Table 3).

Fig. 4.

Effect of treatment. Fixed-effects difference map for 6 patients after a minimum of 8 wk of treatment with positive airway pressure, associated with subjective resolution of symptoms. New areas of activation are seen especially in the posterior parietal cortex but not in the prefrontal areas, and objective performance remains unchanged. All activations are Bonferroni corrected for multiple comparisons, P < 0.05. Talairach space z-coordinate is 37. Statistical color scale: red, r = 0.4; yellow, r = 0.8.


We demonstrate absent dorsolateral lateral prefrontal activation in OSDB patients performing a working memory task regardless of the presence of nocturnal hypoxia. After positive airway pressure therapy, there was a complete subjective clinical recovery but activation remained impaired. In grouped averages, the extent and bilaterality of posterior parietal activation was also reduced relative to healthy subjects performing a very similar task, similar to the pattern recently reported with sleep deprivation using the Sternberg task (45). Thus prefrontal activation impairment was relative and not absolute, as other components of the executive network were also affected. Caudate activation was detected in our paradigm only on the contrast of pre- and posttreatment, with activation seen posttreatment. The percentage of correct responses was high, reducing the possibility that observed effects simply reflected inattentiveness. Although it is difficult to completely control for time, session, and practice effects, our subjects practiced the tasks until performance reached a stable state. The small number in this posttreatment set suggests caution but makes a case for further study of cerebral recovery in sleep apnea. In the exploratory multiple regression analysis, age, RDI, AHI, BMI, percent BOLD signal increase in the posterior parietal cortex or anterior cingulate, and hypoxia severity did not significantly predict baseline or 2-back task performance; this was not surprising because significantly larger numbers of subjects will likely be required given the variation in disease severity and possibly individual tolerance to sleep fragmentation similar to that well demonstrated after sleep deprivation (13, 69).

The nature and extent of executive (vs. attention-based) cognitive dysfunction in OSDB remain unsettled (4, 71). Meta-analytic reviews are confounded by variations in task sets, patient selection criteria, and disease severity (21). Short-term memory is part of the executive dysfunction that has been demonstrated, but there are very few data directly showing impairment in working memory. Naegele and colleagues (47, 48) showed that defects in short-term and working memory were not corrected by continuous positive airway pressure, whereas Lauer et al. (38) showed no significant difference between controls and patients in digit span or Corsi block tapping performance. Ferini-Strambi et al. (20) showed increased errors on the Stroop but not the backward digit span. Verstraeten et al. (71) found vigilance decrements and attentional capacity deficits (slowed information processing and decreased short-term memory span) but no specific clinical indications for executive attentional deficits, such as disinhibition, distractibility, perseveration, attentional switching dysfunction, decreased design fluency, or an impaired central executive of working memory. Naismith et al. (49) used principal components analysis to show that poorer sleep quality was related to slower processing speed, whereas nocturnal hypoxemia was related to visuoconstructional abilities, processing speed and mental flexibility. Lee et al. (39) demonstrated greater deficits in the retrieval of information from semantic memory (Controlled Oral Word Association task) and in shifting responses in the face of error (Wisconsin Card Sort Test), but differences in working memory were not observed. Our result using the 2-back task showed slowing of processing speed relative to attentional responding within the same task and is thus different from the general tone of working memory dysfunction reported to date. It is possible that the specific demands of the “back” task tap into dysfunction in executive control processes not identified by other tasks.

The potential mediators of sleepiness and executive dysfunction in OSDB include sleep fragmentation from repetitive arousals, hypoxia, some degree of chronic partial sleep deprivation, and cytokine effects (12, 35, 58, 72). Performance in a state of hypoxia, such as at high altitude, adversely impacts executive functioning (2, 31). Animal experiments support a role of intermittent hypoxia in the induction of sleepiness and working memory deficits (70). The clinical observation of persistent executive dysfunction despite use of positive airway pressure therapy may reflect inadequate compliance, residual flow limitation-related sleep fragmentation, or structural and functional alterations in brain neurocircuitry that may be slow to reverse (22). Subjective and objective scales of sleepiness correlate imperfectly with usually measured markers of disease severity such as arousal, desaturation, and respiratory indexes (36), and additional measures of disease impact on the brain such as functional neuroimaging could have practical clinical and experimental utility and help differentiate effects of sleep fragmentation and hypoxia. The division into hypoxic and nonhypoxic groups is somewhat arbitrary, and patients had a range of polysomnographic severity. They were, however, relatively homogenous in terms of clinical impact of the disease (sleepiness). It will be critical in future studies to evaluate patients who are restricted on severity measures as determined by respiratory indexes and related measures but demonstrate a range of symptoms. This could provide complementary information and possibly open new avenues to understand the neurobiology of sleepiness and individual differences in tolerating sleep fragmentation and nocturnal hypoxia.

Activation performance correlations are complex. In a study of thalamic influences, activity in the ventrolateral thalamus was modulated by the arousal state (57). The highest level of attention-related thalamic activity was seen under conditions of low arousal (after sleep deprivation) compared with high arousal (secondary to caffeine administration). In the neuroimaging literature, prefrontal activation has been reported to both increase (a possible marker of compensatory activity) and decrease in relation to slowing of task performance. Physiological activity is decreased during active depression in dorsal prefrontal cortical areas implicated in language, selective attention, and visuospatial or mnemonic processing, but these abnormalities reverse with symptom remission (14). Functional magnetic resonance imaging studies have demonstrated reduced lateral prefrontal activation in schizophrenia (3), in multiple sclerosis (areas not typically associated with working memory showed greater activation) (73), and during executive tasks after sleep deprivation (16, 28) but compensatory increases during verbal learning (15). Reports of activation increases may be confounded by circadian effects (8), because performance after 35–40 h of sleep deprivation could be in the zone of maximal circadian drive. A recent report of activation during a working memory task after 30 h of sleep deprivation, but with scanning done in the morning, showed predominantly a reduction in activation in the lateral prefrontal but even greater effects in the posterior parietal cortex (45). Task characteristics such as total and baseline vs. active block duration, degree of sleep deprivation, circadian phase, design (single trial vs. block design, response vs. process analysis, fixed vs. random effects,) and interindividual differences likely contribute to reported differences reported in the literature. Our results add to the growing database of functional imaging results in states of excessive sleepiness.

We consider possible reasons for the abnormality we have demonstrated and alternative explanations. The nature of the BOLD signal itself is controversial but appears to be correlated with local field potentials. Thus it represents an indirect measure of regional brain activity, and its alteration in our patients could be secondary to changes in regional task-specific neurocircuitry. Such examples are increasingly found in the recent functional neuroimaging literature, as in Parkinson’s disease and functional recovery after focal lesions (50, 60). There are other possible explanations for a reduction or loss of activation of specific brain structures on functional imaging tasks, including major structural neural loss, baseline blood flow and volume alterations, cerebrovascular reactivity, and altered regional catecholamine neurochemistry (51, 56, 59). Because those with hypoxic and nonhypoxic disease showed similar patterns of activation, it is less likely that effects of hypoxia or possibly hypercarbia on vascular reactivity, best described in association with altitude exposure (33, 34), were responsible.

Another potential factor is the task design (block); the baseline periods were not fixation but a symbol discrimination task that demanded sustained attention. In sleepy individuals, there are both attentional and executive functioning deficits. Because the baseline task required sustained attention, it may have been difficult for the patients, and there may have been less than average cognitive process differences between the baseline attentional and working memory tasks. The task had differences in duration and baseline periods between the patients and controls, which may have contributed to differences in activation. The longer task could have been much more fatiguing for the patients. Even though the baseline periods were slightly different and cerebrovascular reactivity was not directly tested, such as carbon dioxide inhalation responses (11), we believe the results showing differences between patients and healthy subjects are valid for the following reasons. 1) Healthy subjects show robust activation on tasks run as long as 30 min (unpublished observation). 2) Patients did show maintained posterior parietal and anterior cingulate activation during the course of the task. 3) Six sleep apnea patients had testing done with the shorter version of the task identical to that used in controls; the activation pattern was identical, in individual subjects, to that using the longer task. It is unlikely that the small age difference between controls and patients was responsible, because older individuals tend to show greater activation for a level of performance (62). The possibility of subtle anatomical differences between patients and control was considered, because a recent report demonstrated a possible loss of gray matter in multiple sites, including the prefrontal cortex in patients with severe OSDB (41). We performed voxel-based morphometric analysis on our data set (unpublished). The only positive result was a reduction in gray matter signal in the left hippocampus in the healthy subjects vs. hypoxic patients, which is similar to the conclusions of another report (44).

What are the implications of our results for the diagnosis and treatment of OSDB? The evidence supports roles for sleep fragmentation and hypoxia. For example, experimental auditory sleep fragmentation, which can be induced by using an auditory stimulus to disrupt sleep, impairs mood, decreases mental flexibility, and decreases sustained attention (43). Patients with hypoxemic chronic obstructive pulmonary disease show abnormal cognitive executive functioning, although the reasons could include sleep-disordered breathing and effects of hypercapnia; partial improvements have been noted after oxygen therapy (30). In tetraplegic patients, neuropsychological variables correlate with measures of sleep hypoxia (63). Murine models of cyclic hypoxia have shown clear evidence of sleepiness and learning deficits (61, 70). However, it is widely recognized that patients with symptomatic OSDB may have no hypoxia or may not have apneas and hypopneas as conventionally described. Thus our results of similarly impaired prefrontal activation maps irrespective of nocturnal hypoxia in consistent with the concept that sleep fragmentation alone may be sufficient to induce cognitive dysfunction. Further experiments would be required to evaluate the impact of chronic nocturnal hypoxia. In our small posttreatment sample, subjective clinical recovery was not matched by performance speed improvements, and functional activation in the lateral prefrontal cortex remained impaired. Our numbers are too small to make any definitive statements, but a longitudinal study of a larger group of patients would provide a more definitive answer regarding the extent and time course of recovery. This demonstration of decreased prefrontal activity could also be followed up by investigating the effect of adjunctive pharmacotherapy for those who have residual cognitive symptoms in spite of optimal clinical use of therapy (53). The demonstration of decreased prefrontal activity in OSDB may also provide a link between sleep disorders, depression, and attention deficit hyperactivity disorder (27).

In conclusion, we provide a direct demonstration of altered functional neurocircuitry in the lateral prefrontal cortex in patients with OSDB. Our preliminary results raise the possibility of tardy or incomplete neurobiological and objective performance as contrasted with subjective clinical recovery after treatment. Sleep fragmentation, the presumed impact of nonhypoxic OSDB, was sufficient to induce clinical and functional deficits, but an additional or independent role of hypoxia cannot be determined from this data set. The results provide a speculative although biologically plausible mechanism for the clinical overlap of obstructive sleep apnea with disorders of mood and attention.


Some of the healthy subject data were obtained as part of a separate study sponsored in part by Cephalon Inc.


  • The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.


View Abstract