Main

Previous studies have suggested that deficient arousal responsiveness occurs in apnea of infancy and may also be important in the pathogenesis of SIDS (1,2). However, how arousal should be defined and identified in infants is controversial. The central issue of this controversy is identification of the minimal behavioral or electrophysiologic events that should be regarded as significant in the arousal process. It is now clear that arousal as indicated by overt change in EEG activity (3) is not a prerequisite for recovery from apnea in adults, children, or infants (4–7). Recent reports suggested that sighs (6) (i.e. augmented breaths) and startles, generated in the brainstem, appear at the onset of behavioral arousal in infants during rebreathing (8), and also during internal (6) or external airway obstruction, even in the absence full behavioral arousal (9). However, it remains unclear if infantile sighs and/or startles are closely correlated with the electrophysiologic changes marking initiation of arousal in the brainstem.

In the present work we have used changes in sleep spindle activity as an electrophysiologic indicator of arousal onset. In invasive animal studies it has been demonstrated that either spontaneous arousal or localized stimulation of brainstem arousal centres coincides with termination of sleep spindles (12-14 Hz). The blockage of spindle rhythms reflects a "desynchronization" of neuronal discharge in the thalamocortical system. Such desynchronization reflects onset of arousal and is essentially different from the arousal related appearance of "synchronized" oscillations in higher EEG frequencies (10–12).

We hypothesized that, if sighs and startles reflect the initial events in a sequential arousal from sleep process in infants, then they might be associated with a transient blocking of EEG sleep spindles, because activation of the primary arousal mechanisms in the brainstem RE activating system eliminates spindle activity generated in the thalamus. In the present study we evaluated spontaneous sighs and startles as well as those evoked by a respiratory stimulus (airway occlusion). Our findings indicate that sighs and startles are closely related to sleep spindle suppression in terms of simultaneous occurrences and a correlation between startle magnitude and duration of spindle suppression.

METHODS

Subjects. We studied 12 normal healthy infants, five female and seven male, between 10 and 19 wk of age (mean 101.5 d ± 34 SD) during daytime naps. Two infants were preterm (34 and 36 wk of gestational age) but had no history of perinatal complications. These infants were studied at a postconceptional age of 11 and 13 wk. Because EEG sleep spindle activity in preterm infants has appeared by 11 wk postconceptional age (13,14), we included these infants in our study. Recordings started after feeding during a time, the infants usually were sleeping. The study was approved by the Washington University Committee for human studies. Informed consent was obtained from the parents, one of whom was present during the recordings.

Polygraphic recordings. Studies were performed shortly after a feeding. Infants slept in their usual sleeping position (five prone, seven supine) during the study, which lasted between 2 and 3 h. In prone-sleeping infants the infant's airway was never covered by bedding in a way that could cause rebreathing of expired air.

Polygraphic recordings included EEG (C3-P3), electrooculogram (outer canthi-A1) and ECG (Beckman R 611 8 channel polygraph). For the EEG a bandpass of 0.5-100 Hz was maintained. Surface EMG of the right arm (musculus biceps brachii and musculus triceps brachii) and of the diaphragm (electrode position on the right chest wall, subcostal, and near the medial clavicular line was obtained using a four-channel EMG amplifier (Coulbourn S75-04, Lehigh Valley, PA) and fiberoptic preamplifiers (Coulbourn S75-04B) with a band-pass filter setting of 50-3000 Hz (Coulbourn S75-36). Abdominal and thoracic breathing movements were obtained using the inductive plethysmography technique with bands placed around the chest and abdomen (Respitrace System; Ambulatory Monitoring, Inc., Ardsley, NY). Oxygen saturation was measured at the right toe (Pulseoxymeter; Nellcor, Inc., Hayward, CA). All infants were constantly observed for activity as well as videotaped (JVC Professional Products, Elmwood Park, NJ) with an infrared camera, and simultaneously the polygraph signal was filmed and displayed on a split video monitor (Videomix, Inc., Campbell, CA). Sleep stages were differentiated according to conventional scoring methods (15).

Respiratory measurements. Respiratory measurements were performed by using a Bennet face mask attached to a Fleisch pneumotachograph and a Statham PM15E differential pressure transducer. The flow signal was integrated to give tidal volume. The air pressure within the mask was measured with a Statham Gould PM6 pressure transducer. Occlusions were performed by placing a fingertip over the opening of the face mask. The oxygen saturation always remained above 94%. The biphasic character of sighs was determined by analyzing the diaphragmatic EMG and the face mask air pressure.

EEG and EMG analysis. EEG recordings were digitized with a sample rate of 1000 Hz using a CED 1401 AD converter (Cambridge Electronic Design, Cambridge, UK) and was stored on an external hard disk (JAZ Drive 1GB, USA). A high sample rate provides a smoother digitized EEG signal, which is helpful for visual analysis of the raw computerized EEG signal and has no adverse effect on EEG power measurements (16). For evaluating the EEG information we performed spectral band filtering in the frequency bands 0.5-3 (δ), 4-7 Hz (θ), 8-13 Hz (α), and 14-18 Hz (β) and particularly in the σ frequency band of 12-14 Hz for sleep spindle identification. This strategy was viewed as appropriate because sleep spindles have been shown to have a peak energy in the 12-14-Hz frequency range (17). This peak was also confirmed in the present data by spectral analysis (fast Fourier transformation).

The diverse spectral filtering was made by fast Fourier transformation frames with 4096 samples using a Kaiser window of 40 dB and an overlap of 60%. In each time frame the complex demodulation (Hilbert transformation) (18) of the band-filtered signal was performed. This demodulation separates amplitude modulation and frequency modulation of the given frequency band. The separation of the frequency modulation enables one to get an envelope of the filter output that is independent of frequency shift in the chosen band. Thus the time function of the computed magnitude has no time shift and gives the clear development of the filter output, which is in our study in the σ frequency range of 12-14 Hz, the range of sleep spindle EEG activity in analogy to a compressed spectral band array (19). In a further step we detected groups of σ spindles by using a threshold over the magnitude of the 12-14-Hz filtered signal and finally calculated the duration of spindles and interspindle intervals (Fig. 1).

Figure 1
figure 1

The occurrence of σ sleep spindles was documented by an energy increase of the EEG signal in the frequency range of 12-14 Hz. The detection of the sleep spindle EEG pattern was performed by using a threshold over the band-passed EEG based on fast Fourier transformation and Hilbert transformation analysis techniques.

Full size image

To compare this spindle-based electrophysiologic approach for documenting arousals with criteria more generally used for cortical EEG arousal, we scored EEG activity in accordance with the EEG arousal criteria of the American Sleep Disorders Association (3). These criteria specify frequency shifts as the major indicator for EEG arousal. According to these criteria other associated phenomena such as high amplitude δ EEG waves, sharp, high amplitude EEG patterns similar to K-complexes, and artifacts such as pen-blocking phenomena cannot be interpreted as indicative of arousal if they are not accompanied by a frequency shift of >3 s. A frequency shift was defined as an increase of EEG power as represented in the compressed band array in the 4-12-Hz or 14-18-Hz frequency range.

Clearly the recording of only one channel (C3-P3) cannot exclude possible cortical arousal related EEG changes in other areas of the cortex in terms of conventional definitions of frequency as proposed by the American Sleep Disorders Association. This problem could not have been completely eliminated even by using multiple EEG channels. Our primary intention was not to definitively compare different EEG techniques for evaluating cortical arousal, but to analyze the possible effects of sighs and startles on ongoing EEG sleep spindles because suppression of spindles is a marker for activation of the ascending RE activating system. Because the healthy infants that we studied were volunteered for study by their parents, we wished to reduce the time and adverse stimuli inherent in placing multiple electrodes, and therefore we did not employ a full EEG montage. Occasionally, spindle activity in infants may occur unilaterally in one cortical hemisphere. Occurrence of this phenomenon would not have altered our arousal analysis of the data inasmuch as our conclusions are based on the arousal related suppression of ongoing spindle activity identified before arousal activation. The EMG analysis technique and the ECG artifact elimination procedure of the limb EMG and diaphragmatic activity is described elsewhere (6).

Spontaneous and occlusion evoked sighs (augmented breaths) and startles. Isolated spontaneous sighs and sighs accompanied by a startle were identified from the EMGdia pattern which had a distinct biphasic appearance and much more pronounced maximal activity compared with ordinary breathing efforts (6,20). The duration of these inspiratory efforts was longer than ordinary breaths. Spontaneous as well as occlusion-evoked startle responses were typically characterized by a sudden extension of two arms, which then was followed by a slower adduction having a pattern similar to the Moro reflex (21). Moro responses soon change their appearance and finally vanish after 6 mo of life (21), whereas the startle response, however, with some maturational modifications, will remain throughout the life time (22). We quantified the intensity of startles by grading them from I to III. Grade I startles were characterized by a sudden but small movement observed in one limb. Grade II designated startles with a sudden extension, abduction, and elevation of two limbs, usually both arms (21,23). Grade III was used for exceptional strong responses with movements in at least three limbs. This response often was accompanied by a brief opening of the eyes (less than 2 s) but without sleep state change.

We used brief occlusion of the airway of the infant's face mask as a respiratory stimulus to evoke sighs and startles. A total of 128 airway occlusion maneuvers that stimulated a sigh and a startle during N-REM sleep in 12 infants (age 9-19 wk) were analyzed. Of these 117 occlusion-related startles, 26 were grade I, 55 were grade II, and 36 were grade III. Three occlusions led to full awakening and were not included in the EEG sleep spindle analysis. Fifty-three brief occlusions did not evoke a sigh or a startle and are termed "short occlusions." We also analyzed 35 spontaneous sighs and 56 spontaneous sighs accompanied by a startle. Most of these occurred while the infant slept undisturbed without a face mask in place, and at least one of them was observed in each infant. In each infant during periods of undisturbed sleep we recorded 5-10-min periods of spindle activity and a total of 236 interspindle periods which were used as control values.

Statistical analysis. To evaluate the statistical significance of the prolongation of sleep spindle intervals, we compared meaned values for each infant with the infant's own control values using the paired t test. Similarly, the interspindle interval that coincided with the evoked startle in each infant was correlated with the startle intensity score by using the Spearman rank correlation.

RESULTS

In general we found a prolongation of the interspindle that coincided either spontaneous or evoked sighs and startles (Figs. 2–4). The spindle occurrence was immediately blocked even during the course of an ongoing sleep spindle (Fig. 4). Prolongation of interspindle intervals was limited to the interval coinciding with a sigh or startle; after this sleep spindles reappeared with the same rhythm as prior sighs and startles. Occlusion-related startles were always accompanied by a sigh.

Figure 2
figure 2

In each infant spontaneous sighs as well as spontaneous sighs accompanied by startles lead to an increase of the mean sleep interspindle intervals of the EEG compared with control intervals during undisturbed N-REM sleep. This transient suppression of sleep spindles reflects the effect of an arousal-related RE activation on RE thalamic cells, the site of sleep spindle generation. Short occlusions, which do not result in a sigh or a sigh and a startle, were not followed by a prolongation of these intervals.

Full size image
Figure 4
figure 4

Sighs coupled with a startle (st) (reflected by the activation of diaphragmatic and limb EMG) led to an immediate suppression of an ongoing sleep spindle and also lengthening of the interspindle interval.

Full size image

Effect of augmented breaths and startles on the EEG sleep spindle intervals. The interspindle interval increased after 82.8% of spontaneous sighs (Figs. 2 and 3, Table 1). When spontaneous sighs were accompanied by a startle the interval was prolonged in all instances and was further increased over that after sighs alone. After occlusions the interspindle interval was prolonged if a sigh occurred. Similar to the findings with sighs and startles, the interval was further increased if the occlusion-evoked sigh was accompanied by a startle (Fig. 5). The intensity of a startle was correlated with the interval between spindles with the more intense startles producing the longest suppression of spindle activity (Spearman rank correlation p < 0.01) (Fig. 5). In many grade III startles (14 cases) there was a brief opening of the eyes for 1-2 s without interruption of N-REM sleep as reflected in the EEG. To control for the influence of tactile stimuli from the face mask, we also analyzed the interspindle intervals during short occlusions in which no sigh or startle occurred. (Figs. 2 and 6). The spindle interval was not significantly prolonged in these cases (Mann Whitney rank sum test, NS).

Figure 3
figure 3

Below the raw EEG signal the energy of the EEG in the frequency range of 12-14 Hz is shown (σ sleep spindle frequency range). The periodic appearance of the sleep spindles was interrupted by a spontaneous isolated sigh (s) (reflected by the diaphragmatic EMG activation and the airflow) with a cessation of the energy of the 12-14-Hz band-passed EEG for a period of about 20 s.

Full size image
Table 1 Occurrence of EEG changes in terms of sleep interspindle interval prolongation and/or frequency shifts (range 0.5-18 Hz) of more or less than 3-s duration after spontaneous as well as evoked sighs and startles during N-REM sleep
Full size table
Figure 5
figure 5

The intensity of a startle (grades I-III) stimulated by airway occlusions showed a significant correlation (Spearman rank correlation p < 0.01) with the extent of interspindle interval prolongation. In contrast to startle-related sighs after airway occlusion, isolated sighs were followed by less marked spindle suppression.

Full size image
Figure 6
figure 6

After some short occlusions (occ) (reflected by the face mask air pressure) finally a sigh (s) and startle (st) were evoked, and this was immediately followed by transient sleep spindle suppression. Below the raw EEG signal in the first row the energy of other EEG frequencies is shown. To demonstrate the startle-related artifact due to EMG activation in the EEG signal (Eartf) the frequency range of 60-100 Hz is added. Beside the effect on sleep spindles there is only a short, moderate energy shift in the frequency range of 0.5-8 Hz after sigh and startle. However, an EEG activity change in terms of a shift to higher frequencies as to meet the 3-s EEG arousal criteria did not occur in this example.

Full size image

Other EEG changes after sighs and startles and progression to full cortical arousal. Some sighs and startles were followed by more general frequency shifts in the EEG pattern lasting longer than 3 s (Table 1). But such changes meeting the standard criteria for arousal accompanied only 20.3% of occlusion-related startles and were even less frequent after spontaneous startle-related sighs (5.4%). Spontaneous isolated sighs were never followed by this EEG pattern. EEG frequency shifts less than 3 s more commonly followed spontaneous and occlusion-related sighs and startles but in at least half of these cases even this minimal amount of general EEG change was absent (Table 1). In no case did we observe the progression of spontaneous sighs and startles to full cortical arousal with awakening. However, after three occlusion-related startles sustained desynchronization of the EEG and termination of N-REM sleep with awakening did occur. There was no correlation of startle intensities with the duration of EEG frequency shifts.

DISCUSSION

EEG spindle suppression as a measure of brainstem arousal. Our findings suggest that occurrence of sighs and startles is an indication of transient arousal without awakening. This follows from our observation of the close association between sighs and startles and suppression of spindles and from the known physiology of arousal and spindle suppression. These findings suggest a close link between sighs and startles and the activation of the ascending RE system responsible for arousal initiation (24,25).

Sleep spindles: incidence, site of generation, and influence of arousal-related brainstem activation. In reaching this conclusion, we relied on several well documented findings concerning the physiology of spindles. Sleep spindle oscillations first appear in humans during the second month of life during N-REM sleep and become clearly prominent against background activity at 3 mo of age. The duration of sleep spindles as well as the duration of the interspindle intervals during undisturbed sleep described in our study are in accordance with other results (13,14).

The spindle oscillations are the result of synaptic interactions in a network including inhibitory neurons of the RE thalamus, thalamocortical cells, and cortical pyramidal neurons (26). The arousal-related ascending pathways involved in spindle suppression are thought to arise mainly from cholinergic neurons at the pontomesencephalic junction (24,27). More precise studies revealed two main sites of the posterior midbrain, the laterodorsal nucleus and the pedunculopontine nucleus, which have widespread ascending projections, of which the most significant is the thalamus (28). The suppression of the σ sleep spindles happens at the site of their genesis, the RE thalamic nucleus. Brainstem cholinergic neurons induce a hyperpolarization of RE cells and, in this way impede spindle generating oscillations in these neurons (24). Other activating neuromodulators, serotonin (raphe nuclei) and norepinephrine (locus cerulus) hyperpolarize RE neurons, and this action promotes single spike activity of RE neurons (29), contributing to the arousal initiation. Although the relation of spindle suppression to the initiation of arousal has been fully characterized in these detailed neurophysiologic studies in animal preparations, to our knowledge this is the first report of arousal-related sleep spindle suppression in infants.

The blocking of spindle EEG activity was characterized by an immediate suppression of activity, which strongly suggests that this suppression reflects a very early stage of the sequence of events after initiation of arousal. Even ongoing spindle oscillations appeared to be blocked at the appearance of sighs and startles (Fig. 3). A similar occurrence has been demonstrated in cats after electrical stimulation of the mesencephalic RE formation with a latency of 200-500 ms (24,30,31). In contrast to other arousal-related EEG changes, we found that spindle suppression could persist up to 45 s after an occlusion-related stimulus. This observation indicates that brainstem arousal-related changes frequently have a much longer lasting effect on subcortical arousal pathways than on cortical activity as reflected by conventionally defined EEG arousals.

Recent experiments demonstrate that the conventional notion of a totally "desynchronized" cortical activity as observed during sleep arousal-related spindle blocking should be revised. Steriade and Amzica (32) have shown that, during arousal, fast rhythms (within γ EEG frequency range, mainly 30-40 Hz) are enhanced and "synchronized" within intracortical networks. Thus the term EEG "desynchronization" can no longer be used exclusively for an aroused state of vigilance. However, a desynchronization of EEG activity with regard to the blocking of spindle EEG activity that is based on highly synchronized thalamocortical neuronal discharge and interplay still occurs upon arousal. In the higher EEG frequency ranges we also have observed enhanced EEG activity in frequencies above 20 Hz (γ EEG frequency range) during as well as after sighs and startles. But because we used surface EEG electrodes and not intracellular and extracellular needle EEG techniques (11), we have not been able to absolutely exclude the interference of artifacts through EMG activity represented in the frequency range of 60-100 Hz (see Eartf channel (Fig. 6). Thus we excluded this frequency range from further analysis in the present study.

Spectrum of arousal related electrophysiologic phenomena on EEG. The present findings further illustrate the broad range and variety of arousals in infants. The electrophysiologic manifestations of arousal involve a range of changes extending from full cortical EEG arousal at one end of the spectrum (33,34) to a transient motor activity or desynchronization of the EEG (35,36) affecting spindles as a result of subcortical activity at the other. In recent years there has been a trend to define arousals more narrowly, particularly with regard to clinical studies. Transient arousals during sleep not progressing to awakening are evidenced by a shift to higher frequency ranges lasting at least 3 s according to the proposed criteria of the American Sleep Disorders Association (3). In our study, albeit limited somewhat by the inclusion of only one EEG channel, cortical arousal characterized by these defined criteria could be found during only a minority of sigh and startle events, whereas a blocking of sleep spindles occurred in 100% after startles, either spontaneous or evoked. These findings show that arousals evidenced by sighs and startles and initially confined to brainstem centers may or may not subsequently spread throughout the brain and lead to cortical EEG changes, some of which may meet conventional EEG arousal criteria and some not. Furthermore, if EEG activation using the 3-s criteria of the American Sleep Disorders Association was used, there remained 40% of sigh startle events that were unassociated with EEG activation. Thus with regard to the analyzed EEG channel, an arousal-related EEG frequency shift was not a very sensitive indicator of subcortical arousals that were substantial enough to significantly effect motor and respiratory behaviors.

This finding suggests that the clinical definition of arousal in infants could appropriately be expanded to include brainstem and subcortical arousal activation as represented by sighs and startles and reflected by sleep spindle suppression. Our results support the hypothesis that arousal related activation of the ascending RE activating system can be confined to the brainstem. Sighs (6) and startles (8) are initial reflex motor behaviors that accompany brainstem arousal. Consequently our results are in accordance with our own and other recent studies of infant apnea concerning autonomic and respiratory evidences of brainstem arousal in the absence of cortical activation (4,6,37).

The appearance of biphasic sighs and arousal. Orem and Trotter's (37) detailed investigation of central respiratory neuronal activity in intact unanesthetized cats has revealed early and late firing inspiratory medullary cell populations whose activity corresponds to the "breath on breath character" of an augmented breath or sigh. In this feline model this pattern was associated with arousal from sleep (37). These findings are in accordance with our finding of a close relationship between sighs and arousal-related brainstem activation in infants reflected by spindle suppression.

Another study of spontaneous sighs in sleeping cats has demonstrated changes in hippocampal bioelectrical activity immediately preceding spontaneous sighs (38). A widespread circuit in generation of sighs was proposed including vagal afferents as well as hypothalamic and forebrain influences. This study did not report changes in predominant EEG pattern and did not include the analysis of sigma spindle activity. Thus it does not directly speak to the proposed relation of sighs to arousals, however, it does indicate that forebrain influences as well as those of the brainstem may be involved in sigh generation.

The appearance of startles and arousal. Spontaneous startles are a common finding in newborns. They occur periodically during all sleep stages (39), but have not traditionally been related to arousal in infants (40). It is well established now that the system subserving the motor limb of the startle reflex in humans originates, as in animals, in the RE formation of the caudal brainstem (41). The observation that the startle reflex exists in anencephalic infants supports this concept (42). In various animal studies the startle response has been related to an escape reaction from unexpected danger and can be potentiated by fear (43). But a relation between arousal from sleep and startle in these models remains controversial (44).

In contrast the frequent simultaneous appearance of sighs and startles in our study indicates a close relationship in terms of synchrony between sigh-, startle-, and arousal-related brainstem centers. Moreover, our study shows that brainstem arousal is not an "all or none" phenomenon. The intensity of the arousal response in terms of duration of spindle suppression is paralleled by a graduation in intensity of respiratory and behavioral response, which extends from an isolated sigh to sighs accompanied by startles with incremental magnitudes. This clear correlation of the startle intensity with the duration of spindle suppression more precisely defines the close link between startle and arousal from sleep in infants.

Arousal failure hypothesis and SIDS. It has been hypothesized that failure in arousal from sleep during life-threatening events either related to apnea or to positional airway compromise may play an important role in SIDS. Absent or delayed arousal to an asphyxial stimulus has been identified in many infants with a recent history of an apparent life-threatening event (45). Furthermore, decreased spontaneous motor activity reflecting decreased arousals has been observed in infants who later died of SIDS (46,47). The present work suggests that SIDS could occur as a result of partial or weak arousals in addition to total failure of arousal. Head lifting and turning to end asphyxia when bedding covers the infant's external airway depends on the startle response (8). Therefore the intensity of startles might be critical in determining success of recovery from such positional airway compromise.

In summary this study has shown evidence that sighs and startles, either spontaneous or evoked, are related to an activation of arousal-related structures within the RE formation reflected by EEG spindle suppression. The magnitude of the startle response correlated with the duration of EEG spindle suppression, and it is proposed that the intensity of brainstem arousals could be important in recovery from life-threatening events during sleep.