Previous Article |
Table of Contents
| Next Article 
BIOLOGICAL SCIENCES / PHYSIOLOGY
Repeated sleep restriction in rats leads to homeostatic and allostatic responses during recovery sleep
Center for Sleep and Circadian Biology, Department of Neurobiology and Physiology, Northwestern University, 2205 Tech Drive, Hogan 2-160, Evanston, IL 60208
Edited by Bruce S. McEwen, The Rockefeller University, New York, NY, and approved May 3, 2007 (received for review December 4, 2006)
 |
Abstract |
|---|
 Top Abstract ResultsDiscussionMethodsAcknowledgementsReferences |
|---|
35 h of sleep over 5 days of sleep restriction, animals regained virtually none of their lost sleep, even during a full 3-day recovery period. These data demonstrate that the compensatory/homeostatic sleep response to acute SD does not generalize to conditions of chronic partial sleep loss. We propose that the change in sleep–wake regulation in the context of repeated sleep restriction reflects an allostatic process, and that the allostatic load produced by SD has direct effects on the sleep–wake regulatory system.
allostasis | homeostasis | sleep deprivation | stress | rodents
In contrast, rat studies by Rechtschaffen et al. (6) have shown that the compensatory sleep response to acute TSD does not generalize to longer durations of sleep loss. For example, after 48–96 h of continuous TSD, rats exhibit a large REM rebound but fail to generate any positive rebound in EEG NREM delta power, and NREM sleep time is actually decreased to below baseline (BL) levels, indicating that they regain none of what was lost in these two sleep parameters (6). These data provide evidence that the sleep homeostatic process may be fundamentally different between acute and chronic TSD conditions.
Interestingly, recent studies in humans also indicate that the sleep-recovery process may be altered during conditions of chronic partial sleep loss. In one clinical study, in which healthy young adults were restricted to 4, 6, or 8 h of sleep over 14 consecutive days, there was no indication that cumulative sleep loss over days was compensated by increases in either sleep time or NREM EEG delta power during the daily sleep opportunities (8). In addition, self reports of subjective sleepiness appeared to undergo adaptation and did not increase in severity after the second or third day of sleep restriction. Another study using 3, 5, or 7 h of sleep restriction for 7 consecutive days in healthy young adults showed that, despite an accumulating sleep debt, there was no change in sleep–wake patterns across sleep restriction days or during a recovery opportunity (9). One postulate for these findings is that once some core sleep need is achieved, possibly as little as 3–4 h per night, additional sleep is dispensable and fulfills no pertinent function (10). However, recent experimental studies in humans have shown that chronic partial sleep loss of even 2–3 h per night leads to impairments in cognitive performance, as well as cardiovascular, immune, and endocrine function (8, 9, 11–13), indicating that even mild-to-moderate chronic sleep restriction is not inconsequential.
Despite the fact that chronic partial sleep restriction is a hallmark of life in modern society, few attempts have been made to develop animal models of repeated sleep loss. This is surprising in view of the fact that it was demonstrated 50 years ago that, although rats could tolerate prolonged (10- to 18-wk) partial sleep restriction (sleep allowed for only 4 h per day), such loss led to reduced growth and made the rats extremely irritable and reactive (14). We have identified only one study where sleep homeostasis was specifically assessed in animals exposed to a mild chronic sleep restriction protocol (15). When rats were deprived of sleep during the 12-h light phase for 5 consecutive days, they generated a similar homeostatic response (i.e., increased NREM delta power and sleep amount) during each of the 12-h dark-phase recovery opportunities, suggesting that animals fully compensated to this mild SD protocol on a daily basis.
We have developed a rodent model of repeated sleep restriction (RSR) in which EEG/electromyographic (EMG) activity was recorded continuously for a 24-h BL day, 5 consecutive days of sleep restriction (20 h of SD and 4 h of sleep opportunity per day), and 3 days of full-recovery sleep opportunity. The major and surprising finding to come out of this study was that, after the first day of sleep restriction, animals failed to express a compensatory response in NREM delta power, the primary measure of sleep homeostasis. This lack of rebound, in the context of accumulating sleep debt across days, indicates a change in the homeostatic process to a strategy that is more reflective of an allostatic response to repeated sleep loss.
|
Results |
|---|
|
TopAbstractResultsDiscussionMethodsAcknowledgementsReferences |
|---|
Total Sleep Time (TST). On the BL day, rats slept 10.6 ± 0.4 h across the 24 h, including 8.0 ± 0.4 h during the 20-h time block [zeitgeber time (ZT) 4–24] and 2.6 ± 0.1 h over the 4-h time block (ZT0–4) (Fig. 1A). Significant time effects occurred for TST over the 24-h (F(8, 56) = 152.3, P < 0.001), 20-h (F(8, 56) = 173.8, P < 0.001), and 4-h (F(8, 56) = 2.6, P < 0.05) time blocks. On each day of sleep restriction (SD1–SD5), 24-h TST was significantly decreased compared with the BL day (all post-hoc comparisons, P < 0.001). Similarly, during the 20-h (ZT4–24) sleep-deprivation periods on SD1–SD5, TST was lower than the corresponding 20-h BL period (all post-hoc comparisons, P < 0.001). The average amount of sleep achieved in the wheel during the 20-h sleep-deprivation block was relatively small and ranged from an average of 0.2 ± 0.1 h on SD1 to 1.4 ± 0.4 h on SD5, although there were no significant differences between SD1–SD5 days. During the 4-h (ZT0–4) sleep-opportunity blocks on SD1-SD5, TST showed a small but statistically significant increase on SD1 (P < 0.01) and SD2 (P < 0.05) compared with corresponding 4-h BL levels, whereas no increase in TST was noted on SD3–SD5. Even though there was no change in TST even on the last SD day (SD5), TST over the entire 24 h and the 20-h time block on R1 was significantly increased compared with BL.
|
REM Sleep Time. Over the 24-h BL period, rats produced 1.9 ± 0.1 h of REM sleep, comprised of 1.5 ± 0.1 during the 20-h time block and 0.4 ± 0.03 over the 4-h time block (Fig. 1C). A significant time effect was noted for REM sleep time over the 24-h (F(8, 56) = 193.6, P < 0.001), 20-h (F(8, 56) = 311.2, P < 0.001), and 4-h (F(8, 56) = 27.0, P < 0.001) time blocks. On each day of sleep restriction (SD1–SD5), 24-h REM sleep time was significantly reduced compared with the BL day (all post-hoc comparisons, P < 0.001). During the 20-h (ZT4–24) SD block on SD1–SD5, REM sleep time was almost completely eliminated, with only minimal amounts appearing on SD2–SD5. During the 4-h (ZT0–4) sleep opportunity, there was a significant increase in REM sleep time on each sleep restriction day (SD1–SD5) compared with the corresponding BL 4-h time block (all post-hoc comparisons, P < 0.001). The increase in REM sleep time during the 4-h time block on SD5 was continued into the R1 day 20-h time block, where REM sleep was elevated over the corresponding BL 20-h interval (P < 0.001). REM sleep time was reduced during the last 4 h of R1 (P < 0.05), whereas no changes occurred on R2 and R3 compared with the BL condition.
NREM Delta Power. NREM delta power is commonly used as a quantitative measure of EEG sleep intensity and homeostatic sleep drive. We analyzed NREM delta power during each 20-h (ZT4–24) and 4-h (ZT0–4) time block across BL, SD1–SD5, and R1–R3 days (Fig. 2). During the 20-h wheel deprivation period on SD1–SD5, there were many days when rats achieved no sleep in the wheel or slept only a small amount of time. Furthermore, when a 10-sec epoch was scored as NREM sleep, it normally contained a mixture of wake and NREM EEG signals, because sleep in the wheel was very fragmented. Therefore, the specific measure of NREM delta power could not be accurately determined during the 20-h time block (however, see the alternative analysis in the following paragraph). During the 4-h time blocks, a significant time effect was detected (F(8,56) = 16.4, P < 0.001), such that NREM delta power was significantly increased (+28%) on SD1 compared with the corresponding 4-h BL period (Fig. 2). Thereafter, NREM delta power returned to BL levels and was comparable to BL levels on SD2–SD5. A comparison between BL and R1–R3 days showed that NREM delta power was significantly reduced on R1 (ZT0–4, P < 0.001), R2 (ZT4–24, P < 0.001; ZT0–4, P < 0.05), and R3 (ZT4–24, P < 0.05) compared with the corresponding BL values.
|
BL vs. Recovery. In addition to examining sleep and wake during the 20- and 4-h time blocks, to more fully understand recovery sleep after 5 days of RSR, we compared NREM sleep time, NREM delta power, and REM sleep time over 2-h intervals for the 3 recovery days to the pattern during BL sleep. In Fig. 3, the BL and R1–R3 recordings are divided into 2-h intervals beginning at light onset. It should be noted that hours 0–4 on the first full recovery day (Fig. 3) are the same as the 4-h sleep opportunity on SD5 (Fig. 1), the last day of sleep restriction. On the first recovery day, there was a trend for decreased NREM sleep time in the first 6 h of recovery (ZT0–6), followed by a significant increase in NREM sleep time in hours 3–8 of the dark period (ZT15–20) compared with corresponding BL levels (condition main effect, F(1, 7) = 22.4, P < 0.01; condition x time interaction, F(11, 77) = 4.9, P < 0.001) (Fig. 3A). On R2, NREM sleep time was significantly increased during the first 2 h of the light phase (ZT0–2) and during hours 3–6 of the dark phase (ZT15–18) (condition main effect, F(1, 7) = 8.2, P < 0.05; condition x time interaction, F(11, 77) = 2.8, P < 0.01). On R3, no differences in NREM sleep time were observed between BL and recovery conditions.
|
A clear increase in REM sleep time occurred during most hours of R1 compared with corresponding BL levels (condition main effect, F(1,7) = 112.3, P < 0.001; condition x time interaction, F(11, 77) = 2.3, P < 0.05) (Fig. 3C). A combination of positive and negative rebounds occurred on R2 (condition x time interaction, F(11,77) = 3.0, P < 0.01). On R3, there were no differences between BL and R conditions.
Cumulative Sleep Loss and Recovery. The amount of total sleep loss during the ZT4–24 block was determined for each sleep restriction day (SD1–SD5) and accumulated over the 5 days (SI Fig. 5). Animals lost a mean of 35.8 ± 2.8 h of sleep over the five 20-h blocks of SD. Similarly, the amount of sleep gained over BL levels was determined for each 4-h sleep opportunity (SD1–SD5) (SI Fig. 5). Animals accumulated 0.8 ± 0.4 h of sleep over BL levels during the five 4-h sleep opportunities, even though there was a total potential net recovery of 7.0 ± 0.5 h over BL levels across the five 4-h time blocks. The net result was an average loss of 35.0 h of sleep over the 5-day (SD1–SD5) period. Yet, by the end of R3, rats had recovered only a total of 2.6 h of sleep (vs. BL), consisting of 2.3 h during the ZT4–24 time block and 0.3 h over the ZT0–4 time blocks. This represented only a small portion of the potential net recovery for the combined 20- and 4-h recovery opportunities across R1–R3 (40.2 ± 1.2 h). At the end of R3, animals had a remaining sleep debt of 32.4 h. The cumulative sleep loss and recovery values for individual NREM and REM sleep states are presented in Table 1.
|
|
Discussion |
|---|
|
TopAbstractResultsDiscussionMethodsAcknowledgementsReferences |
|---|
After the first block of 20-h SD (SD1), rats generated a positive rebound (i.e., increase over BL levels) in both EEG NREM delta power and REM sleep time during the 4-h sleep opportunity, compared with the corresponding 4-h BL interval; these increases are indicative of an acute homeostatic response (Figs. 1 and 2). The compensatory increase in NREM sleep intensity is consistent with previous in vivo data from acute (6- to 24-h) TSD studies in rats (4–6). NREM delta power is normally the first component of the compensatory sleep response to be expressed after acute TSD, indicating that physiological processes linked to EEG delta wave generation are vital for the recuperative response to sleep loss.
The most remarkable finding in this study was that, during the 4-h recovery opportunities on SD2–SD5, animals failed to exhibit the same positive rebound in EEG NREM delta power that appeared on SD1, despite having accrued an even greater amount of sleep debt each day. In fact, during the 4-h recovery periods on SD2–SD5, NREM sleep intensity returned to original BL levels. These data demonstrate a clear change in the homeostatic sleep response in the context of acute sleep loss vs. RSR conditions.
One possible reason for the loss in sleep homeostatic drive was that sleep drive was dissipated while the animals were being sleep-deprived, either through microsleeps or through increased delta activity in wakefulness. We determined that the amount of delta activity (i.e., delta energy) that accrued across the 20-h (ZT4–24) SD procedure was similar to corresponding BL levels (SI Table 2). Therefore, animals did not compensate for the significant loss of sleep by increasing total delta energy across wakefulness and sleep during the SD procedure. This finding indicates the reduction in NREM delta power during the 4-h sleep opportunities was not accounted for by a compensatory increase in cumulative delta activity during the course of SD.
Another remarkable finding in this study was that, despite the accumulated 35 h of sleep loss across SD1–SD5, animals failed to recover any amount of NREM sleep time during the 4-h sleep opportunities and did not increase the magnitude of REM rebound across days (Fig. 1). In fact, on SD4 and SD5, NREM sleep time was significantly lower compared with corresponding BL levels, indicative of a negative rebound. In total, animals had the capacity to regain 7 h of NREM and REM sleep time over corresponding BL levels between SD1 and SD5; however, they averaged only 0.8 h of sleep rebound (Table 1 and SI Fig. 5). Therefore, animals failed to use all of the time that was available to them to sleep during the 4-h time blocks. One explanation for the lack of a positive NREM rebound could be that the 4-h sleep opportunity was not long enough for an increase in NREM sleep amount to be expressed. That is, the animals exhibited a selective preference to increase REM sleep rather than NREM sleep time. Previous studies in rats have shown that after 24 h of TSD, a pronounced rebound in NREM sleep time can be delayed by many hours after the termination of SD (16).
After the SD block on SD5, animals were given 3 full days of an ad-lib recovery opportunity to detect a possible delay in the recovery pattern to the previous days of sleep loss (Fig. 3). After a brief increase in NREM delta power during only the first 2 h of R1, levels quickly returned to BL levels and showed a dramatic and sustained negative rebound during the dark phase on R1 and the light phases of R2 and R3. Therefore, even during an extended recovery opportunity, animals showed virtually no tendency to gain back lost sleep by increasing sleep intensity. NREM sleep time was increased during the dark phase on R1 and R2; however, the amount of sleep regained hardly approached the amount that was previously lost and sleep intensity was no greater than BL levels (Table 1). Because the rebound in NREM sleep time during the recovery days was small and showed no increase in EEG sleep intensity, it is unlikely that the negative rebound in NREM delta power could be accounted for by the increased NREM sleep time. Therefore, we conclude that the negative rebound in NREM delta power is not a function of the homeostatic process but is related to another regulatory process, as described later in Discussion. The rebound in REM sleep was primarily restricted to R1 and accounted for only a small proportion of REM sleep that was lost during the deprivation episodes. Therefore, even during the full recovery sleep opportunities on R1–R3, animals used only a small portion of the time that was available on the recovery days to regain sleep or pay off their accumulated sleep debt.
The two-process model of sleep regulation has been used to predict the homeostatic response to SD, using NREM delta power as a quantitative measure of sleep homeostasis (1, 17). The model predicts that, after prolonged or repeated SD, NREM delta power will reach a saturating exponential or a "peak" level that cannot be surpassed, even in the face of additional sleep loss. Our results are strikingly different from this prediction, because on SD2–SD5, NREM delta power did not remain at a saturated level but dissipated to pre-deprivation BL levels (Fig. 2). Furthermore, on the full recovery days (R1–R3), NREM delta power was predominantly maintained at or below original BL levels (Fig. 3). Therefore, our data suggest that during the course of RSR, sleep falls under the regulation of a process other than homeostatic regulation. The rest of this discussion will develop a rationale for what this alternative process may be.
Clinical studies have shown that chronic partial sleep loss results in significant alterations in energy metabolism and cardiovascular function. For example, sleep restriction in young healthy subjects to 4 h per night for 5 consecutive nights results in symptoms of early-stage diabetes (13). In addition, experimental sleep restriction has been associated with increased risk factors for cardiovascular disease, such as high levels of C-reactive protein (11). We have recently used the RSR paradigm to demonstrate that rats allowed to sleep for only 4 h per day for 8 consecutive days have increased basal corticosterone levels and develop pronounced changes in the reactivity of the hypothalamo-pituitary-adrenal axis to stress (18). RSR and other forms of SD have also been shown to alter regulation of corticosterone (18–20) and proinflammatory cytokine levels (21, 22), elevate markers of oxidative stress (23), and promote neurodegenerative processes in the hippocampus (24, 25).
Taken together, data from human and animal studies indicate that RSR results in two notable consequences, (i) a change in the sleep response to cumulative sleep loss and (ii) changes in the regulation of multiple physiological systems. Although most studies have examined specific physiological pathways affected by SD, few models to integrate the multiple effects of sleep loss have been available. McEwen (26) has recently proposed that the biological model of allostasis and allostatic load may be used to explain the deleterious effects of sleep loss. The concept of allostasis refers to the maintenance of stability within the organism through change, a process that allows for the integration of physiology and behavior in response to a changing environment (27). Allostasis is mediated primarily by stress responses that, in the short term, lead to coordinated changes in physiological systems that result in adaptive behavior. However, when mediators of allostasis, such as glucocorticosteroids, autonomic nervous system activity, or inflammatory cytokines, are chronically high, an allostatic load or overload will develop that takes the animal out of an adaptive response mode and leads to multiple pathologies (27).
The stress response elicited by SD has long been considered a confounding factor in studies attempting to isolate the specific role of sleep loss itself on any independent variable under investigation. Indeed, RSR is associated with elevations in corticosterone (18–20), sympathetic tone, and cytokine levels (13, 21, 22), important mediators of allostasis and allostatic load. We have used the current RSR model (4-h sleep per day for 5 consecutive days) to demonstrate elevated corticosterone levels on SD1 and SD5, even when blood samples were collected after the 4-h sleep opportunity (unpublished data). Rather than discarding the stress response as an unwanted experimental confound, the allostatic model incorporates stress as a critical component of the physiological state of SD (26). In this framework, the condition of TSD or RSR represents a stress that adds to the allostatic load on various physiological systems and results in nonadaptive responses and adverse health consequences (e.g., metabolic, cardiovascular, immune, and neurocognitive impairments).
Our finding that RSR results in a change in the homeostatic sleep response raises the intriguing possibility that the sleep–wake regulatory system itself is significantly impacted by the allostatic load of cumulative sleep loss. That is, the allostatic load resulting from RSR feeds back onto the sleep–wake system, in turn changing the acute compensatory sleep response into an allostatic response. Specifically, we interpret the loss of NREM delta power rebound on SD2–SD5, the prolonged negative rebound in NREM delta power during R1–R3, and the reduction in NREM sleep time on SD4–SD5, as signs of allostasis. REM sleep, which exhibited a consistent positive rebound, is possibly less susceptible to allostasis and is able to remain under some degree of homeostatic control around the original set point.
Saper et al. (28) have speculated that the homeostatic and circadian drive for sleep can be overcome for brief periods of time by an allostatic drive when external conditions demand a change in the sleep–wake regulatory system. Such a change might be adaptive for an animal in the wild, because TSD or RSR would be expected to occur only under emergency environmental conditions when a change from normal life events to a survival mode would be necessary. The homeostatic pressure to sleep, along with subsequently falling asleep when an animal must maintain wakefulness under emergency circumstances (e.g., flooding), is of obvious detriment for survival, and thus the need for an allostatic response to repeated sleep loss might be more beneficial for survival. Of course, there could be a limit to the amount and/or duration of sleep loss that is tolerable to meet environmental demands; anything beyond this limit would then lead to nonadaptive physiological changes and negative consequences to the organism. Indeed, it is conceivable that allostasis in the sleep recovery process could itself add to allostatic load in the stress response system, setting up a positive feedback series of events with negative health consequences in organisms partially sleep-restricted for long periods of time. Using the conceptual structure of the allostatic model, many hypotheses can be addressed regarding the relationship between chronic sleep loss and health, including investigations into the physiological, molecular, and genetic mechanisms that link sleep with specific physiological processes and disease states. The growing number of epidemiological and experimental studies demonstrating close reciprocal relationships between sleep and a diverse range of disease states (29) points to the importance of developing animal models and conceptual models to investigate these important links.
|
Methods |
|---|
|
TopAbstractResultsDiscussionMethodsAcknowledgementsReferences |
|---|
Surgical Procedures. Rats were anesthetized (ketamine 87 mg/kg and xylazine 13 mg/kg, i.p.) and surgically implanted with EEG and EMG electrodes for sleep–wake recording. EEG electrodes (Small Parts, Miami Lakes, FL) were placed contralaterally on the skull surface (1.0 mm anterior to Bregma/1.0 mm right of the central suture and 1.0 mm posterior to lambda/on the extended line of the central suture). EMG activity was monitored by using stainless steel Teflon-coated wires placed bilaterally in the nuchal muscle in the dorsal neck region. After surgery, animals remained in their home cage for 2 weeks to recover.
Experiment Design. After recovery, animals were transferred to sleep-recording chambers and connected to a wire tether/commutator system (Plastics One, Roanoke, VA) for the collection of EEG/EMG signals. Each animal was placed in an individual chamber that contained a sleep-recording cage and a SD wheel. The chambers were light-, temperature-, and sound-controlled. A 7-day adaptation period to the recording environment was allowed before the 9-day experimental protocol was performed. On day 1, sleep was recorded for a 24-h BL period beginning 4 h after light onset (ZT4), as depicted in SI Fig. 4. For the next 5 consecutive days (SD1–SD5), animals were sleep-deprived for 20 h (ZT4–24), followed by a restricted 4-h ad lib sleep opportunity. The protocol was designed so that the 4-h blocks of restricted sleep on SD1–SD5 began at light onset (ZT0–4). After the last day of sleep restriction, animals were allowed a 3-day recovery period (R1–R3). EEG/EMG recordings were collected throughout the entire protocol, including the SD blocks and ad lib sleep opportunities.
SD. Animals were sleep-deprived by placing them in a slowly rotating wheel [(Techniplast USA, Exton, PA) stainless steel, 13.5 inch in diameter, 3.75 inch in width] maintained at a constant speed (1.5 rpm) by a motor. Food and water were freely available. EEG/EMG was continually recorded in the wheel. After each 20-h SD period, animals were quickly placed back into their sleep-recording cage for the 4-h sleep opportunity.
Sleep Data Collection and Analysis. EEG signals were amplified x10,000 with high- and low-pass filters set at 1 and 30 Hz, respectively. EMG signals were amplified x5,000 with high- and low-pass filters set at 3 and 100 Hz. Both signals were then digitized at 102.4 Hz by an analog-to-digital converter (Data Translation, Inc., Marlboro, MA, model DT-01EZ) and stored on a Dell Pentium IV (ABT Electronics, Chicago, IL) computer. Waveforms were collected by using Multisleep (Actimetrics, Evanston, IL), a software system designed for gathering and analyzing rodent sleep data. After data collection, the EEG and EMG signals were reconstituted on a computer screen in 10-sec epochs and visually scored as either wake NREM or REM sleep. For quantitative analysis of the EEG signal in the frequency domain, each 10-sec scored epoch was subjected to fast Fourier transformation. In particular, for epochs of NREM sleep, the EEG power in the delta (1- to 4-Hz) frequency range was calculated. During the 20-h SD blocks and corresponding BL and R periods, total delta energy [i.e., summed delta power (µV2) across all epochs] was determined. Epochs containing artifact were eliminated from power spectral analysis.
Postscoring analysis was performed by using SleepReport (Actimetrics) for the determination of sleep-structure parameters. Depending on the particular analysis, wake, NREM, and REM sleep amounts, as well as NREM delta power, were determined in 2-, 4-, 20-, or 24-h intervals. For statistical comparisons of sleep–wake parameters across BL, SD1–SD5, and R1–R3 conditions, a repeated-measures analysis of variance (Statistica, StatSoft, Tulsa, OK) was used. Post-hoc comparisons were made by using the least standard difference test, when indicated. A value of P < 0.05 was considered significant for all comparisons.
|
Acknowledgements |
|---|
|
TopAbstractResultsDiscussionMethodsAcknowledgementsReferences |
|---|
|
Footnotes |
|---|
Abbreviations: REM, rapid eye movement; NREM, non-REM; TSD, total sleep deprivation; RSR, repeated sleep restriction; EEG, electroencephalogram; EMG, electromyographic; TST, total sleep time; ZT, zeitgeber time; BL, baseline; SD, sleep deprivation; Rn, recovery day n.
*To whom correspondence should be addressed. E-mail: fturek{at}northwestern.edu
Freely available online through the PNAS open access option.
Author contributions: A.D.L. and F.W.T. designed research; Y.K. performed research; Y.K., A.D.L., B.M.B., and F.W.T. analyzed data; and A.D.L. and F.W.T. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
This article contains supporting information online at www.pnas.org/cgi/content/full/0610351104/DC1.
© 2007 by The National Academy of Sciences of the USA
|
References |
|---|
|
TopAbstractResultsDiscussionMethodsAcknowledgementsReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg What's this?
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
