Procedures and apparatus

Sixteen subjects (21 to 46 years of age) were recruited to participate in the study from an electronic posting at Rensselaer Polytechnic Institute in Troy, N.Y. All subjects were screened for major health problems and except for women taking birth control pills, subjects reported not taking any pharmaceuticals or medications. Every subject completed a Munich Chronotype Questionnaire (MCTQ) prior to the study [21]; those who were late or extremely late chronotypes were excluded from the experiment. All subjects provided an informed consent approved by Rensselaer's Institute Review Board. Subjects were asked to refrain from alcohol and caffeine on the days of the experiment and were asked not to sleep after awakening for the day. Of the sixteen subjects, nine males and five females completed the entire experiment, and the results of their data are reported here.

Four experimental lighting conditions, two spectra (blue and red) at two levels (10 and 40 lx), were delivered to individual subjects from 0.6 × 0.6 × 0.6 m light boxes, each fitted with arrays of light-emitting diodes (LEDs). The arrays (ICove, Color Kinetics) were located behind the front box apertures to be outside the subject's direct view, thereby creating a uniform, non-glaring distribution of light within the box. During light exposures, subjects placed their chin on a rest mounted near the front of the box, ensuring delivery of the prescribed light exposure. The spectral emissions of the blue LEDs peaked at 470 nm with a full width at half maximum (FWHM) of 25 nm. Light from the red LEDs peaked at 630 nm with a FWHM of 25 nm. Before the experiment, each of the light boxes was calibrated using a Gigahertz illuminance photometer to provide the prescribed corneal illuminance levels when subjects were positioned in the chinrest. The spectral radiances of the red and blue conditions were measured prior to the experiment with a calibrated spectroradiometer (Photoresearch, model PR705a) and diffuse white reflectance standard (Labsphere, model SR 099) and were used to calibrate the Gigahertz illuminance readings. Two boxes provided blue light (40 μW/cm² at 40 lx and 10 μW/cm² at 10 lx) and two emitted red light (19 μW/cm² at 40 lx and 4.7 μW/cm² at 10 lx); light levels could be adjusted with an electronic dimmer to reach the prescribed light levels without significantly affecting the relative spectral distributions of the LED emissions. Measurements of pupil area completed after the experiment with a different group of subjects (n = 5) were: red at 10 lx, 34 mm²; red at 40 lx, 22 mm²; blue at 10 lx, 10 mm²; blue at 40 lx, 6.5 mm².

The study was conducted over the course of several weeks in April/May 2008 and in March 2009. Groups of four subjects participated in two sessions separated by at least one week. Subjects were asked to arrive at the laboratory at 22:00 to receive instructions, become familiar with the performance tests, and be fitted with scalp electrodes for EEG recordings. Because only one EEG machine was available, data collection was staggered. The first subject in a session started at 23:00, the second at 23:10, the third at 23:20, and the last at 23:30; the last subject completed the experiment at 03:45. During every session, each subject was presented a high (40 lx) and a low (10 lx) light exposure condition of the same spectrum (blue or red). The presentation order of the light levels was counterbalanced across sessions for a given subject; light spectra were counterbalanced across subjects within sessions. Every 45-minute light exposure condition was preceded by a 45-minute period of inactivity in a dark anteroom (< 1 lx of red light at the cornea). During the inactive dark periods, subjects remained quiet and were not allowed to perform any task (e.g., talk, read, or computer work) except for the prescribed data sampling specified in the experimental protocol. Each data collection period (after each dark and after each light exposure condition), which included EEG measurements, saliva collection and PVT tests, lasted about 15 minutes. In sum then, in each nighttime session there was a total of four, 45-minute light-and-dark conditions (a dark condition always preceded one of the four light conditions), and each condition was continued for 15 minutes for data collection. Figures 1 and 2 show the experimental design and the data collection activity periods, respectively.

PVT and KSS

Self-luminous personal digital assistants (PDA; Tungsten E2 models from Palm Inc.) were used to record subjects' performance levels and sleepiness self-reports [22, 23]. A battery of three psychomotor vigilance tests (PVT; Brain Checkers 2.75 from Behavioral Neuroscience Systems LLC) were used to measure performance: (a) a Simple Reaction Times (RT) test where the subject had to tap a symbol displayed on the PDA screen with a stylus as soon it appeared; (b) a Two-choice Reaction Time (TCRT) test where the subject had to choose, by tapping the screen with a stylus, whether a previously, briefly displayed numeral on the screen was either a "2" or a "3" or a "4" or a "5"; and (c) a Matching-to-Sample (MTS) test where the subject had to indicate with the stylus which of two specific patterns of squares had previously been displayed on the screen. The KSS was used to probe self-assessments of sleepiness [24]. Subjects performed tests every 15 minutes (three times for every light-and-dark condition) throughout a four-hour session. Tests were performed while subjects were sitting comfortably in a chair, either adjacent to the light box or in the dark anteroom.

PVT and KSS data were downloaded to an Excel spreadsheet for subsequent analyses using the BCDataMan 2.0.10 software from Behavioral Neuroscience Systems LLC. Both a response time and a score were determined for the TCRT and MTS tasks. Both scores were calculated by dividing the percentage of correct responses by the mean response times for each test during a given recording period.

Melatonin

Melatonin, a hormone produced at night and under conditions of darkness, is used as a marker of circadian system timing. In darkness, melatonin levels rise in the evening, reach a peak between 02:00 and 04:00 and decline to daytime levels between 07:00 and 09:00. Light can attenuate the rate of melatonin production during the nighttime; depending upon the characteristics of light exposure (intensity, spectrum, duration), melatonin synthesis will continue to rise, but at a slower rate, in the early part of the night and continue to fall, but at a faster rate, toward the end of the night. In the current study, 1 to 2 ml saliva samples were collected at the end of each dark and each light exposure period using the Salivette system from Alpco Diagnostics (four samples per session per subject). The vessels containing the suspended saliva-impregnated cotton swabs were then spun in a centrifuge at 1000 × g for five minutes, causing the saliva to collect at the bottom of the centrifuge vessel. Saliva samples were frozen for transport to a laboratory for melatonin radioimmunoassay (Pharmasan, Osceola, WI). The sensitivity of the saliva assay was 0.7 pg/ml and the intra- and inter-assay coefficients of variability (CVs) were 12.1% and 13.2%, respectively.

EEG and ECG

The Biosemi ActiveTwo system with active electrodes was used for EEG recordings. This system is battery powered, minimizing electrical interference from alternating current (ac) during recording sessions. Electrodes were placed on subjects' scalps according to the International 10-20 system at Oz, Pz, Cz, and Fz [25]. Two additional electrodes serving as virtual reference electrodes for those attached to the scalp were attached to the right and to the left earlobes. Another electrode was placed approximately 5 cm below the left clavicle to measure an electrocardiogram (ECG) signal.

Near the end of each dark and each light exposure period, just before collecting saliva samples, the scalp electrodes on each subject were attached to the EEG recording system. Six minutes of data were collected: three one-minute periods with the subject's eyes closed alternating with three one-minute periods with the eyes open. When the eyes were open and subjects were not sitting at the light box (dark condition), the subjects were asked to fixate on a specific marked point approximately one meter away. Similarly, when sitting at the light box, subjects fixated on specific point on the far wall of the box approximately 0.6 m away. Subjects were visually monitored by an experimenter to ensure compliance with the protocol.

The EEG signals were sampled at 16384 Hz and then low-pass filtered and downsampled to 2048 Hz for electronic storage by the Biosemi system. All subsequent EEG data processing and analyses were performed with Matlab, version R2008a, by The Mathworks™. The signals recorded from the two reference channels were averaged and these values were subtracted from those obtained from all of the other channels. The direct current (dc) offset of each channel was eliminated by subtracting the mean value of each channel from itself. A low-pass finite impulse response (FIR) filter (f-_3dB = 50 Hz) was applied and the data were downsampled to 512 Hz. Then a high-pass, 3^rd order Butterworth filter (f-_3dB = 4 Hz) was applied to the downsampled signals from each channel to eliminate slow trending in the data.

Another program divided the filtered data into 5-second epochs, segregated by periods when the eyes were open and when they were closed during the six-minute recording period. Eye blink artifacts were eliminated by removing epochs from all channels where voltage fluctuations of any epoch exceeded ± 100 μV. A Blackman window followed by a fast Fourier transform (FFT) was then applied to the data segments. This process yielded spectral power distributions from 1-50 Hz. The power spectra for each one-minute segment were then combined to give an average spectral power distribution for each trial. The relative power levels for eyes open in the theta (5-7 Hz), alpha-theta (5-9 Hz), alpha (8-12 Hz), beta (12-30 Hz), and gamma (30-50 Hz) ranges were calculated as a percentage of overall power from 1-50 Hz [4]. These calculations were not performed for those intervals when the eyes were closed because these data were only used for the AAC (below).

The AAC was computed from the ratio of the alpha power of eyes closed to the alpha power of eyes open for each of the three sessions in the six-minute recording period [26]. The three AAC values for a single measurement period were then averaged to arrive at an AAC value for that condition. One AAC value was calculated for each of the four channels.

The ECG data were digitally processed the same way as the EEG data up to the high-pass filtering. For the ECG analysis, the high-pass filtering -3dB cut-off was lowered to 0.2 Hz. Heart rates corresponding to the filtered ECG data were determined by two methods: 1) by taking the FFT of the ECG, whereby the frequency having the peak power within the range from 40 to 120 beats/minute is the heart rate, and 2) determining the mean elapsed time between successive QRS complexes of the ECG. The QRS complex represents ventricular depolarization. It is called a complex because there are three different waves in it (Q-wave, R-wave, and S-wave). The QRS complexes were located by the first derivative of the ECG falling below a negative threshold value (corresponding to the R-wave to S-wave transition) after individual normalization of the first derivative of the ECG. The values of the first derivative of the ECG at the QRS complexes were typically five to 10 times greater in magnitude than at any other part of the waveform, which enabled this simple algorithm to reliably find the QRS signatures. To further guard against spurious values and artifacts in the ECG data from affecting the mean heart cycle period and corresponding heart rate, only those periods between the 5% and 95% quantiles were included in the calculations of the reported means.

EEG

Since the a priori hypothesis was that the circadian system mediated light-induced nocturnal alertness, the experiment was designed so that the two levels of red light exposure would serve as controls for the two levels of blue light exposure. Initially then, a pair of repeated measures ANOVAs (two [light spectra] by two [light levels] by four [channels]) was performed, one using the percent power in the EEG recordings for the alpha frequency range (8-12 Hz) and the other using the percent power for the beta frequency range (12-30 Hz), as recorded from four scalp electrode channels (Oz, Pz, Cz and Fz). There was no significant main effect of light spectra (F_1,13 = 0.005, ns, and F_1,13 = 1.09, ns, for alpha and beta power, respectively) or of light levels (F_1,13 = 0.03, ns, for both measures), nor was the interaction between these independent variables statistically significant for either dependent variable (F_1,13= 3.27, ns, and F_1,13 = 3.65, ns, for alpha and beta, respectively). There was a significant effect of channels (F_3,39 = 38.5, p < 0.0001, for alpha and F_3,39 = 4.33, p < 0.01, for beta). As a check against potentially large individual differences masking the underlying treatment effects of interest, a second pair of ANOVAs was performed comparing the difference, for each subject, in alpha power and the difference, for each subject, in beta power between measurements made in the dark and in the subsequent light conditions. Again, there were no significant main effects for light spectra (F_1,13 = 0.16, ns, and F_1,13 = 0.89, ns, for alpha and beta, respectively) or light levels (F_1,13 = 0.61 and F_1,13 = 0.19, ns, for alpha and beta, respectively) nor was the interaction statistically significant using the differences in EEG power for either the alpha or the beta frequencies (F_1,13 = 2.78, ns, for both frequencies). Again, however, there was a significant effect of channels.

These findings show, in effect, that light of either spectrum or either level is equally effective for inducing nocturnal alertness because there was no statistically reliable difference between the blue light and the red light exposures at either light level. In other words, the EEG power averages in the alpha frequency range and in the beta frequency range were statistically the same for all four of the lighting conditions. Importantly, the relative alpha power averages obtained under each of the four lighting conditions were always lower than those obtained in the preceding dark conditions, and the relative beta power averages were always higher than those obtained in the previous dark conditions.

Since there was no difference among the four lighting conditions in their effectiveness for inducing alertness at night compared to the previous darkness, the aggregated effects of light versus dark were examined using a third pair of repeated measures ANOVAs (two [dark versus light] by two [first and second sessions within the same night] by two [first and second nights] by four [channels]) was performed using the relative alpha power and the relative beta power as dependent variables. There was a significant main effect of dark versus light for both the alpha and the beta frequencies (F_1,13 = 6.34, p = 0.03 for alpha and F_1,13 = 17.2, p = 0.001 for beta). No other term in the two ANOVAs interacted with this key variable. This analysis demonstrates, as inferred indirectly from the initial ANOVAs, that, simply, the four lighting conditions had an alerting effect on subjects.

Since there were no significant interactions between the variable dark versus light and any of the other variables in the first two ANOVAs, the aggregate alpha power means for dark and for light and the aggregate beta power means for dark and for light were determined. These averages are shown in the top pair of panels in Figure 3; the comparison between light and dark for the alpha frequencies is shown in the left panel and for the beta frequencies in the right panel. To more precisely understand the effects of light spectra and light levels on nocturnal alertness, the aggregated mean alpha and beta power values for light and for dark were decomposed into smaller averages whereby, for example, the mean alpha power for red could be compared to the mean alpha power for the associated dark condition preceding that light condition. Figure 3 shows, in each pair of panels, how the aggregated mean values from all four channels for light and for dark were broken down. As this figure illustrates, eight pairwise comparisons were performed. To correct for multiple comparisons, the criterion alpha level (i.e., p < 0.05) was adjusted in accordance with the Bonferroni/Dunn method to p < 0.00625. Two-tail paired t-tests were performed for the alpha power and for the beta power using the combined data from all four channels.

The first t-test showed that the power in the alpha frequencies from all channels for combined dark conditions was higher than that for the combined light conditions (p < 0.0001); the power in the beta frequencies for the combined dark conditions was lower than the beta power for the combined light conditions (p < 0.0001). These two statistical comparisons simply mirror the results of the third ANOVA where the main effect of the variable dark versus light was significant. These effects are illustrated in the pair of panels labeled A in Figure 3; the comparison between light and dark for the alpha frequencies is shown in the left panel and for the beta frequencies in the right panel.

With regard to the combined red light conditions compared to the combined dark conditions preceding the red light exposures, the power in both alpha and beta frequencies were significantly different, in the expected direction (alpha, p < 0.0001; beta, p = 0.0002). These effects are illustrated in the pair of panels labeled B in Figure 3. The combined blue light conditions were significantly different than the combined dark conditions that preceded the blue light conditions for both alpha and beta power, in the expected direction (alpha, p < 0.0001; beta, p < 0.0001). These effects are illustrated in the pair of panels labeled C in Figure 3. Data for the 10 lx conditions were significantly different, in the expected direction, than the data associated with the preceding dark conditions using both alpha power (p < 0.0001) and beta power (p < 0.0001), as illustrated in the pair of panels labeled D in Figure 3. Similarly, the data for the 40 lx conditions were significantly different, in the expected direction, than the data associated with the preceding dark conditions for both alpha power (p = 0.0002) and beta power (p < 0.0001), as shown in the pair of panels labeled E in Figure 3.

Alpha power was higher for the dark condition preceding the blue-10 lx than for the blue-10 lx condition, although this difference (p = 0.007) did not reach the adjusted alpha criterion. The difference in alpha power was, however, significantly higher for the dark condition preceding the blue-40 lx condition than for the blue-40 lx condition (p < 0.0001). These differences are illustrated in the left panels of H and I, respectively, in Figure 3. For beta power, the same ordering effects were observed, but in the opposite, expected direction; beta power was significantly lower for the dark condition preceding the blue-10 lx than for the blue-10 lx condition (p = 0.0006), and lower for the dark condition preceding the blue-40 lx condition than for the blue-40 lx condition (p < 0.0001), as illustrated in the right panels of H and I, respectively, in Figure 3. Power in the alpha and beta frequencies were not significantly different for the red-40 lx condition and the previous dark condition (panel G of Figure 3), whereas power in both, alpha and beta frequencies were significantly different in the red-10 lx condition compared to the preceding dark condition (p < 0.0001 for alpha and beta) (panel F of Figure 3). The findings for the blue light conditions were expected, but those for the red light conditions were not expected; a greater effect should have been seen for the 40 lx exposure than for the 10 lx exposure. This lack of statistical significance may, in part at least, reflect the lower (p < 0.05) alpha power level associated with the dim condition preceding the R40 condition than with any of the other the dim condition.

A repeated measures ANOVA (two [dark versus light] by two [first and second sessions within the same night] by two [first and second nights] by four [channels]) was performed using the AAC values computed from the alpha power with the eyes closed to the alpha power with the eyes opened (data not shown). Several maximum amplitude criteria were applied to the raw alpha power (47 μV, 75 μV, 100 μV) to minimize artifacts in the data (however, AAC for one subject could not be determined, so the ANOVAs included data for 13 subjects). As expected, the channels (F_3,36 = 21.28, p < 0.0001) were statistically different. The main effect of night (F_1,12 = 5.61, p = 0.04) was also statistically significant. From this measure it would seem that subjects were more tired on the second night than on the first night. There was a statistically significant interaction between night and channel (F_3,36 = 3.83, p = 0.02). This result reflects the fact that the differences in AAC among the four channels were smaller on the second night than on the first night. The interaction between light and channel (F_3,36 = 3.21, p = 0.03) was also statistically significant. Parallel to the night by channel interaction, the differences in AAC among the four channels were smaller for the light condition than for the dark condition. These significant interactions do not have any obvious implication and since there was no main effect of light, the findings from the AAC will not be discussed further.

ECG

As mentioned in the Methods section, the heart rate data were analyzed using two different techniques, FFT and QRS. Given the results of the series of statistical analyses conducted for the EEG data, two repeated measures ANOVAs (two [dark versus light] by two [first and second sessions within the same night] by two [first and second nights]) was performed using data from both the FFT and the QRS techniques. Only the main effect of sessions was statistically significant (F_1,13 = 16.3, p = 0.004 using the FFT technique; F_1,13 = 12.0, p = 0.005 using the QRS technique) indicating, as might be expected, that heart rate was significantly lower during the second session (later at night) than during the first. Although there was no statistically significant effect of dark versus light, paired one-tail t-tests were conducted to determine if specific lighting conditions were statistically different than their previous dark condition. The first t-test showed a statistically significant increase in heart rate after exposure to the blue-40 lx condition compared to the prior dark period for both techniques (FFT, p = 0.003; QRS, p = 0.04). The mean ± standard error of the mean (s.e.m.) heart rate in the blue-40 lx condition was 64.4 ± 1.2 for the FFT and 63.6 ± 1.1 for the QRS. The mean ± s.e.m. heart rate in the dark preceding the blue-40 lx condition was 61.8 ± 1.0 for the FFT and 62.3 ± 1.1 for the QRS. For the red-40 lx condition, only the QRS method showed a significant increase in heart rate relative to the previous dark condition (p = 0.04). The mean ± s.e.m. heart rate in the red-40 lx condition was 61.9 ± 1.1 for the QRS, and in the dark preceding the red-40 lx condition it was 61.1 ± 1.1. No significant differences in heart rate were found using either technique at the lower level (10 lx) of either blue or red light. Although the ECG results are weaker than those from the EEG data, they do suggest that sufficient irradiance at the cornea of either red or blue light exposure can increase alertness at night.

PVT and KSS

Each of the outcome measures obtained from the PDA (RT, TCRT, MTS, and KSS) were submitted to the same series of repeated measures ANOVAs as the ECG variables, with one additional factor, trial: (i.e., two [dark versus light] by two [first and second sessions within the same night] by two [first and second nights] by three [trials]). Three successive trials were conducted during every light and every dark condition, one at the beginning, one in the middle, and one at the end of the 60-minute light exposure or dark condition. One subject did not follow instructions; that subject's data were not included in the analyses.

For KSS, the main effects of session (F_1,12 = 46.14, p < 0.0001), trial (F_2,24 = 22.06, p < 0.0001), and light (F_1,12 = 34.98, p < 0.0001) were statistically significant, as was the session by light interaction (F_1,12 = 11.19, p = 0.006). The three main effects demonstrate increasing sleepiness throughout the experiment. The second session was associated with higher values of KSS than in the first session; Paired two-tail t-tests showed that KSS values associated with the second trial and with the third trial were significantly higher than those in the first trial (p < 0.0004 and p < 0.0001, respectively) although values for the second trial were not significantly different than those associated with the third trial. KSS values were significantly higher (p < 0.0001) in the light exposure condition than in the dark, but since every light exposure condition followed a dark condition, even the significant main effect of light very likely reflects growing sleepiness by the subjects throughout the experiment. Interestingly, however, the significant interaction between session and light suggests that light, in fact, served as a countermeasure for fatigue: a two-tail paired t-test revealed that the difference in reported sleepiness between session 1 and session 2 was significantly greater (p < 0.0001) in the dark than in the light.

In terms of the performance measures, only the TCRT task showed a significant main effect for trial with response time (F_2,24 = 6.06, p < 0.007) and with score (F_2,24 = 8.43, p < 0.002). For both TCRT measures, performance was consistently worse as the trials progressed (i.e., response time increased and score decreased). Consistent with the KSS data, these results indicate that fatigue played an important role in this study. Moreover, the trial by night interaction was statistically significant for both TCRT measures (F_2,24 = 4.98, p < 0.02, for response time; F_2,24 = 4.08, p < 0.03, for score), suggesting that, consistent with the AAC results, fatigue played a greater role in the second night than in the first night; paired two-tail t-tests showed that the difference in TCRT for response time from the first trial to the third trial was significantly greater (p = 0.02) on second night than on the first night, but the difference in TCRT for score did not reach significance (p = 0.1). No effects of light exposure, either among the different light exposure conditions or relative to the previous dark conditions, were demonstrated by any of the performance measures.

Melatonin

Melatonin suppression for each lighting condition was calculated for every subject using the following formula: 1 - (melatonin in the light/melatonin in the dark). If light exposure is a strong stimulus to the circadian system, it will suppress melatonin levels below those measured in the previous dark condition, whereas if it is a weak stimulus, light exposure may not fully counteract the natural rise in melatonin levels that occurs during the early nighttime, in which case negative suppression will be observed. The suppression values associated with three of the four lighting conditions were then determined from those obtained from all 14 subjects; melatonin levels could not be detected from one subject's saliva sample for the blue-40 lx condition, so the average suppression for this condition was based on 13 subjects. Using these data, a two [light spectra] by two [light levels] repeated measures ANOVA was performed; neither the main effects were significant (light color, F_1,12 = 0.3, ns; light level F_1,12 = 1.9, ns), nor was the interaction (F_1,12 = 1.1, ns). However, melatonin concentrations were suppressed by 18% ± 15% (average ± s.e.m.) after exposure to the blue-40 lx condition relative to the preceding dark condition, whereas negative suppression values were determined for the other three lighting conditions (-62% ± 33% for red-10 lx, -39% ± 24% for red-40 lx, and -96% ± 82% for blue-10 lx). In other words, melatonin levels were significantly lower for the blue-40 lx condition than for the dark conditions preceding the blue-40 lx exposures, supporting the literature [12] that this lighting condition was the strongest circadian stimulus. In fact, post hoc one-tail paired t-tests revealed a significantly higher level of melatonin suppression for the blue-40 lx condition than for either the red-40 lx condition (p = 0.03) or the red-10 lx condition (p = 0.02). There was no significant difference between the blue-10 lx condition and the blue-40 lx condition (p = 0.09).

It must be noted that, because of the counterbalanced experimental design, the average suppression values used in these statistical analyses were determined from combined suppression values obtained at two different circadian times. In the counterbalanced experimental design, half the subjects would have been presented with the 10 lx conditions before the 40 lx conditions, and half the subjects would have been presented with the blue light conditions before the red light conditions. Thus, average suppression values for every lighting condition were based upon data obtained during both the early and the later exposure times in the sessions. Since the rate of melatonin synthesis changes throughout the night, melatonin suppression values will be differentially affected by the sample time. Specifically, the same light stimulus presented early in the night will result in less suppression than when it is presented later in the night because the rate of melatonin synthesis is high in the early night, reaches a peak value in the middle of the night, and then decreases until the end of the night [27].