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 (F1,13 = 0.005, ns, and F1,13 = 1.09, ns, for alpha and beta power, respectively) or of light levels (F1,13 = 0.03, ns, for both measures), nor was the interaction between these independent variables statistically significant for either dependent variable (F1,13= 3.27, ns, and F1,13 = 3.65, ns, for alpha and beta, respectively). There was a significant effect of channels (F3,39 = 38.5, p < 0.0001, for alpha and F3,39 = 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 (F1,13 = 0.16, ns, and F1,13 = 0.89, ns, for alpha and beta, respectively) or light levels (F1,13 = 0.61 and F1,13 = 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 (F1,13 = 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 (F1,13 = 6.34, p = 0.03 for alpha and F1,13 = 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 (F3,36 = 21.28, p < 0.0001) were statistically different. The main effect of night (F1,12 = 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 (F3,36 = 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 (F3,36 = 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 (F1,13 = 16.3, p = 0.004 using the FFT technique; F1,13 = 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 (F1,12 = 46.14, p < 0.0001), trial (F2,24 = 22.06, p < 0.0001), and light (F1,12 = 34.98, p < 0.0001) were statistically significant, as was the session by light interaction (F1,12 = 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 (F2,24 = 6.06, p < 0.007) and with score (F2,24 = 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 (F2,24 = 4.98, p < 0.02, for response time; F2,24 = 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, F1,12 = 0.3, ns; light level F1,12 = 1.9, ns), nor was the interaction (F1,12 = 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].