In order to examine the dynamics of frequency and topography across successive oscillations within spindle discharges, wavelets were used to estimate the power in one Hertz steps from 10–16Hz, from the beginning to end of each individual spindle, separately for EEG and MEG. As is illustrated by the example spindle in , many EEG spindles showed a typical spatiotemporal pattern, wherein higher frequencies occurred earlier in the spindle and in more posterior contacts, with lower frequencies in more anterior contacts being relatively more prominent later in the spindle.
Spatiotemporal evolution of different frequency during an example spindle
In order to quantify this phenomenon, power in low (12 Hz) and high (14 Hz) frequency bands was measured in each spindle burst during an early (25th to 45th percentile of spindle duration) and late (55th to 75th percentile of spindle duration) period, at each sensor, for each spindle. Topographical maps of these measures for EEG clearly show very regular differences (). Specifically, higher frequencies are maximal in the central midline, and are larger at the beginning of the spindle (), whereas lower frequencies are maximal in the frontocentral midline, and are larger at the end of the spindle ().
Topography of power in higher and lower frequencies over the course of the spindle discharge
The significance of these differences were examined with an ANOVA using factors of frequency (low and high), period in the spindle (early and late), anterior-posterior location, and medio-lateral location. For the location analysis, power was averaged for all of the channels in each of three regions: anterior, middle and posterior. Significant main effects were found for frequency (F(1, 6100)= 308.08, p<0.00001) and region (F(2,6100 = 47.17, p<0.00001) but not for period (F(1,6100)= 1.58, p<0.208). Examination of the marginals suggests that these effects were due to an overall greater amplitude of the higher frequencies (), in the central midline (). Also significant were the interactions of frequency with period (F(1,6100) = 21.13, p<0.00001), and frequency with region (F(2,6100) = 21.89, p< 0.00001). These interactions reflect the change in frequency from high to low over the course of the spindle burst, and the predominance of higher frequencies posteriorly and lower frequencies anteriorly. However, neither the interaction of region with period (F(2,6100) = 0.25, p = 0.78), nor the three way interaction (F(2,6100) = 0.43, p = 0.65), were significant, indicating that for a given frequency, changes between periods were similar in different regions. Thus, for example, although high frequencies were greatest in central sites early in the spindle, they decreased everywhere to about the same degree over the time-course of the spindle burst.
Power in different frequencies periods and locations during spindles
MEG gradiometers were examined to determine if the same spatiotemporal evolution of power in different frequency bands could be found over the course of spindle bursts. In contrast to EEG, gradiometers showed relatively subtle differences between high vs. low frequencies or early vs. late periods ( and ). The same conclusion, that the modulation of MEG by frequency is present but less pronounced than for EEG, was confirmed using quantitative measures. While EEG low frequency power increased 35.1% from early to late in the spindle (), MEG only increased 6.9% (). Similarly, while EEG low frequency power decreased 18.4% from early to late in the spindle (), MEG only decreased 9.4% (). Furthermore, there was a general tendency for MEG power to be approximately equal for low and high frequencies, whereas EEG was much larger at the higher frequency. Specifically, early in the spindle, 14 Hz power was 231% greater than 12 Hz in EEG, but only 9% larger in MEG. Similarly, late in the spindle 14 Hz power was 100% greater than 12 Hz for EEG, but was 8% smaller for MEG.
ANOVA was again performed with factors of frequency, period within the spindle, and three regions. Unlike EEG, the main effect of frequency (F(1,6552) =0.21, p = 0.64) was not significant. Like EEG, the main effect of period (F(1,6552) = 0.53, p = 0.46) was also not significant, but the effect of region was (F(2,6552) =106.02, p <0.00001). However, for EEG, this reflected the concentration of power in midline sites, whereas for MEG, this reflected the concentration of power in lateral sites (). Also like EEG, the period by region (F(2,6552)= 0.49, p = 0.61), and three-way interactions (F(2,16552) =0.20, p = 0.81) were not significant, but the period by frequency (F(1,6552) = 13.35, p < 0.0003) interaction was significant. The period by frequency effect is due to an increase of low frequency power and decrease of high frequency power from early to late within the spindle, i.e., the same pattern as found for EEG (). Unlike EEG, which showed a more posterior localization of high frequencies, the MEG frequency by region interaction was not significant (F(2,6552) = 1.57, p = 0.20).
The lack of significance in the period × region, and or period × region × frequency interactions, suggests that the topography of EEG and MEG power in each frequency remains constant between periods, even though their amplitudes change. This was further examined by plotting the topographical differences between higher and lower frequencies, separately for early and late in the spindle, after normalization for overall power changes between periods (bottom row of ) The normalized topographical differences between high and low frequencies are remarkably similar during the early and late periods of the spindle for EEG () as well as for MEG ().
All of the above analyses were carried out by analyzing the mean and variance of the entire sample of spindles. A simple analysis was performed to determine whether this pattern of early/high-frequency, late/low-frequency could also be discerned in individual spindle bursts. Of the 183 spindles, 48% of EEG recordings and 34% of MEG recordings were found to have greater 14 Hz power (averaged across all sensors) early as compared to late in the spindle, and greater 12 Hz power late as compared to early in the spindle (). Under the null hypothesis of no consistent temporal evolution of the spindle burst, the number of spindles that would be expected to show both phenomena is 25%, from which the observed proportion is significantly different at p<.0.00001 for EEG and p = 0.0017 for MEG (binomial test).
Spindles categorized by evolution of high and low frequency power