Overall sleep spindle density and duration were compared across nuclei performing repeated-measures ANOVA with the factors spindle type (slow vs. fast) × nucleus (ANT vs. MD), where missing values were handled as missing data. Comparison of SP(ripple/pure) density and duration were performed by repeated-measure ANOVA with the factors spindle type (slow vs. fast) × association (ripple vs. pure) × nucleus (ANT vs. MD). Post-hoc tests were conducted by Tukey’s HSD.
Time-frequency wavelet analysis was performed on 4600 ms long epochs extracted in the -2300 to 2300 ms latency range around sleep spindle onsets (separately for the slow and fast SP(ripple/pure)) on all ANT and MD derivations. Furthermore, time-frequency analysis was also performed on frontal and parietal scalp records using the same time intervals as for the ANT and MD sleep spindles (scalp EEG activity around thalamic sleep spindles). Data from the F3 and F4 electrodes were analyzed for the frontal region, except for patient #15, where Fp1 was used instead of F3 due to the low signal quality. Parietal electrodes were P3 and P4, except for patient #2, where Pz was used instead of P3 and P4 as these electrodes were missing for this patient. Data were averaged across the frontal and parietal electrodes, separately. Time-frequency analysis was performed by Morlet wavelet transformation with linearly increasing cycle numbers from 4 cycles to 20 cycles for the 1–450 Hz frequency range using 0.5 Hz frequency resolution in the 1–40 Hz range, 1 Hz frequency resolution in the 41–80 Hz range and 5 Hz frequency resolution in the 81–450 Hz range. The baseline interval was set at -2000 ms to 0 ms before sleep spindle onset. Time-frequency power spectra were averaged across all derivations within the same nucleus, separately for the fast and slow SP(ripple/pure). Statistical comparison of SP(ripple) vs. SP(pure) was performed by nonparametric randomization test separately for the slow and fast spindles and for the ANT, MD and scalp (frontal and parietal), applying the Monte-Carlo estimate of significance probabilities (1000 permutations), using the Fieldtrip toolbox (Oostenveld et al. 2011), applying FDR correction.
For assessing phase-amplitude coupling between sleep spindle phases and ripple amplitudes in the thalamus, the modulation index (MI) was calculated based on the method by Tort et al. (2010), separately for each thalamic derivation. First, 4600 ms long epochs were extracted around individual sleep spindles with co-occurring ripples (SP(ripple)), in the -2300 to 2300 ms latency range where 0 ms corresponds to the onset of sleep spindles. For all detected SP(ripple) data were band-pass filtered in the frequency bands of interest, 80–200 Hz for extracting the ripple amplitude and individual slow and fast spindle bands, detected by the IAM of sleep spindle analysis (see above) for extracting instantaneous phase of sleep spindles (FIR filter). Bandpass-filtered data were than Hilbert transformed using the 0–2000 ms time range for avoiding edge effects. MI were then calculated for each electrode using 18 phase bins from –π to +π, pooling all phase and amplitude values extracted from the epoched data. The observed MI were then subjected to permutation testing in order to quantify the difference between the observed MI and the distribution of shuffled coupling values. Shuffled MI distribution were calculated by measuring the MI between the original phase time series and permuted amplitude time series where amplitude data points were randomly shuffled, using 1000 iterations for each electrode. The observed MI values were then z-standardized to the shuffled phase-amplitude coupling distribution, where normalized z-values directly reflect p-values, MI(z) equal to 1.645 corresponds to the 5% p-value. Thus, MI(z) values larger than 1.645 reflects a significant spindle-ripple coupling in the thalamus.
We aimed to estimate the time dynamics of spindle co-occurrence between thalamic and cortical channels. Spindle co-occurrence was defined in the following way: when the initiation of a sleep spindle was detected on any channel (cortical or thalamic), all subsequent spindles initiating on any other channel before the end of the original spindle were considered to co-occur, comprising a single spindle event involving multiple channels with a time lag on each channel, defined as the time difference of spindle initiation relative to the first spindle. For thalamocortical co-occurrence analysis, we selected all instances when sleep spindles co-occurred on both 1) a selected scalp channel (F3 or F4 for frontal spindles, P3 or P4 for parietal spindles, for patient #2 Fp2-Fpz and Pz-Oz instead) and 2) on a channel localized in a specific thalamic nucleus (ANT or MD). That is, the analysis included spindles which could originate elsewhere, but were later detected on both specific scalp channels and in the thalamus. We defined thalamocortical spindle lags as the time lag (relative to the first spindle) of the scalp channel minus the time lag of the thalamic channel. We modeled spindle lags with a linear mixed model implemented in the MATLAB 2017a fitlme() function using lag as the dependent variable, spindle type, thalamic nucleus and scalp channel as fixed effects with random intercepts by patient (Ujma et al. 2022).
For assessing the potential correlates and functions of thalamic ripples and spindles Pearson correlations coefficient were calculated between sleep spindle density (scalp-detected parietal fast sleep spindle density, overall thalamic spindle density, thalamic SP(ripple) and SP(pure) density), clinical epilepsy characteristics (years since epilepsy onset and seizures/month (before DBS)), and general intelligence (Table 3 shows the WAIS IQ scores ); separately for the ANT and MD, and for the slow and fast spindles. Relationships between scalp-detected, parietal fast sleep spindle density were measured based on the average recording locations P3 and P4, except Patient #2, where Pz was used.