Coordination of cortical and thalamic activity during non-REM sleep in humans

Every night, the human brain produces thousands of downstates and spindles during non-REM sleep. Previous studies indicate that spindles originate thalamically and downstates cortically, loosely grouping spindle occurrence. However, the mechanisms whereby the thalamus and cortex interact in generating these sleep phenomena remain poorly understood. Using bipolar depth recordings, we report here a sequence wherein: (1) convergent cortical downstates lead thalamic downstates; (2) thalamic downstates hyperpolarize thalamic cells, thus triggering spindles; and (3) thalamic spindles are focally projected back to cortex, arriving during the down-to-upstate transition when the cortex replays memories. Thalamic intrinsic currents, therefore, may not be continuously available during non-REM sleep, permitting the cortex to control thalamic spindling by inducing downstates. This archetypical cortico-thalamo-cortical sequence could provide the global physiological context for memory consolidation during non-REM sleep.


Supplementary Figure 2. The probability of a thalamic downstate increases as the number of cortical channels participating in a cortical downstate increases.
For each subject, the number of downstates occurring within 500ms after one other cortical channel, up to the maximum number of cortical channels, was calculated. For each thalamic channel, the probability of thalamic downstates occurring within 500ms after cortical downstates was calculated for all combinations of cortical channels. When a downstate occurred in two or more cortical channels, the downstate peak of the locked cortical channel was used to search for thalamic downstates. Each colored line represents the mean ± SEM for each thalamic channel. In most thalamic channels, as more cortical channels participate in a downstate, the probability of a thalamic downstate increases. Figure 3. On average, spindles start at the thalamic downstate peak, and end at the following thalamic upstate peak. Histograms of thalamic spindle onsets (first column) or spindle terminations (second column) in relation to the thalamic downstate peak at 0ms (orange vertical lines) for each thalamic channel in 50ms bins. Spectral power from 5-100Hz (third column), or from 5-20Hz (fourth column) averaged on the downstate peaks at 0ms, baseline corrected over entire epoch, thresholded at p<0.01, uncorrected. Channels plotted here and in Fig. 2B comprise all thalamic channels. Waveforms show the averaged local field potential in each channel. Figure 4. High gamma decreases in the thalamus are related to the end of spindles. In Subject 3, the local field potential (0.1 to 4Hz bandpass, thick black line), high gamma amplitude (Hilbert analytic amplitude on data bandpassed from 250 to 500Hz and then bandpassed from 0.1 to 4Hz, blue dotted line), and spindle (10-16Hz bandpass, thin black line) are plotted for the medial pulvinar/lateral pulvinar channel. The spindle trace has been multiplied by 10 for display purposes. Blue arrows indicate the high gamma increases or decreases in each subplot. A. The local field potential (LFP), high gamma amplitude, and spindle locked to the start of spindles. There is a clear high gamma increase after the downstate LFP peak that coincides with the start of spindles. B. The LFP, high gamma amplitude, and spindle locked to the end of spindles. A clear high gamma decrease occurs at the end of spindles. Figure 5. Downstate versus spindle relationship in N2 versus N3 for Subject 1. In each row, the spindle starts occurring within ±2 seconds of N2 downstate (left) or N3 downstate (right) peaks are plotted in 50ms bins for each channel. Vertical orange lines mark the local downstate peaks at 0ms for each channel. The overall shape of the spindle distributions locked to downstates appears similar in N2 and N3 for each channel: thalamic spindles occur clustered around the downstate peak (black histograms), while cortical spindles arise after the downstate peak (red histograms). For each channel, the correlation coefficient between N2 and N3 spindles occurring in relation to downstates was calculated. The location, correlation coefficient, r, and the p-value for the correlation, are calculated for each channel. P-values<0.05, corrected for multiple comparisons, are marked with an asterisk and indicate that the distributions arising in relationship to downstates are significantly correlated between N2 versus N3 spindles. Here, one middle frontal gyrus channel and one post central channel do not have significant correlations; however, the pattern of spindles related to downstates is less clear for this middle frontal gyrus channel than other channels and the number of spindles arising in N2 for this post central channel is small, although they both show a tendency to occur after the downstates. Figure 6. Downstate versus spindle relationship in N2 versus N3 for Subject 2. The spindle starts occurring within ±2 seconds of N2 downstate (left) or N3 downstate (right) peaks are plotted in 50ms bins for each channel. Vertical orange lines mark the local downstate peaks at 0ms for each channel. The overall shape of the spindle distributions locked to downstates appears similar in N2 and N3 for each channel: thalamic spindles occur clustered around the downstate peak (black histograms), while cortical spindles arise after the downstate peak (red histograms). For each channel, the correlation coefficient between N2 and N3 spindles occurring in relation to downstates was calculated. The location, correlation coefficient, r, and the p-value for the correlation, are calculated for each channel. P-values<0.05, corrected for multiple comparisons, are marked with an asterisk and indicate that the distributions arising in relationship to downstates are significantly correlated between N2 versus N3 spindles. All channels show a significant correlation. Figure 7. Downstate versus spindle relationship in N2 versus N3 for Subject 3. The spindle starts occurring within ±2 seconds of N2 downstate (left) or N3 downstate (right) peaks are plotted in 50ms bins for each channel. Vertical orange lines mark the local downstate peaks at 0ms for each channel. The overall shape of the spindle distributions locked to downstates appears similar in N2 and N3 for each channel: thalamic spindles occur clustered around the downstate peak (black histograms), while cortical spindles arise after the downstate peak (red histograms). For each channel, the correlation coefficient between N2 and N3 spindles occurring in relation to downstates was calculated. The location, correlation coefficient, r, and the p-value for the correlation, are calculated for each channel. P-values<0.05, corrected for multiple comparisons, are marked with an asterisk and indicate that the distributions arising in relationship to downstates are significantly correlated between N2 versus N3 spindles. All channels except for the lateral pulvinar/nucleus reticularis have a significant correlation. In this channel, the spindles are still clustered around 0 for N2 and N3, but appear more clustered to the left of 0 in N3.  Figure 10. Thalamic spindles still start at the thalamic downstate peak when stricter thalamic downstate detection parameters are applied. When downstate detection parameters are changed to only select the bottom 20% of downstate peaks whose zero crossings occur between 0.25 to 1secs, the thalamic spindles still start during the thalamic downstate. Histograms of thalamic spindle onsets (left column) in relation to the thalamic downstate peak at 0ms (vertical orange lines) for each thalamic channel in 50ms bins. Spectral power from 5-100Hz (right column), averaged on the downstate peaks at 0ms, baseline corrected over entire epoch, thresholded at p<0.01, uncorrected. All 8 thalamic channels are plotted. Waveforms show the averaged local field potential in each channel. Figure 11. Cortical spindles still start during the cortical down-to-upstate transition when stricter cortical downstate detection parameters are applied. When downstate detection parameters are changed to only select the bottom 20% of downstate peaks whose zero crossings occur between 0.25 to 1secs, the cortical spindles still start during the cortical down-to-upstate transition. Cortical downstate peaks are locked at 0ms (vertical orange lines) and the corresponding cortical spindle onsets are plotted in red 50ms bins for each channel. Waveforms show the averaged local field potential in each channel. All 22 cortical channels are plotted.

Supplementary Table 1.
Discrimination table of the two spindle detection methods. For each subject, the calculations were performed over 10 minutes of N2 and 10 minutes of N3 that had been manually marked for spindles. Cortical and thalamic channels are considered separately for each subject. "Previous" refers to the method adapted from (1), which is described in the Methods. "Current" refers to our current spindle detection method. The Hit Rate was calculated by dividing the number of True Positives by the sum of the True Positives and False Negatives. The False Alarm Rate was calculated by dividing the number of False Positives by the sum of False Positives and True Negatives. True negatives in the current method were candidate epochs that were not automatically selected as spindles and were not manually marked as spindles. In order to approximate the number of true negatives in the previous method, the sum of the False Negatives, False Positives, and True Positives for the previous method was subtracted from the sum of the True Negatives, False Negatives, False Positives, and True Positives separately for the cortex and thalamus for each subject. The d', which measures discriminability, and C, which measures bias, were calculated using a d' calculator (http://memory.psych.mun.ca/models/dprime/). thalamic (B) channels: 0-750ms around the downstate peak for cortical channels and -500ms to + 250ms around the downstate peak for thalamic channels. The proportion of spindles occurring with downstates was then calculated by dividing the number of spindles that occurred within these specified time windows divided by the total number of spindles. The thalamic channels exhibited a higher proportion of downstate-related spindles (0.43 ± 0.18) compared to cortical channels (0.35 ± 0.07). When these proportions were normalized to account for the larger number of downstates occurring in the cortex (see Methods), the normalized proportion of spindles occurring with downstates is even larger between thalamic channels (0.63 ± 0.21) and cortical channels (0.38 ± 0.07). The enrichment factor measures how spindle density increases in a channel when spindles occur in relation to downstates compared to the channel's overall spindle density (see Methods). The thalamic channels show greater enrichment factors (5.62 ± 2.13) compared to the cortical channels (3.15 ± 0.9), further indicating that spindles in the thalamus are more highly modulated by the downstate than in the cortex.

Supplementary Note 1. Convergence of cortical downstates leads to a thalamic downstate
The binomial results indicate a statistically significant tendency for cortical downstates to lead thalamic downstates; however, examination of connected thalamocortical pairs still finds that the number of cortical downstates leading to thalamic downstates remains low ( Supplementary Fig.  1A, boxed histograms). For example, even though the posterior insula downstates strikingly cluster just prior to the central lateral/medial pulvinar downstates, only 1258 out of 5079 posterior insula downstates, or ~25%, are followed by central lateral/medial pulvinar downstates within 500ms. The annectant gyrus and medial/lateral pulvinar pair also exhibit downstates that are tightly coupled in time to one another; however, only 400, or ~15% of the 2622 downstates in the annectant gyrus are followed by downstates in the medial/lateral pulvinar. We therefore hypothesized that a convergence of cortical downstates may be required for thalamic downstates to occur. Due to limited cortical sampling, however, a complete examination of this hypothesis is also limited. For the following analyses, only cortical bipolar pairs that were separated by at least two contacts were included. The hypothesis that convergence from multiple cortical downstates induces thalamic downstates was tested in three ways. First, the percentage of thalamic downstates without a prior cortical downstate (in any cortical channel) within 500ms was examined for each thalamic channel. It was predicted that if a subject has more cortical channels, the percentage of isolated thalamic downstates would decrease. Subject 3 only had two cortical channels and consistent with this prediction, both thalamic channels showed a high percentage of downstates without a preceding downstate in at least one cortical channel: 67% for the Medial/Lateral Pulvinar channel and 89% for the Lateral Pulvinar/Nucleus Reticularis channel. Subject 1 included six cortical channels and Subject 2 included ten cortical channels. In both subjects, the percentage of isolated thalamic downstates decreased by ~50% compared with Subject 3: 37%, 39%, and 42% for each of the three Medial Pulvinar channels in Subject 1, with a continued drop to 35% for the Central Lateral, 30% for the Central Lateral/Medial Pulvinar, and 36% for the Medial Pulvinar/Lateral Posterior channels in Subject 2. We would predict that with robust cortical sampling, thalamic downstates would not occur without preceding cortical downstates.
Second, we examined whether cortical pairs that produce downstates together are more likely to lead to thalamic downstates, compared with cortical downstates produced in only one of these cortical channels. For each pair of cortical channels, the number of downstates occurring within 500ms of the locked channel was calculated. The probability of thalamic downstates occurring within 500ms after these paired cortical downstates was calculated for each thalamic channel; the downstate peak of the locked cortical channel was used to search for thalamic downstates. The probability of thalamic downstates in each thalamic channel occurring within 500ms of isolated cortical downstates was also calculated for each cortical channel individually. Over all thalamic channels, the mean probability of thalamic downstates occurring with cortical pairs (0.14) was significantly greater than when thalamic downstates occur with a single cortical channel (0.10) (paired t-test, p=0.011, 8 thalamic channels). A Chi-squared test was performed between the proportion of thalamic downstates relative to isolated versus paired cortical downstates. Out of 364 unique cortical /thalamic channel combinations, 54 (~15%) were significant after Bonferroni correction at p<0.05. For an example, see Fig. 1G (in the main text). Of these, one showed a greater probability of a thalamic downstate when downstates occurred on the individual cortical channel, but the remaining 53 showed a greater probability of a thalamic downstate when the cortical downstate occurred in both cortical channels. Overall, these results indicate that thalamic downstates are more likely to occur when cortical downstates occur in two related cortical channels.
The third way we tested the convergence hypothesis was by examining the probability of a thalamic downstate as a function of the number of cortical channels showing a downstate. With small and occasional exceptions, the addition of multiple channels participating in a cortical downstate further increased the probability of a following thalamic downstate, as shown in Supplementary Fig. 2. In summary, our results indicate that as a greater number of related cortical channels participate in a downstate, the more likely a downstate will follow in the thalamus. Overall, however, the percentages are relatively low. This may suggest that very widespread convergence is needed, which was not possible to observe given our limited cortical sampling. Alternatively, upstates or another ongoing process not measured here may also be contributing to the generation of thalamic downstates.