Coordinated human sleeping brainwaves map peripheral body glucose homeostasis

Abstract

Insufficient sleep impairs glucose regulation, increasing the risk of diabetes. However, what it is about the human sleeping brain that regulates blood sugar remains unknown. In an examination of over 600 humans, we demonstrate that the coupling of non-rapid eye movement (NREM) sleep spindles and slow oscillations the night before is associated with improved next-day peripheral glucose control. We further show that this sleep-associated glucose pathway may influence glycemic status through altered insulin sensitivity, rather than through altered pancreatic beta cell function. Moreover, we replicate these associations in an independent dataset of over 1,900 adults. Of therapeutic significance, the coupling between slow oscillations and spindles was the most significant sleep predictor of next-day fasting glucose, even more so than traditional sleep markers, relevant to the possibility of an electroencephalogram (EEG) index of hyperglycemia. Taken together, these findings describe a sleeping-brain-body framework of optimal human glucose homeostasis, offering a potential prognostic sleep signature of glycemic control.

Keywords: NREM sleep; autonomic nervous system; diabetes; glycemia; heart rate variability; insulin resistance; sleep spindles; slow oscillations.

Graphical abstract

Figure 1

Slow oscillations are functionally coupled with sleep spindles (A) In human NREM sleep, slow oscillations (SOs; <1 Hz) are functionally coupled with sleep spindles, such that the phase of the SO modulates the amplitude of the spindle-related frequency band (12–16 Hz). The plot shows the average peak-locked SO calculated across all the participants (black thick line) and the associated time-frequency representation of the coupling strength. Warmer color indicates higher phase-amplitude coupling. The strongest coupling between SO and spindle-related activity occurs ∼0.4 s after the negative peak of the SO. (B) Histogram of the average SO-spindle coupling strength across all participants. The coupling strength is calculated using the normalized direct phase amplitude coupling (ndPAC) method. The circular plot shows the histogram of the preferred phase of the coupling. For most individuals, the maximum coupling occurs near the up phase of the SO (0°). (C) Example of a coupled SO. The thick black line shows the SO-filtered signal (0.3–1.5 Hz), whereas the orange lines show the associated spindle-filtered (12–16 Hz) signal, scaled by a factor of 4 for illustrative purposes. (D) Example of an uncoupled SO from the same individual as in (C). No statistical SO-spindle coupling was detected for this SO (see STAR Methods).

Figure 2

SO-spindle coupling predicts lower next-day fasting glucose in the CFS dataset (A) Partial correlation adjusted for age between the extent of SO-spindle coupling (i.e., the proportion of SOs that are significantly coupled, see STAR Methods) and next-day fasting blood glucose levels. (B) Partial correlation adjusted for age between SO-spindle coupling strength and next-day fasting blood glucose levels. Translucent bars represent 95% bootstrapped confidence intervals. Fasting glucose levels were normalized using a square-root transformation (see STAR Methods). Of note, both coupling measures remained significantly correlated with fasting glucose levels when removing fasting glucose values above 12 (= 144 mg/dL; r = −0.20, p < 0.001 and r = −0.15, p < 0.001, respectively).

Figure 3

SO-spindle coupling during sleep is a prominent marker of glucose homeostasis, in an independent (MESA) dataset (A) Histogram of the average SO-spindle coupling strength across all participants in the MESA dataset. The coupling strength is calculated using the ndPAC method. The circular plot shows the histogram of the preferred phase of the coupling. For most individuals, the maximum coupling occurs near the up-phase of the SO (0°). (B) Partial correlation adjusted for age between the extent of SO-spindle coupling (i.e., the proportion of SOs that are significantly coupled, see STAR Methods) and next-day fasting blood glucose levels in the MESA dataset. (C) Partial correlation adjusted for age between SO-spindle coupling strength and next-day fasting blood glucose levels. Translucent bars represent 95% bootstrapped confidence intervals, in the MESA dataset. Fasting glucose levels were normalized using a square-root transformation (see STAR Methods).

Figure 4

Insulin resistance is significantly correlated with the coupling between SOs and spindle-related activity in the CFS dataset (A) Partial correlation adjusted for age between the extent of SO-spindle coupling (i.e., the proportion of SOs that are significantly coupled, see STAR Methods) and next-day HOMA-IR. (B) Partial correlation adjusted for age between SO-spindle coupling strength and next-day HOMA-IR. Translucent bars represent 95% bootstrapped confidence intervals.

Figure 5

SO-spindle coupling is the top sleep predictor of next-day glucose homeostasis (A) Top sleep predictors of lower next-day fasting glucose, ranked in descending order of significance (negative log10 p value). (B) Top sleep predictors of lower next-day insulin resistance (HOMA-IR) ranked in descending order. The proportion of SOs with significant coupling was the best sleep predictor of both fasting glucose and insulin resistance. Unadjusted two-tailed p values were obtained by fitting, for each sleep predictor separately, a multilevel regression model adjusted for age, gender, BMI, race/ethnicity, hypertension, and family identification. A total of 47 sleep parameters were included in the rank analysis. NREM refers to N2 + N3 sleep (N1 excluded). A full description of these parameters is provided in Tables S9 and S10.