Figure 1.

Physiological negative feedback loops show outputs with characteristic dynamic signatures; dysregulation of the circuit causes a shift in dynamics that can be characterized by autocorrelation—either stronger or weaker, depending upon the type of dysregulation. To illustrate a shift towards autocorrelation that is stronger than optimal, here we show three age and gender-matched subjects’ glucose time-series using an implantable MedTronic device, sampled every 5 min over 6.25 days. The glucose time-series produced by the Type 1 diabetic patients are more auto-correlated (self-similar, fractal) than those of the healthy control, in this case reflecting impaired negative feedback as glucose boluses trigger excitatory responses that are only weakly suppressed by insufficient insulin. As shown, detection sensitivity for differences in glucose amplitude varied dramatically during the day, as well as between days; thus, acquisition of random mean values over short periods of time (as typical for functional magnetic resonance imaging (fMRI) experiment, 10 min with TR = 2000 ms yields ~300 samples, which is roughly equivalent to 1 day of glucose measurements) would yield highly variable accuracy. However, even over this same period, patients showed markedly less complexity in their time-series than the healthy control. Using the Hurst exponent, in which maximum complexity is achieved at H = 0.5 with >H corresponding to stronger auto-correlation, our healthy control showed H = 0.68, with patients showing H = 0.82 and H = 0.83, respectively. A similar shift towards autocorrelation is seen in heart rate variability of heart disease patients, for whom the vagus nerve only weakly suppresses sympathetic excitatory responses. In contrast to the two above examples, in which circuit dysregulation is caused by changes in feedback strength, our neurobiological results suggest different, more complex, types of control circuit dysregulation caused by changes in gating and anatomical connectivity that affect feedback lag. These result in time-series with autocorrelations that are weaker than optimal, as shown in Figures 4, 6.