Multi-scale glycemic variability: a link to gray matter atrophy and cognitive decline in type 2 diabetes

Abstract

Objective: Type 2 diabetes mellitus (DM) accelerates brain aging and cognitive decline. Complex interactions between hyperglycemia, glycemic variability and brain aging remain unresolved. This study investigated the relationship between glycemic variability at multiple time scales, brain volumes and cognition in type 2 DM.

Research design and methods: Forty-three older adults with and 26 without type 2 DM completed 72-hour continuous glucose monitoring, cognitive tests and anatomical MRI. We described a new analysis of continuous glucose monitoring, termed Multi-Scale glycemic variability (Multi-Scale GV), to examine glycemic variability at multiple time scales. Specifically, Ensemble Empirical Mode Decomposition was used to identify five unique ultradian glycemic variability cycles (GVC1-5) that modulate serum glucose with periods ranging from 0.5-12 hrs.

Results: Type 2 DM subjects demonstrated greater variability in GVC3-5 (period 2.0-12 hrs) than controls (P<0.0001), during the day as well as during the night. Multi-Scale GV was related to conventional markers of glycemic variability (e.g. standard deviation and mean glycemic excursions), but demonstrated greater sensitivity and specificity to conventional markers, and was associated with worse long-term glycemic control (e.g. fasting glucose and HbA1c). Across all subjects, those with greater glycemic variability within higher frequency cycles (GVC1-3; 0.5-2.0 hrs) had less gray matter within the limbic system and temporo-parietal lobes (e.g. cingulum, insular, hippocampus), and exhibited worse cognitive performance. Specifically within those with type 2 DM, greater glycemic variability in GVC2-3 was associated with worse learning and memory scores. Greater variability in GVC5 was associated with longer DM duration and more depression. These relationships were independent of HbA1c and hypoglycemic episodes.

Conclusions: Type 2 DM is associated with dysregulation of glycemic variability over multiple scales of time. These time-scale-dependent glycemic fluctuations might contribute to brain atrophy and cognitive outcomes within this vulnerable population.

Figure 1. The Multi-Scale Glycemic Variability method applied to 3-day continuous glucose monitoring (CGM).

The decomposition is based on the EEMD (Ensemble Empirical Mode Decomposition) technique. (A) The original CGM time-series from a representative control subject (male, 72 years old, HbA1c = 5.2%, SD (standard deviation) = 10.44, MAGE (mean average glycemic excursions) = 31.61) is decomposed into five glycemic variability cycles (GVCs) that are each characterized by fluctuations within a specific frequency band. The bold black lines along the X-axis denote sleep periods defined by actigraphy and patient records. (B) Comparison of raw CGM signals and selected GVCs between the control subject in (A) and a representative patient with type2 DM (male, 62 years old, HbA1c = 9.4%, SD = 71.36, MAGE = 161.85). The shading areas denote sleep for the type 2 DM patient, while the bold black lines along the X-axis denote sleep for the control subject.

Figure 2. Day and night Multi-Scale Glycemic Variability in older adults with and without type 2 DM As compared to controls, the type 2 DM group had greater variability during the day in GVC_2–5, and night GVC_3–5.

At night, glycemic variability declined in type 2 DM in GVC_2–4 and in controls in GVC₃. ‘*’ (P = 0.002) and ‘‡’ (P<0.0001) indicate significant differences between diabetics/day and controls/day; ‘∥∥’ (P<0.0001) indicates significant differences between diabetics/night and controls/night; ‘†’ (P = 0.003) and ‘§’ (P<0.0001) indicates significant differences between diabetics/day and diabetics/night; ‘¶’ (P = 0.028) indicates significant difference between control/day and control/night. All the P values were obtained by ANOVA. Results are presented as mean ± SEM.

Figure 3. Relationships between the fourth glycemic variability cycle (GVC₄) and conventional measures of glycemic control.

The degree of glycemic variability within GVC₄ was highly correlated with SD (A) and MAGE (B), but the areas under the curves of GVC₄ and GVC₅ were greater than SD and MAGE (C). The degree of glycemic variability within GVC₄ was highly correlated several markers of glucose control including HbA1c (D). As with GVC₄ (the cycle linked with meal intake), the example in this figure, similar relationships were observed for all other GVC cycles. The r² and P values represent the least square model fit.

Figure 4. The brain regions associated with Multi-Scale GV.

Higher glycemic variability of GVC_1–3 (period 0.5–2 hours) were associated with lower gray matter (GM) volume (red color; both hemispheres in the cingulate gyrus, hippocampal gyrus, middle and inferior temporal gyrus, insular cortex, the left superior parietal gyrus and right fusiform gyrus), greater GM volume (blue color; the bilateral supramarginal gyrus, left angular gyrus and left middle orbitofrontal gyrus), and greater cerebrospinal fluid (CSF) in the right lingual gyrus (green color).

Figure 5. Group differences of regional GM volumes in left hemisphere and their relationship with Multi-Scale GV.

‘*’ indicates significant differences between the type 2 DM group (white) and controls (grey) in GM volumes (One-Way ANOVA); regional GM volumes in left hemisphere were correlated with Multi-Scale GV for diabetics and/or controls, blue indicates positive correlation, red indicates negative correlation with each GVC, G' = gyrus, ‘#’ indicates we found similar relationship between Multi-Scale GV and GM volumes in the right hemisphere (r² = 0.26–074, P<0.05). The bar graphs are presented as mean ± SEM.

Figure 6. Examples of least squares models indicating negative relationships between Multi-Scale GV and regional GM volumes as well as cognitive performance.

(A) relationship between GVC₂ and GM volume in the left insular cortex; (B) relationship between GVC₁ and GM volume in the right fusiform gyrus; (C) relationship between GVC₂ and GM volume in the left cingulate gyrus; (D) relationship between GVC₂ and overall cognitive performance (composite T score) (diabetics: triangles; controls: circles). We presented r² for the entire model adjusted for age and sex and group, and P values for the specific effect of Multi-scale GV.