Global and Regional Effects of Type 2 Diabetes on Brain Tissue Volumes and Cerebral Vasoreactivity

Studies were conducted at the Syncope and Falls in the Elderly Laboratory and at the MRI Center at the Beth Israel Deaconess Medical Center using a GE 3-Tesla VHI scanner with a quadrature head coil. All subjects were recruited consecutively and provided informed consent approved by the institutional review board. The study control groups consisted of 25 healthy subjects who were normotensive and not being treated for any systemic disease and 26 subjects with type 2 diabetes (Table 1). All subjects were screened with a medical history and physical and laboratory examinations. Diabetic subjects were treated with insulin (9), oral glucose control agents (sulfonylureas and second-generation agents (7), meglitinides (1), thiazolidinediones (4), and biguanides (8) and their combinations (13) or with diet (4), as well as for hypertension when clinically diagnosed (8). Diabetic retinopathy was diagnosed with video-digital retinal imaging in 10 diabetic patients in the Joslin Vision Network (14). Statins were used in four control and eight diabetic subjects. Urinary albumin, creatinine, and their ratio were not different between the groups. Diabetic subjects reported symptoms of dizziness (4) and syncope (2), orthostatic hypotension (2), numbness (8), and peripheral neuropathy (3). Subjects with a history of stroke, myocardial infarction, congestive heart failure, and other clinically important cardiac diseases, arrhythmias, significant nephropathy, kidney or liver transplant, renal failure, carotid artery stenosis, and neurological or other systemic disorders were excluded. Subjects with MRI-incompatible metal implants, pacemakers, arterial stents, and claustrophobia also were excluded.

Anatomical imaging protocol included T1-weighted inversion recovery fast-gradient echo (IR-FGE) (T_I/T_E/T_R = 600/3.3/8.1 ms, 24 × 19–cm field of view, 256 × 192 matrix size, and 3-mm slice thickness [no skip]) and fluid-attenuated inversion recovery (FLAIR) (T_I/T_E/T_R = 2,250/161/11,000 ms, 24 × 24–cm field of view, 256 × 160 matrix size, and 5-mm slice thickness [1.5 mm skip]). CASL MRI was used for blood flow measurements (12,13). An echo planar imaging sequence was applied with T_E = 31 ms, 24 × 24 cm field of view, 64 × 64 matrix size, and 5-mm slice thickness (5 mm skip) for eight slices starting at the level of ventricles. Tagged and control images were collected over 5-min periods of normal breathing, CO₂ rebreathing of 95% air and 5% CO₂, and hyperventilation. Images were obtained every 8 s and averaged for each condition. End-tidal CO₂ was continuously monitored and averaged over 15-s intervals for all conditions. A CBF map was reconstructed for each condition, as previously described (ml · 100 g⁻¹ · min⁻¹) (12,13), and corrected for the effect of hematocrit on T1 (15).

All image data were processed on a Linux workstation, using tools developed in the IDL programming environment (Research Systems, Boulder, CO). Figure 1A presents an example of IR-FGE (I1) and FLAIR (F1) images and a T2 reference image used for CBF map reconstruction (C1). First, a three-dimensional ROI corresponding to the parenchymal brain was extracted using the “brain extraction tools” algorithm (16) on the IR-FGE and the FLAIR images and by simple thresholding of the T2 reference image for the CBF maps. Each ROI was divided into eight regions. On each axial slice, an ellipse was fitted to the edge of the brain using a nonlinear least-squares method. The medial axis of the smallest rectangle to enclose the ellipse was computed with the ellipse parameters. This first step allowed the delineation of six regions on all axial slices: the frontal, temporal, and parieto-occipital for each hemisphere were separated using the ellipse major axis (Fig. 1A, I2). Supraventricular slices were analyzed as one region and named the cortical region. This region definition was proposed by Dahl et al. (17) for quantification of regional perfusion using SPECT. It was applied to outline the same eight regions on all ROIs without any registration step, as illustrated on Fig. 1A (I2, F2, and C2). Regions outline major intracranial vascular territories and the frontal, temporal, and parieto-occipital lobes. A relative limitation is that these regions do not follow the precise anatomical boundaries and therefore may include small areas with different functions. Accordingly, our goal was to quantify the global differences in distribution of CBF and vascular reactivity to CO₂ rather than responses to local activation.

The expectation-maximization algorithm was used to assess gray matter, white matter, and cerebrospinal fluid (CSF) volumes by segmenting the IR-FGE ROI into three classes (Fig. 1A, I3). The expectation-maximization method (18,19) estimates iteratively the parameters of a model of the image histogram (three Gaussian distributions) by maximizing the likelihood of the distribution. On the FLAIR ROI, WMH seeds were identified using thresholding of hyperintense pixels. Borders of WMHs were then detected using a simple region-growing method applied on WMH seeds, yielding a connected WMH cluster for each seed (Fig. 1A, F3). The volumes of the whole brain and of the eight regions were computed on the IR-FGE images. Gray matter, white matter, and CSF volumes were normalized for the volume of each region on the segmented IR-FGE images (Fig. 1A, I4). WMH volume was normalized for the volume of each region on the segmented FLAIR images (Fig. 1A, F4). Each CBF map (Fig. 1A, C3) was averaged within each region for each breathing exercise (Fig. 1A, C4). CO₂ reactivity was computed in each region as the slope of the linear fit between end-tidal CO₂ and CBF values for each condition. Relative reactivity also was calculated as a percentage change in CBF and CO₂ between hyperventilation and rebreathing.

Descriptive statistics were used to summarize all variables. Demographic and laboratory variables were compared between groups using one-way ANOVA and Fisher's exact test. The volumes (whole brain and in regions), the normalized gray and white matter, CSF, and WMH regional volumes were compared between the control and diabetic groups using the least-square models with adjustments for age, sex, regions, and hemisphere side. CBF and CO₂ reactivity comparisons between the groups and conditions (baseline, CO₂ rebreathing, and hyperventilation) also include hematocrit and BMI as covariants. Stepwise multiple regression and least-square models also were used to test the effects of A1C, triglycerides, systolic blood pressure, and BMI as continuous variables in the model. The effects of diabetic retinopathy, hypertension, and neuropathy were included as nominal variables and evaluated using the same approach.

Table 1 compares the demographic characteristics and intracranial brain volume of the control and the diabetic groups. Age, sex, and race did not differ between the groups. Diabetic subjects had higher BMI (P = 0.01), A1C (P < 0.0001), blood glucose (P = 0.002), systolic blood pressure (P = 0.04), and cholesterol (P = 0.007) but lower hematocrit (P = 0.02).

The regional volumes were similar between the groups, except for the cortical region, which was smaller in the diabetic group (P = 0.007). Figure 1B–E shows normalized volumes of white and gray matter, CSF, and WMHs in the frontal, temporal, parieto-occipital, and cortical regions for both groups. Over all regions, the diabetic group had smaller white (P = 0.006, Fig. 1B) and gray (P = 0.001, Fig. 1C) matter volume but larger CSF volume (P < 0.0001, Fig. 1D) than the control group. This difference also was observed for white matter (−13.1%, P = 0.0008) and CSF (36.3%, P < 0.0001) in the frontal region, for CSF (20.9%, P = 0.0002) in the temporal region, and for gray matter (−3.0%, P = 0.04) and CSF (17.6%, P = 0.01) in the parieto-occipital region. In the control and diabetic groups, normalized WMH volume differed between regions (P < 0.0001). WMH volume was greater in the temporal (P < 0.0001) and frontal regions and smaller in the parieto-occipital (P = 0.03) and cortical (P < 0.0001) regions (Fig. 1E). WMH volume increased with age (P < 0.0001, r = 0.49) and was higher in the left hemisphere (P = 0.004). Presence of diabetic retinopathy was associated with greater CSF within the temporal region (r = 0.58, P = 0.0003). Diabetic subjects with hypertension had increased CSF volume (P < 0.0001) and decreased gray matter volume (P = 0.008) and greater CSF volume in frontal (P < 0.0001) and temporal (P = 0.03) regions and greater gray matter volume in the frontal region (P = 0.05) compared with normotensive diabetic subjects. In the diabetic group, A1C was associated with more atrophy (CSF: r = 0.72, P = 0.002) within the temporal region.

Figure 2 shows T2 reference images and the corresponding CBF maps for baseline, CO₂ rebreathing, and hyperventilation for a control subject (Fig. 2A–D) and a diabetic subject (Fig. 2E–H). CBF differed among test conditions (P < 0.0001) in both groups. During baseline rest, the regional CBF was higher in the parieto-occipital (P < 0.0001) and lower in the frontal (P = 0.0005) and temporal (P = 0.02) regions in both groups (Fig. 2I). Over all regions, CBF was lower in the diabetic group than in the control group during normal breathing (P = 0.006) and CO₂ rebreathing (P = 0.001) but was similar during hyperventilation. Figure 2J shows group CO₂ reactivity values that were lower in the diabetic group than in the control group (P = 0.005) over all regions. Relative CO₂ reactivity values also were lower in the diabetic group (P = 0.02). For all conditions, CO₂ values did not differ between groups (control: 36.1 ± 5.3 mmHg; diabetes: 37.0 ± 4.3 mmHg during baseline rest).

CBF during baseline was positively associated with regional gray matter volume (r = 0.77, P < 0.0001), reflecting CBF decline in frontal regions (P < 0.0001) and increases in the parieto-occipital regions (P = 0.003) in both groups. In the frontal region, steeper slope of regression (4.99 vs. 2.7) indicated greater CBF decline in the diabetic group. WMHs were associated with reduced CO₂ reactivity (r = −0.54, P = 0.0004) in the control group and contributed to regional differences in vasoreactivity (P = 0.001) in the diabetic group. In the diabetic group, retinopathy and hypertension were associated with lower CBF during hypercapnia (r = −0.61, P = 0.0008 and r = −0.61, P = 0.002, respectively) and hypocapnia (r = −0.77, P = 0.05 and r = −0.80 P = 0.003, respectively). These associations were observed within the temporal region, during hypercapnia for retinopathy (P = 0.03), and during hypercapnia (P = 0.02) and hypocapnia (P = 0.04) for hypertension. Higher BMI was associated with lower CBF in both groups (r = −0.66, P < 0.01). In the diabetic group, higher A1C was associated with lower CBF (r = −0.78, P < 0.0001) and lower CO₂ reactivity (r = −0.45, P = 0.009). Higher baseline systolic blood pressure was associated with lower CBF (r = −0.41, P < 0.002). Triglycerides had no significant effect on CBF. Retinopathy was associated with lower CO₂ reactivity (r = −0.47, P = 0.03).

This study demonstrated that cortical and subcortical atrophy in type 2 diabetes is associated with diminished regional cerebral perfusion and vasoreactivity. Brain atrophy, as indicated by CSF volume, was prominent in the frontal and temporal regions. Hypoperfusion during baseline rest and diminished CBF response to hypercapnia affected all regions. A1C levels and diabetic retinopathy were further associated with CBF reduction, altered vasoreactivity, and atrophy that was most prominent in the temporal region. This study focused on neuroasymptomatic subjects; therefore, those with significant late complications of diabetes and history of cerebrovascular accidents that may further affect CBF were excluded. Cerebromicrovascular disease and endothelial dysfunction affect multiple brain regions and vascular territories and are associated with CBF dysregulation beyond the brain atrophy.

Hyperglycemia is a unifying mechanism for diabetic tissue damage in the brain, as glucose utilization requires transport through the blood-brain barrier and metabolism to provide energy supply (20). Altered glucose transporter regulation results in intracellular hyperglycemia in endothelial cells and neurons (21,22). Signaling of oxidative stress (23,24) in mitochondria and the endoplasmic reticulum leads to activation of four major cell-damaging pathways. Superoxide also triggers an inflammatory response through the release of proinflammatory cytokines (i.e., tumor necrosis factor α, endothelin-1, and interleukins) and decreased synthesis of nitric oxide. These cascades activate proapoptotic pathways and ultimately cause neuronal cell damage and death. The finding that N-acetyl aspartate (an indicator of functional neuronal mass) was reduced in hypertensive diabetic patients in the areas affected by WMHs supports the notion that neuronal loss in frontal and temporal associative areas may contribute to functional decline in type 2 diabetes (25). In our study, WMHs were associated with regional differences in white matter volume, vasoreactivity, and higher A1C. These results support the theory that hyperglycemia in association with other mechanisms (i.e., those related to oxidative stress) may contribute to neurodegeneration in diabetic brain in addition to endothelial dysfunction.

Our results demonstrating reduced CBF in type 2 diabetes are consistent with CBF abnormalities reported in earlier studies. Semiquantitative assessments of regional CBF using SPECT detected the lower ratio between regions with normal CBF versus areas of reduced CBF in type 1 diabetic subjects (4,5). The ratio was inversely correlated with systolic blood pressure, total cholesterol, and the atherogenic index, and it was positively correlated with HDL and cholesterol. These observations suggest that the age-related CBF reduction may be accelerated by a combination of hyperglycemia plus other risk factors for arteriosclerosis (26,27). The regional differences in cerebral metabolic capacity may explain increased sensitivity to hyperglycemia in the cerebral cortex (28). Previous studies described reduced or normal CO₂ reactivity in diabetes (32,33). This study applying CASL for evaluation of CBF and CO₂ vasoreactivity demonstrated that CASL is a reliable tool for assessment of flow reserve in an elderly diabetic population.

There may be some limitations to this approach. Because of the short decay time of the CASL label (∼1 s), CBF measurements might be affected for longer arterial transit time at low flow as during hypocapnia. It is important to mention that CBF measurements that reflect mainly flow in the gray matter were normalized for the volume of the ROI on which they were computed and also were adjusted for any variability associated with age, sex, hematocrit, and BMI to adjust for the effects of tissue loss and possible effects of other variables and risk factors. Therefore, the observed CBF changes are unlikely to be affected by partial volume effects attributed to an increased CSF. Our region selection approach allowed comparisons among different brain areas’ anatomical and CBF images. As the method did not require any registration step, the definition of identical regions on all ROIs was performed regardless of the differences in matrix and voxel size between IR-FGE, FLAIR, and CASL images. We validated our automated method of WMH quantification, and we found excellent correlations between WMH volume and clinical rating scale (35).

The segmentation method of the IR-FGE image allows accurate classification of noisy pixels and, thus, reliable measurements of CSF volume that have been validated using the phantom model (36). Regional intensity variation caused by an inhomogeneous radiofrequency coil and low signal-to-noise ratio, however, may reduce the accuracy of the brain tissue classification. The electron microscope–based segmentation method was selected because it can estimate the intensity of inhomogenities. In addition, the voxel classification is influenced primarily by the neighboring voxels, reducing the difficulties encountered in noisy regions of the images (37). The relatively thick 5-mm slices and 1.5-mm gaps in the FLAIR images may be a contributor to errors in the measurement of the WMHs. To estimate this uncertainty, a high-resolution model of the WMHs was created with the same parameters. Then, an axial offset to the slices’ position was introduced, and the volume was calculated as a function of this offset. It is estimated that a ±6% uncertainty in the measured WMH volume is introduced by the selected geometry of the FLAIR images (36).

Microvascular and macrovascular disease and neurodegeneration in type 2 diabetes has been associated with accelerated cognitive impairment and dementia in elderly subjects (38). Diagnosis of diabetes doubled the risk for both hippocampal atrophy and lacunar infarcts in a population-based study. Diabetic patients with the longest disease duration, those taking insulin, and those with complications had more severe pathologic brain abnormalities (39). Diabetes also is associated with cortical and subcortical atrophy, WMHs, and impaired cognitive performance (attention and executive function, information processing speed, and memory) (11). Elevated body mass and systemic pressure may further affect cerebral perfusion and contribute to cognitive decline in older people (40).

Some limitations to our approach may arise from the fact that selected regions reflect more vascular territories rather than anatomical regions and that slice thickness on FLAIR images may affect WMH volume estimates. The study was done on an outpatient basis, and laboratory values may include nonfasting samples. This study addressed an important question about the effects of type 2 diabetes on regional CBF distribution and vasomotor reserve in older adults and provided evidence for reduced perfusion beyond gray and white matter atrophy. Higher levels of A1C were associated with lower CBF and greater CSF volume in the temporal region. Reduced resting blood flow may reflect combined effects of microvascular disease and metabolic tissue damage affecting preferentially frontal and temporal regions. These findings are clinically relevant for functional outcomes, such as cognition and balance of elderly people with diabetes. Further study of the mechanisms prospectively relating hypoperfusion to neurodegeneration and functional outcomes in type 2 diabetes is merited.

This study was supported by an American Diabetes Association Grant (1-03-CR-23 and 1-06-CR-25); the National Institutes of Health (NIH), National Institute of Neurological Disorders and Stroke (1R01-NS045745-01A21 [to V.N.]); an NIH Older American Independence Center Grant (2P60 AG08812); an NIH program project (AG004390); and a General Clinical Research Center Grant (MO1-RR01032).