Changes in Regional Brain ¹⁸F-Fluorodeoxyglucose Uptake at Hypoglycemia in Type 1 Diabetic Men Associated With Hypoglycemia Unawareness and Counter-Regulatory Failure

We recruited 13 men with C-peptide−negative type 1 diabetes from the Diabetic Clinic at Guy’s Hospital, London. Based on their glycemic control and hypoglycemic experiences, they were characterized as hypoglycemia aware or unaware. This grouping was done using clinical history, measurement of HbA_1c, and home blood glucose monitoring, performed by the patient at least four times daily. A clinical diagnosis of hypoglycemia unawareness required confirmation of at least three episodes every 2 weeks, in which routine home glucose monitoring showed a glucose value ≤3 mmol/l when the patient was asymptomatic (5,11). Hypoglycemia awareness was diagnosed only if there were no such episodes on similar monitoring and no severe hypoglycemia (defined as an episode of hypoglycemia requiring outside intervention) (3). The hypoglycemia-aware group was comprised of six men (age 35.8 ± 4.6 years and diabetes duration 16.3 ± 3.6 years) with good awareness of occasional hypoglycemia. The hypoglycemia-unaware group was comprised of seven men (age 36.9 ± 9.1 years and diabetes duration 16.6 ± 6.1 years), all of whom had a history of severe hypoglycemia and were experiencing recurrent mild-to-moderate hypoglycemia, as defined in our protocol. This was reflected in this group’s lower HbA_1c values compared with the hypoglycemia-aware group (7.4 ± 1.6 vs. 8.9 ± 2.0%; P < 0.05; nondiabetic range >6.1%). No patient in either group had clinically significant chronic diabetic complications. A retrospective check on the accuracy of the classification of our patients was available in the form of catecholamine levels measured during the hypoglycemic studies.

The study was approved by the ethics committees of Guy’s and St. Thomas’ Hospitals and by the Administration of Radioactive Substances Advisory Committee. Each subject gave written informed consent.

Each patient underwent two FDG-PET scanning sessions. One session was performed during controlled normoglycemia (plasma glucose of 5 mmol/l) and the other during moderate acute hypoglycemia (plasma glucose of 2.6 mmol/l). Each session occurred on different days, at the same time of day, 3−6 weeks apart, in random order. The patients were not informed of their glucose levels during scanning.

On a further separate occasion, each subject had a T₁-weighted magnetic resonance image (MRI) whole-brain scan to guide region of interest (ROI) placement. There was no specific attempt at precise glycemic control for this scan.

To achieve normal blood glucose values for the start of each study, all patients were admitted to the hospital the evening before each scan and omitted their usual evening intermediate-acting subcutaneous insulin injection. An intravenous cannula was placed in a vein in the arm, using 1% lidocaine to anesthetize the skin, and was used to sample blood for glucose measurement every 30−60 min. Regular insulin was infused through the cannula to control blood glucose at 4−8 mmol/l, avoiding all nocturnal hypoglycemia. The patients fasted from 10:30 p.m. onward.

In the morning, after checking for the presence of collateral circulation, the radial artery of the nondominant hand was cannulated for blood sampling. A primed continuous infusion (maintenance rate 1.5 mU · kg⁻¹ · min⁻¹) of regular insulin (Human Actrapid; Novo-Nordisk, Sussex, U.K.) in a 4% solution of autologous blood was administered through the intravenous line. Arterial plasma glucose was sampled every 5 min, measured at the bedside, and maintained at 5 mmol/l for 1 h before scanning by adjusting the rate of an intravenous infusion of 20% glucose solution (Baxter Health Care, Thetford, Norfolk, U.K.) (12). During this hour, the patient was made comfortable in the scanner. Plasma glucose was then held at 5 mmol/l (euglycemic study) or reduced to 2.6 mmol/l over 40 min and held at this level for 30 min before and throughout scanning (hypoglycemic study) (13).

On completion of each scan, the insulin infusion was stopped and the subject was given an individualized dosage of subcutaneous regular insulin and lunch. Advice was given about hypoglycemia risk on the next day. After the second scan, hypoglycemia-unaware subjects enrolled in a program of hypoglycemia avoidance.

PET scans were acquired with a CTI ECAT 951R PET camera (CTI/Siemens, Knoxville, TN), with an axial field of view of 10.8 cm and an intrinsic inplane spatial resolution of 6.5 mm (full width at half-maximum [FWHM]). The subject lay with his head in the scanner in a supportive head rest to minimize movement, aligned axially and to the orbitomeatal line (approximating the anterior-posterior commisural line of the Talairach and Tournoux brain atlas [(14]), using two perpendicular laser lines located on the camera gantry. A 10-min transmission scan was performed before tracer injection to correct for photon attenuation. Then ∼185 MBq FDG were injected intravenously over 30 s, and dynamic PET scanning was performed over the following 60 min (15). Arterial activity concentration was measured continuously in blood drawn past a scintillation detector (Allogg, Stockholm) initially at 5 ml/min and then at 2 ml/min by a peristaltic pump (IVAC 572).

All images were reconstructed by filtered back-projection and smoothed with a Hanning filter, so that the spatial resolution was 8.5 mm (FWHM) transaxially and axially. Reconstructed images were displayed in a matrix of 128 × 128 × 31 voxel format, each voxel measuring 2.0 × 2.0 × 3.43 mm (15). FDG uptake images were formed by summation of images acquired at 30−60 min postinjection.

Whole-brain activity of FDG in the two groups (aware and unaware) and the two conditions (euglycemia and hypoglycemia) was determined by summing the activity in all brain slices after application of a 40% threshold to exclude extracerebral activity. FDG uptake values were corrected for injected activity and body weight to remove some of the dependence on these variables, as follows (16):

To examine for regional differences in the FDG uptake images between the two groups (aware and unaware) and the two conditions (euglycemia and hypoglycemia), statistical parametric mapping (SPM) was performed using the SPM96 program (Wellcome Department of Cognitive Neurology, London, implemented in Matlab 4.2c [Mathworks, Sherborn, MA]). SPM is a standard technique for the analysis of brain imaging studies. All images in a study are transformed into a common standard anatomical space. Statistical tests comparing the images are performed at every voxel position. Regions of the brain where there are statistically significant differences are thus identified and displayed in the form of a new parametric image, where the image intensity at any point is related to statistical significance at that point. The SPM map can be displayed superimposed on a standard anatomical image to identify the anatomical locations of statistically significantly different regions.

Each subject’s scan was transformed into standard stereotactic space using a 12-parameter linear transformation, a 6-parameter quadratic deformation, and a fluid nonlinear 3D deformation on a slice-by-slice basis to allow for intersubject averaging and comparisons (17,18). An FDG-uptake template in Talairach space (data from 13 subjects averaged into standardized space) was kindly supplied by Dr. Eraldo Paulesu (San Raffaele Hospital, Milan, Italy). The stereotactically normalized scans contained 26 planes with a voxel size of 2 × 2 × 4 mm, corresponding to the Talairach and Tournoux brain atlas (14). Smoothing using a Gaussian kernel of 20 mm was performed to allow for intraindividual differences in gyral anatomy and to increase the signal-to-noise ratio. The effect of variation in whole-brain FDG uptake among different scans from differing injected activities, plasma glucose concentrations, and plasma FDG concentrations, was removed using a voxel-by-voxel analysis of covariance, with values for mean whole-brain uptake as the confounding covariate, analogous to the procedure described by Reed et al. (19).

Comparisons to determine the effect of hypoglycemia and hypoglycemia unawareness at hypoglycemia were made by definition of the appropriately weighted linear contrasts within SPM96. SPM{t} maps of voxel-by-voxel t statistics were generated and subsequently transformed into SPM{Z} maps (20). In this exploratory analysis, clusters of voxels surviving a threshold Z score >3.09 (omnibus threshold P < 0.001), uncorrected for multiple comparisons, were considered to show significant differences (20,21). This analysis conformed to a mixed random and fixed effects model, as each subject was studied once only in each condition (euglycemia and hypoglycemia). To enhance anatomical localization, the locations of volumes of change were displayed by rendering the maximum intensity map onto orthogonal planes of a high-resolution T₁-weighted MRI brain scan in Talairach space provided with the SPM96 software. For display purposes, a threshold Z score >2.81 (P < 0.0025) was used.

ROIs were placed over the FDG uptake images in the regions identified by the SPM analysis. PET and MRI images for each individual were coregistered using an automatic algorithm (22). Small circular ROIs were placed on each individual’s MRI, and transferred onto the registered PET images. Normalized uptake values for these regions were obtained by dividing the regional uptake values by the whole-brain values, without correcting for injected activity or subject weight. This method has been shown to be a highly sensitive measure of regional variation in FDG uptake (19)

Plasma glucose was measured at bedside using a glucose oxidase method (Yellow Springs Instruments, Yellow Springs, OH). Plasma epinephrine was measured using high-pressure liquid chromatography with electrochemical detection (23), and free insulin was measured by radioimmunoassay (24).

Data are presented as means ± SD unless otherwise stated. Metabolic and hormonal data were compared using unpaired t tests. Whole-brain and regional FDG uptake data were compared between the hypoglycemia-aware and -unaware groups at euglycemia and hypoglycemia by analysis of variance (ANOVA), using SPSS for Windows 8.0 (SPSS, Chicago) and Minitab (Minitab, version 13.30). Subsequent t testing was performed where a significant interaction was found to indicate its source. Paired t tests were performed for euglycemia/hypoglycemia comparisons and unpaired t tests for aware/unaware comparisons.

During PET scanning, arterial plasma glucose was maintained at 5 (euglycemia) or 2.6 mmol/l (hypoglycemia), with a coefficient of variation of 0.03 (Fig. 1A). Steady-state plasma free insulin levels were 82.2 ± 15.2 and 77.3 ± 13.8 μU/ml during euglycemia and hypoglycemia, respectively (NS). During the hypoglycemic studies, all subjects in the aware group reported having had “typical hypoglycemic” symptoms and were observed to sweat. Sweating was observed in only one of the unaware subjects at hypoglycemia, and none of this group could distinguish the hypoglycemic from the euglycemic study.

There was no significant difference between the two groups in epinephrine levels at euglycemia (P = 0.4). As shown in Fig. 1B, epinephrine levels in the hypoglycemic studies rose briskly in the hypoglycemia-aware group (from 0.44 ± 0.23 to a mean peak of 7.52 ± 2.9 nmol/l) but rose only minimally in the unaware subjects (from 0.37 ± 0.15 to a peak of only 0.77 ± 0.39 nmol/l at hypoglycemia; P < 0.0003 versus aware group).

Whole-brain FDG uptake data are shown in Table 1 and Fig. 2. The values at hypoglycemia were approximately twice those at euglycemia. In addition to this highly significant effect of condition (euglycemia versus hypoglycemia, P < 0.001), the ANOVA analysis revealed a significant effect of group (aware versus unaware, P = 0.027). This corresponded to ∼20% lower whole-brain values for the hypoglycemia-unaware subjects relative to the aware subjects at both glucose levels. No significant group versus condition interaction was detected for the whole-brain data (P = 0.227); that is, the effect of hypoglycemia on global FDG uptake was not different between the aware and unaware groups.

SPM analysis of the FDG uptake images identified a region of brain where FDG uptake, normalized to whole-brain uptake, exhibited a highly significant group by condition interaction; that is, the effect of hypoglycemia on FDG uptake in this region was significantly different between the aware and unaware groups. The initial group by condition interaction analysis used a height threshold of P < 0.001 and detected a single region of differences with a cluster size of 178, Z = 3.40, and Talairach coordinates of maximal effect (x, y, and z = 22, −22, and −12, respectively). The anatomy of the identified brain region is more clearly seen when the analysis was performed at a slightly lower height threshold of P < 0.0025. Fig. 3 shows the Z score map resulting from this analysis, coregistered with a standard MRI in Talairach space. This clearly shows the cluster in the subthalamic region involving the ventral and medial aspects of the hypothalamus bilaterally below the third ventricle.

Regional normalized uptake values (NUVs) from the subthalamic ROI in the aware and unaware groups at euglycemia and hypoglycemia are shown in Table 2 and Fig. 4. The ANOVA revealed a significant overall effect of glucose level (P < 0.001) and an additional significant interaction between glucose level and awareness status (P = 0.046). Specifically, the relative regional NUV at euglycemia was higher in the unaware than in the aware group (0.715 ± 0.015 vs. 0.654 ± 0.016; P = 0.028) and showed a significantly greater fall with hypoglycemia (from 0.715 ± 0.015 to 0.623 ± 0.012 [P = 0.001] vs. from 0.654 ± 0.016 to 0.636 ± 0.013; [NS]) to a level that was not significantly different from that of the aware group at the same hypoglycemia (P = 0.631). When the average differences in global brain FDG uptake were taken into account, the absolute rates of FDG uptake in the subthalamic region in the unaware group were 14.9% lower at euglycemia and 20.4% lower at hypoglycemia than in the aware group.

Hypoglycemia unawareness is a major risk factor for severe hypoglycemia in the treatment of type 1 diabetes (1). It is associated with a failure of the normal counter-regulatory mechanisms that alert the diabetic patient to a falling blood glucose concentration (4). Most notably, the onset of the catecholamine and sympathetic responses to hypoglycemia is significantly delayed, with reduced magnitude of these responses at any given glucose level (5). The warning symptoms of hypoglycemia are similarly delayed and diminished, allowing the patient to slip into severe neuroglycopenia without subjective warning.

We used the FDG-PET method to investigate brain glucose metabolism in hypoglycemia-aware and -unaware diabetic patients. There are methodological issues that must be born in mind when interpreting the results of our study. Deoxyglucose is taken up by brain tissue as glucose and is similarly phosphorylated, but it does not move further along the glucose metabolic pathway. Its accumulation in brain tissue, imaged by PET, has been extensively used to calculate the glucose metabolic rate, using a formula that requires knowledge of a lumped constant to relate the behavior of FDG to that of native glucose (25). Although in one human study of acute hypoglycemia it was suggested that this constant was robust in moderate hypoglycemia (26), the literature from animal studies show a clear change in the constant when plasma glucose is lowered (10). For these reasons, we confined our subsequent analysis to the FDG uptake data, changes in which may reflect changes in the rate of glucose utilization, changes in the value of the lumped constant, or both.

The FDG method requires that the brain glucose levels remain constant throughout the duration of the PET study. Although the induction of hypoglycemia perturbs the steady state, the duration of induction of hypoglycemia and the delay between reaching the target glucose and collecting PET data make this an unlikely problem. Observed effects are also unlikely to relate to regional alterations in blood flow. Despite evidence for a small increase in brain blood flow with greater degrees of hypoglycemia (27) which may be different in the hypoglycemia-unaware group (28), a variation in regional blood flow is an unlikely explanation for our findings, as FDG uptake is known to be insensitive to moderate changes in blood flow (29,30)

Two effects were seen in our whole-brain uptake data. First, whole-brain FDG uptake rose approximately twofold when plasma glucose was lowered by about half, for both groups. Although there were several complicating factors (i.e., changes in the shape of the tracer input function and in the value of the lumped constant), this inverse relationship was probably primarily attributable to straightforward competition between FDG and glucose and entirely predictable. The second result is that whole-brain FDG uptake (reflecting either an absolute change in overall glucose utilization rate or in the lumped constant) was significantly and consistently lower in the unaware group than in the aware group, both at euglycemia and hypoglycemia.

Recent studies in animal brain exposed to prolonged hypoglycemia (which would be expected to induce counter-regulatory failure comparable to that shown by our hypoglycemia-unaware patients) have shown increased expression of the glucose transporters GLUT1 and GLUT3 (31,32). Therefore, the reduction in FDG uptake in our unaware group might have reflected a change either in glucose phosphorylation or in the lumped constant, rather than in glucose uptake. It is also possible that the decrease in FDG uptake (seen both at euglycemia and hypoglycemia) in our unaware patients was reflecting reduced neuronal density, but this possibility needs further investigation.

Recent human studies of global brain glucose uptake using arteriovenous difference techniques have found a preservation of global brain glucose uptake in hypoglycemia in unaware subjects but not in aware subjects (33,34). As discussed above, absolute rates of glucose utilization cannot be reliably derived from our data; however, there is nothing in our FDG data to suggest that the whole-brain response to hypoglycemia differed between the two groups (i.e., there was no significant group versus condition interaction in our whole-brain data). Another recent study using ¹¹C-labeled glucose and PET also failed to find any evidence for a difference in global brain glucose uptake at hypoglycemia between hypoglycemia-aware and -unaware subjects (35). In that study, hypoglycemia unawareness was induced in healthy volunteers by a single prolonged exposure to mild hypoglycemia. This may have different effects from the intermittent recurrent exposure to more profound hypoglycemia experienced by our diabetic patients, so the data are not necessarily comparable; however, neither of the PET studies suggested a global increase in the capacity of the brain to take up glucose.

In contrast to [¹¹C]glucose PET, FDG-PET data can be analyzed regionally. SPM compares images on a voxel-by-voxel basis, and results in a single parametric image displaying regions where statistically significant differences are found. We used SPM to identify brain regions where there was a significant condition (euglycemia/hypoglycemia) versus group (aware/unaware) interaction in our data; that is, regions where the two groups responded differently to the hypoglycemic challenge. The SPM analysis showed that in our small group of type 1 diabetic patients, the failure of the symptomatic counter-regulatory responses to a fixed moderate hypoglycemic challenge in the unaware patients was associated with a regional difference in brain FDG uptake compared with that of aware patients. This was independent of the changes in whole-brain uptake and was localized to the subthalamic brain centers, including the regions of the ventromedial and lateral hypothalamus, and did not extend to the rest of the brain. A subsequent ROI analysis performed specifically on the region identified confirmed the SPM result, and furthermore showed that the interaction corresponded to a larger fall in normalized FDG uptake at hypoglycemia from a higher baseline level in the unaware group.

The brain region indicated by the SPM analysis as different in the hypoglycemia unaware included those areas of brain known to contain glucose-sensing neurons and already implicated in hypoglycemia detection and the triggering of the autonomic responses. In experimental animals, when glucose levels in these brain regions are maintained by microdialysis techniques, the responses to systemic hypoglycemia are markedly attenuated (36). Conversely, when localized glucodeprivation is created in these brain regions by infusion of deoxyglucose, an autonomic response is generated, similar to the response expected for systemic hypoglycemia, despite maintenance of peripheral glucose levels (37). There has been a case report of a sarcoid lesion in this region adversely affecting hypoglycemia sensing, but the present study is the first demonstration of a specific regional metabolic defect in diabetic men with counter-regulatory failure (38).

We must be cautious about the interpretation of our regional normalized FDG uptake values, as these were calculated relative to the whole-brain data, which were 20% lower in the unaware group. Thus, although the normalized FDG uptake rate in the region at hypoglycemia was not significantly different between the two groups, the absolute rate of FDG uptake in the subthalamic region of the unaware group was 20% lower than in the aware group. This may have been directly related to the lesser epinephrine response in this group. However, it is likely that control of counter-regulatory hormone release is mediated indirectly from changes in neuronal activation in the glucose sensors through neurotransmitter release. Furthermore, because of the limitations of performing SPM on absolute data (19), the most robust statistical finding of our study was the difference in the magnitude of the fall of subthalamic FDG uptake rates between our two groups. How might a greater fall in FDG uptake in the subthalamic brain region be associated with a lesser epinephrine response?

One possible interpretation is that there was higher glucose uptake in the subthalamic brain region in the hypoglycemia-unaware group at euglycemia, but that this increased capacity was not maintained during hypoglycemia. As with the global data, FDG uptake may be affected by changes in the rate of glucose uptake and utilization, the lumped constant, or both. Despite the uncertainties in the exact interpretation of our data, they clearly show a local change in the way the tissue metabolized glucose at euglycemia and how that was altered during hypoglycemia in association with counter-regulatory failure.

The FDG uptake in the subthalamic regions remained virtually unchanged with hypoglycemia in association with the release of epinephrine in the aware group. This much smaller fall of FDG uptake in the subthalamic regions of these patients may have reflected a compensatory metabolic response to hypoglycemia, switched on to maintain cellular metabolic rates, which itself triggers autonomic activation. The greater fall in the region in the unaware group may then have represented an inability to switch on the same mechanism. Alternatively, the maintenance of FDG uptake in the face of falling glucose in the hypoglycemia-aware group may have reflected activation of the neurons of this glucose-sensing region to trigger counter-regulation. The recruitment of these additional glucose-sensitive neurons, perhaps while other cells were failing to maintain their metabolism, may explain the lack of significant reduction in glucose metabolism in the region as a whole. The failure to activate the glucose-sensitive neurons in the hypoglycemia-unaware group may explain the localized fall in FDG uptake there. Because hypoglycemia unresponsiveness is reversible (15,39), we would expect this regional change in FDG handling at hypoglycemia in hypoglycemia-unaware patients likewise to be reversible, but we have not tested this hypothesis.

Our data support the hypothesis that the human hypoglycemia sensor is located in the ventromedial hypothalamus, as has been shown for rodents. The data indicate altered glucose uptake, metabolism, or both in this region, associated with a loss of subjective and neuroendocrine responses to hypoglycemia. The extension of the defect into areas related to hunger and appetite control (the ventrolateral hypothalamus) is also important, as hunger is an important warning symptom of hypoglycemia, which is lost in hypoglycemia unawareness (40). We may need to revisit current dogma about the control of plasma glucose in humans if we are to successfully devise therapies to protect diabetic patients against asymptomatic and severe hypoglycemia.

This study was funded by the Wellcome Trust. Andrew Pernet was supported by the Juvenile Diabetes Foundation International.

The authors thank the nuclear medicine physicians and staff of the Clinical PET Center of St. Thomas’ Hospital, London, and Prof. Michael Maisey, its director; Prof. Peter Sonksen for access to the Clinical Diabetes Center at St. Thomas’ Hospital; Bernadette Cronin for her support in running the PET scans; David Forster and Prof. Ian Macdonald of Nottingham University for the measurements of the catecholamines; and Gary Chusney, Jill Lomas, and Andrew Pernet for laboratory and nursing support. We would also like to thank the patients who so kindly gave us their time and help in conducting these studies.