A Functional Magnetic Resonance Imaging Study
The authors’ aim was to examine the regional anatomy of brain activation by cognitive tasks commonly used in hypoglycemia research and to assess the effect of acute hypoglycemia on these in healthy volunteers. Eight right-handed volunteers performed a set of cognitive tasks—finger tapping (FT), simple reaction time (SRT), and four-choice reaction time (4CRT)—twice during blood oxygen level–dependent (BOLD) functional magnetic resonance imaging of the brain on two occasions. In study 1 (n = 6), plasma glucose was maintained at euglycemia (5 mmol/l) throughout. In study 2 (n = 6), plasma glucose was reduced to 2.5 mmol/l for the second set. Performance of the tasks resulted in specific group brain activation maps. During hypoglycemia, FT slowed (P = 0.026), with decreased BOLD activation in right premotor cortex and supplementary motor area and left hippocampus and with increased BOLD activation in left cerebellum and right frontal pole. Although there was no significant change in SRT, BOLD activation was reduced in right cerebellum and visual cortex. The 4CRT deteriorated (P = 0.020), with reduction in BOLD activation in motor and visual systems but increased BOLD signal in a large area of the left parietal association cortex, a region involved in planning. Hypoglycemia impairs simple brain functions and is associated with task-specific localized reductions in brain activation. For a task with greater cognitive load, the increased BOLD signal in planning areas is compatible with recruitment of brain regions in an attempt to limit dysfunction. Further investigation of these mechanisms may help devise rational treatment strategies to limit cortical dysfunction during acute iatrogenic hypoglycemia.
The brain is normally dependent on glucose for oxidative metabolism and function. Acute iatrogenic hypoglycemia, occurring as a result of insulin excess during the treatment of type 1 diabetes, can cause clinically significant cognitive impairment. In health, such hypoglycemic cognitive dysfunction does not occur. Endogenous counterregulatory mechanisms are activated in response to a small reduction in circulating glucose, preventing more profound hypoglycemia. In diabetes, the inability to reduce circulating insulin levels (1), the failure of glucagon responses (2), sometimes accompanied by defective autonomic and adrenergic responses (3), can allow plasma glucose levels to fall low enough to result in detectable disturbance of cognitive function, ranging from a subtle slowing of reaction times (3) to coma (4).
A variety of measures of cognitive function have been used in the investigation of the responses of the brain to acute hypoglycemia, and these show different susceptibilities to hypoglycemia. In health, autonomic activation occurs at arterialized plasma glucose of 3.0–3.6 mmol/l (5,6,7). Choice reaction times (3,8) and performance of Stroop interference tests become slower at ∼3 mmol/l; short-term memory deteriorates at 2.5 mmol/l (9); and finger tapping, a simple motor task, slows at ∼2.3 mmol/l (10). The glucose level at which some brain functions alter, such as the activation of autonomic counterregulatory responses (3,8,11,12,13) (a hypothalamic action ) or deterioration of the performance of the Stroop and memory tests (10,11,13), may change according to prior experience of hyper- or hypoglycemia, with higher or lower plasma glucose associated with the onset of change, respectively. Such changes may leave those diabetic patients with prior experience of hypoglycemia at high risk for severe hypoglycemia with no warning hypoglycemia unawareness (15).
Functional magnetic resonance imaging (fMRI) detects regional changes in brain oxygenation state during activation by a task (16,17,18,19,20). This allows exploration of physiological mechanisms underlying cognitive dysfunction during hypoglycemia. This study uses fMRI to test the hypothesis that different functional brain areas are involved in the performance of the different cognitive functions tests used in hypoglycemia research, and these will respond in distinct ways during a single hypoglycemic challenge, depending on the cognitive load of the task.
RESEARCH DESIGN AND METHODS
Eight right-handed healthy subjects (three women, age 28.6 ± 1.9 years, BMI 22.1 ± 0.95 kg/m2, and five men, 33.3 ± 1.1 years of age, BMI 23.3 ± 1.3 kg/m2) were recruited. They were screened for cardiovascular risk factors and conditions that would predispose to central nervous system abnormalities, and those with metal implants or other contraindications to magnetic resonance imaging were excluded. They were allocated to study 1 (prolonged euglycemia) and study 2 (euglycemia followed by hypoglycemia). This study design was to provide an internal control, eliminating any potential differences in performance or brain activation occurring in any one individual on different scanning days. Four subjects underwent study 1 and study 2 in random order, 3 weeks apart. Four further subjects underwent either study 1 or study 2. All studies were performed in the morning after an overnight fast and after 3 days of caffeine abstention. The subjects were blind to their glucose levels at all times. The protocol was approved by the Ethics Committees of King’s College Hospital and of the Maudsley Hospital, both of London, U.K. Each volunteer gave written informed consent for the studies, and consent for fMRI was obtained independently by the Neuroimaging Department.
A set of cognitive function tests, comprising finger tapping (FT), simple reaction time (SRT), and four-choice reaction time (4CRT), was applied twice in each study. During each set of tests, performance was measured and brain activation was determined using fMRI. The subjects practiced the three tests to stable performance 1 day before the studies and three times on the morning of each study before entering the magnetic resonance (MR) scanner. During scanning, plasma glucose was controlled by hyperinsulinemic glucose clamping either at euglycemia (5 mmol/l) throughout (study 1) or at euglycemia followed by controlled hypoglycemia (2.5 mmol/l, study 2).
On the day of the study, at 0700 h, subjects were admitted to the Neuroimaging Center and made comfortable, supine on the scanner table. A head mold was used to support the head, although this was only placed firmly in position when the subject was moved into the scanner.
Two intravenous catheters were inserted, using 1% lidocaine intradermally to anesthetize the skin. One catheter, for administration of insulin and glucose, was placed in the left antecubital fossa. The other, for blood sampling, was placed in a retrograde direction distally in the left forearm and kept patent with a slow infusion of normal saline. The left hand was placed in a warmed box (55°C) to arterialize the venous blood samples (21). The heating box was exchanged for heated water pads when the subject was moved into the scanner. Throughout, fluids were infused through the catheters via long infusion tubing.
A primed continuous intravenous infusion of regular insulin (Human Actrapid; Novo Nordisk), which made up to 55 ml in 0.9% sodium chloride in which 2 ml of the subject’s blood had been added, was infused at a maintenance rate of 1.5 mU · kg body wt−1 · min−1. Plasma glucose (measured at the bedside every 5–15 min) was kept constant at 5 mmol/l for 1 h before scanning using a variable-rate infusion of 20% glucose (Baxter Healthcare Ltd., Thetford, Norfolk, U.K.) (5). Before and during this time, the subjects performed the set of cognitive tests (see below) on at least two occasions, separated by rest. The glucose infusion rate was then doubled for 30 s, and the subject was disconnected from the infusions to be moved into the MR scanner, taking <1 min. The infusion pumps remained in the adjacent room. The infusions were reconnected to the subject through the dividing wall, so that magnetic materials did not enter the room.
Once the subject was comfortable in the scanner, still with plasma glucose maintained at 5 mmol/l, a localizer image was acquired. This was followed by the performance of the first set of cognitive tasks during fMRI scanning. The subject viewed the tests on a computer screen positioned at the end of the scanner table via a mirror positioned above the head and responded using either a keypad button or a joystick, as appropriate for each task, positioned at the right hand. Response data were recorded electronically. In study 1, a second set of cognitive tasks was performed after a rest period of 15 min, with euglycemia (5 mmol/l) maintained throughout. In study 2, the first set of tests was performed at euglycemia (5 mmol/l), but plasma glucose was reduced to hypoglycemia (2.5 mmol/l) after completion of the first set of tests. The hypoglycemic stage was maintained for ∼40 min while the second set of tests was performed.
On completion of each study, the plasma glucose was restored to euglycemia if necessary, the insulin infusion was stopped, and the subject was removed from the scanner and given a meal. Blood glucose monitoring continued until euglycemia was maintained spontaneously, and then all lines were withdrawn and the subject went home.
Cognitive function tests.
These were based on conventional tasks but adjusted for fMRI use. This requires the active performance of the task to occur in 30-s phases, so the changes in the blood oxygen level–dependent (BOLD) signal that occur with similar periodicity can be correctly attributed to brain activation by the task. The change in performance from the first to the second set of tasks in each study was measured and compared between prolonged euglycemia and sequential euglycemia-hypoglycemia to assess the effect of hypoglycemia.
The subject repeatedly pressed a keypad button, positioned comfortably under the right hand, as quickly as possible for 30 s in response to the computer screen turning from red to green and then rested for 30 s. The test lasted for 5 min. The number of responses per unit of time was recorded.
The subject was presented with four oval symbols arranged in the form of a cross on the screen. When a single oval was lit, the subject moved a joystick in the direction of the indicated symbol and then returned the joystick to the resting position. The same oval was illuminated intermittently over 30 s, followed by a rest period of 30 s, and the sequence then repeated with different symbols over a total of 5 min. Response times were recorded for each joystick movement.
This test was similar to the SRT test, but the visual stimulus was unpredictable, occurring randomly in one of four possible positions. The subject responded by pushing the joystick toward the corresponding position. The time and accuracy of each response were recorded.
The three tasks were performed as a set, with a 5-min break between each task. After the first set was completed, the subject rested for a period of 15 min, after which a second set was performed.
Image acquisition and analysis.
Gradient-echoplanar images were acquired using a 1.5 Tesla GE Signa System software (General Electric, Milwaukee, WI) fitted with advanced nuclear magnetic resonance (ANMR, Woburn, MA) at the Maudsley Hospital. A hundred T2* weighted MR images displaying BOLD contrast (20) were acquired from 14 planes parallel to the plane through the anterior commisure–posterior commisure (AC-PC) line. Scan sequence parameters were as follows: echo time (TE) = 40 ms, repetition time (TR) = 300 ms, flip angle = 90, in-plane resolution = 3.1 mm, slice thickness = 7 mm, slice skip = 0.7 mm. A 43-slice high-resolution inversion-recovery echoplanar image of the whole brain was also acquired in the AC-PC plane (TE = 73 ms, inversion time (TI) = 180 ms, TR = 16,000 ms, in-plane resolution = 1.5 mm, slice thickness = 3 mm).
The MR images were analyzed to determine brain regions or voxels in which the signal appeared to change significantly between active and inactive phases of performance of each cognitive task. After software correction for subject movement (22), the power of the relative changes in response to the phases of performance was estimated by regression of a periodic model (23). This model is described in reference 23.
For each image, maps were registered in the standard space of Talairach and Tournoux (24) and were smoothed by a Gaussian filter (SD = 7 mm) to accommodate variability in the gyral anatomy. Testing to deduce which responses exceeded a statistical threshold was performed to reveal activated regions. This threshold was calculated using a method of nonparametric randomization (22). The median for the cohort was then calculated in the same way, with each individual contributing to the median group response displayed in the group activation maps.
Plasma glucose was measured at the bedside in duplicate using a glucose oxidase technique (Yellow Springs Instruments, Yellow Springs, OH). Catecholamines were measured using high-pressure liquid chromatography with electrochemical detection (25).
Results are expressed as means ± SD, unless stated otherwise. To determine the regional anatomy of brain activation by the three cognitive tasks at euglycemia, the activation maps for the first image data for each of the eight subjects were analyzed using published methodology (23), with P < 0.005 taken to indicate significant activation.
To examine the effects of hypoglycemia on peak catecholamine responses and task performance, the changes between the first and second cognitive function test for each study were calculated for each subject by subtraction, and the changes for all subjects in study 1 (prolonged euglycemia) were compared with the changes in study 2 (sequential euglycemia-hypoglycemia) by using Student’s t test. A P value of <0.05 was taken to indicate a significant difference.
The differences in group brain activation between the first and second sets of tasks were compared between studies 1 and 2 by analysis of covariance, with a P value of <0.005 taken to indicate significance.
During study 1 (prolonged euglycemia), the mean glucose level through the first set of cognitive tests was 5.4 ± 0.3 mmol/l, and during the second set it was 5.3 ± 0.4 mmol/l. During study 2 (sequential euglycemia-hypoglycemia), the mean glucose level during the first set of tests was 5.6 ± 0.6 mmol/l, and during the second set it was 2.5 ± 0.1 mmol/l. There were no significant differences between the glucose profiles of study 1 and the euglycemic level of study 2 (Fig. 1).
Plasma epinephrine concentrations remained stable throughout study 1, with a mean increment of 0.036 ± 0.422 nmol/l over a mean baseline of 1.9 ± 0.09 mmol/l. During the hypoglycemia of study 2, epinephrine levels rose, with a mean increment of 3.7 ± 1.3 nmol/l over a mean baseline of 0.25 ± 0.07 mmol/l, P = 0.0002 vs. study 1. This result was accompanied by a nonsignificant rise in norepinephrine levels.
Performance of cognitive function tasks.
In study 1 (prolonged euglycemia), there was a percentage increase in the number of FT responses from the first to the second set (4.7 ± 9.7%). In contrast, in study 2, the number of responses decreased from the first to the second set (−17 ± 8%, P = 0.026 vs. study 1) (Fig. 2).
Simple reaction time was not significantly affected by hypoglycemia, changing by −12.0 ± 30 ms during prolonged euglycemia compared with 3.7 ± 56 ms in sequential euglycemia-hypoglycemia, P = 0.57. The 4CRT became faster in the prolonged euglycemia of study 1 (29 ± 44 ms) but deteriorated significantly from euglycemia to hypoglycemia in study 2 (−29 ± 25 ms, P = 0.02 vs. study 1).
Regional activation by the tasks at euglycemia.
Similar cognitive networks were activated by the three tasks, but some differences in the specific patterns of activation were observed between tasks, as shown in Fig. 3. Table 1 lists the center of mass Talairach coordinates (which define the regions in three-dimensional space) of activated regions exceeding 10 voxels in size for each task. The corresponding group images are shown in Fig. 3. This shows the mean data for activation for all eight subjects. They are all shown at a voxelwise statistical significance threshold of P < 0.005, giving an estimated number of type I error voxels of 10 per image volume. The numbers of observed activated voxels were 1,372 for FT, 1,223 for SRT, and 1,143 for the 4CRT.
All three tasks resulted in activation of the left precentral gyrus, the right premotor cortex, the supplementary motor area, the left and right supramarginal gyri, the right and left visual cortices, and the right and left cerebellum. The posterior cingulate gyrus was activated in the 4CRT and in FT but not in SRT. The middle temporal gyrus was activated in both reaction time tasks, together with the left postcentral gyrus and the middle temporal gyrus (on the right in SRT and left in 4CRT). Areas activated in one task only were the left inferior-posterior temporal lobe and the left and right parietal association area in FT; the dorsal-lateral prefrontal cortex and the medial frontal area in SRT; and the right frontal pole in 4CRT.
Effect of hypoglycemia on regional brain activation by tasks.
Most areas of brain activated by each task in the euglycemic studies were similarly activated in the hypoglycemic studies (Fig. 4). However, for each task there were significant differences in brain activation that were associated with hypoglycemia. Table 2 details the areas that demonstrated reduced activation during hypoglycemia, and Table 3 details the areas that demonstrated increased activation during hypoglycemia in the three tasks. The coordinates given are the center of mass for each contiguous region in the coordinates system of Talairach and Tournoux (24).
All areas listed exceed a statistical threshold of P = 0.005, with an estimated number of type I error voxels of five per image. Areas not listed in the tables were unchanged.
As shown by Tables 2 and 3, the changes in activation by each task during hypoglycemia were different. In general, regional activation in the cerebral cortex during task performance was unchanged or reduced during hypoglycemia. Three areas showed increased activation during task performance at hypoglycemia; the largest, seen in two contiguous areas, was in the parietal association cortex and was seen only during the 4CRT.
Figure 4 shows an example of the image data from the 4CRT.
Impairment of brain function is a recognized consequence of acute hypoglycemia. It is known that different brain functions have different susceptibilities to acute hypoglycemia; for example, the triggering of autonomic activation, a hypothalamic function, occurs in response to quite modest decrements in circulating glucose concentrations (5,6,7), whereas significant cortical dysfunction requires more profound glucose deprivation (3). Even within different cortical functions, there is variation in hypoglycemia susceptibility (26). The mechanisms of cortical dysfunction during hypoglycemia are not known. Using fMRI, we have been able to identify the regions of brain activated in certain cognitive tasks commonly used in hypoglycemia research and to determine the effect of hypoglycemia upon that regional activation. Our data show that the effect of acute hypoglycemia on the human brain is task- and region-specific. This finding may help explain the differences in sensitivity to hypoglycemia observed in different cortical functions.
fMRI depends on detecting regional changes in the oxygenation status of the brain (in this case, between rest and activation), due to the differences in paramagnetism between deoxy- and oxyhemoglobin (16,17,18,19,20). When an area of brain is activated by a task, there is a subtle brief fall in BOLD signal followed by an increase, attributed to reactive hyperemia (20). This sequence suggests that a task-specific brain region becomes activated during task performance, creating a localized increased oxygen demand and an equally localized reactive hyperemia. The hyperemia results in regional excess oxygen delivery and a strong signal, which delineates the brain structures involved in the task. fMRI therefore gives a map of the brain regions involved in task performance and indicates part of the mechanisms of that brain activation. It has demonstrated reproducibility and reliability for visual, motor, and cognitive tasks (27,28), and we designed our study to eliminate any intraindividual variation from day to day (29).
The BOLD response was studied in three tasks of increasing complexity: FT, SRT, and 4CRT. All three tasks resulted in activation of the visual cortices, although SRT and 4CRT resulted in more extensive activation and involved a different visual network from FT, presumably reflecting the less complex visual cue in the latter. Our group activation maps are consistent with data from other studies showing that the ventral occipito-temporal visual pathway is involved in both the identification of color (as in FT) and the dorsal occipito-parietal visual pathway in spatial perception (as in SRT and 4CRT) (30,31). The SRT task was associated with activation of the left medial frontal lobe and the right dorsolateral prefrontal cortex, implicated in the processing of visual information in working memory (32). This is likely to be because only in this task is there a need to retain an image between stimuli. This task was not associated with activation of the posterior cingulate gyrus. This may be because it requires less activation of attention compared with the other tasks (33). Although all three tasks activated the supramarginal gyrus, which is strongly linked with visual associative function (34), the most complex task, 4CRT, resulted in the largest activation of this area.
All three tests activated the left precentral gyrus, the right premotor cortex and supplementary motor area, and the right and left cerebellum, with FT also activating the left and right precuneus, and 4CRT and SRT activating the somotosensory cortex (postcentral gyrus). All of these areas are involved in the integration of sensory information and execution of motor function (35), and their pathways are described elsewhere (23,36). The premotor cortex in particular has been shown to be involved in the integration of tactile and visiospatial signals to cued movements (37). Our data are consistent with a number of positron emission tomography studies examining the anatomy of brain activation when motor routines are executed in response to visual or somatosensory cues (38).
The activation of the postcentral gyrus in SRT and 4CRT probably reflects the necessary information regarding the position of the joystick at any one time during these tests. The precuneus is involved in FT and is thought to be important in tasks that require an internal cue (39).
Performance in FT and 4CRT both deteriorated in the hypoglycemia of our study, as would be expected from previous studies (3,9,40,41,42). The failure of the performance of SRT to deteriorate significantly, contrasted with the marked change in the FCRT, suggests that the cognitive element of the latter was more affected by the hypoglycemia. However, a relatively small number of challenges are presented during an fMRI-compatible test, so a type 2 error in the SRT data is possible.
With each task, hypoglycemia and/or deteriorated performance was associated with either no change or a reduction in BOLD signal in specific brain areas. A reduction in occipital cortical activation during hypoglycemia in humans has recently been reported in abstract for a simple visual stimulus (43). Another abstract has reported on the correlation of task difficulty rather than hypoglycemia state with focal changes in BOLD signal (44), but a recently published study in rats, showing reduced specific regional brain activation in hypoglycemia in response to median nerve stimulation, suggests that the reduced BOLD signal in brain activation in hypoglycemia is a direct effect of the glucose lack (45). These data are compatible with a failure of the ability to enhance regional brain metabolism to perform the task as a result of glucose deprivation as well as a subsequent failure to stimulate local hyperemia. An alternative explanation might be that the glucose deprivation prevents the hyperemic response alone. In either case, the normal enhancement of regional brain blood flow fails, associated with deterioration in the ability to perform such tasks.
Of the three areas where BOLD signal increased during hypoglycemia, the two small areas seen with FT are widely separated anatomically and functionally and are each seen in one slice only. The enhancement of BOLD signal in areas of cortex associated with planning during 4CRT task in hypoglycemia extended over 16 voxels and involved two contiguous brain slices. This significantly increased BOLD activation suggests recruitment of new areas of cortex to perform this more complex task during hypoglycemia. The increased BOLD signal in these cognitive areas is compatible with increased effort by the subject, perhaps with awareness that the task is increasingly difficult. It is noteworthy that this did not successfully prevent deteriorated performance, but the response may be limiting the degree of deterioration. Certainly, the difference in the response between the SRT and FT as compared with the 4CRT appears to relate to the increased cognitive load.
BOLD signal is a composite of neuronal oxygen consumption and localized hyperemia, and changes in either may contribute to our results. Global cerebral blood flow rises during hypoglycemia in the resting brain, although this is considered to occur at lower glucose levels than used in this study (46,47). Studies of activated brain at hypoglycemia found no associated increment in regional blood flow (48).
Other changes in hypoglycemia may include the degree of neurovascular coupling, and regional differences in this response may also contribute to the differences we saw. Certainly, our technique of insulin clamping does not alter cerebral blood flow or brain glucose metabolism (49), and the composite analysis of all four studies—recurrent euglycemia and sequential euglycemia-hypoglycemia—eliminates the possibility that the observed changes relate to the administration of the clamp. It is also unlikely that the changes in BOLD signal that occurred with hypoglycemia were due to the catecholamine responses, as there is no evidence that catecholamines cross the blood-brain barrier (50).
Our study included both men and women. Sex differences have been demonstrated in the functional organization of the brain for language (51), but sex-related differences in regional brain activation for the motor tasks used in our study have not been identified. We examined the data for differences at euglycemia between male and female subjects for the chosen tasks and found none (data not shown).
In conclusion, fMRI has allowed us to identify the brain regions involved in different brain tasks commonly used to examine cortical function in hypoglycemia research and begin to examine the mechanisms by which they respond to hypoglycemia. While brain activation by simple tasks is diminished during hypoglycemia (associated with a deteriorated performance), tasks that require significant cognitive input have a different response, suggesting a capacity for the brain to limit impaired performance during hypoglycemia. Our data may help explain why different tasks show different sensitivities to hypoglycemia, although these studies are elaborate and highly technical, and extrapolation to the general population from our limited sample size must be done with caution. Further investigation of these responses may help unravel the mechanisms of cognitive impairment in acute hypoglycemia and lead toward rational strategies to protect against severe hypoglycemia in the treatment of people with diabetes.
The Juvenile Diabetes Foundation International funded this study. J.M.R. is a Diabetes U.K. R.D. Lawrence Research Fellow.
We are grateful to the radiographers in the Magnetic Resonance Imaging Unit of the Maudsley Hospital, without whom these studies could not have been done, and to Prof. Iain Macdonald and David Forster of Queen’s University, Nottingham, U.K., for the measurements of catecholamines.
Address correspondence and reprint requests to Dr. Jane Miranda Rosenthal, Department of Diabetes, Endocrinology, and Internal Medicine, GKTSM, Denmark Hill Campus, Bessemer Road, London SE5 9PJ, U.K. E-mail:.
Received for publication 8 May 2000 and accepted in revised form 19 March 2001.
AC-PC, anterior commisure–posterior commisure; 4CRT, four-choice reaction time; BOLD, blood oxygen level–dependent; fMRI, functional magnetic resonance imaging; FT, finger tapping; MR, magnetic resonance; SRT, simple reaction time; TE, echo time; TI, inversion time; TR, repetition time.