Functional Brain Connectivity and Neurocognitive Functioning in Patients With Long-Standing Type 1 Diabetes With and Without Microvascular Complications: A Magnetoencephalography Study

Fifteen type 1 diabetic patients with proliferative retinopathy, an indicator of microangiopathy resulting from chronic hyperglycemia (type 1 diabetes⁺), 29 type 1 diabetic patients without manifest microvascular complications (type 1 diabetes⁻), and 26 healthy control subjects matched for sex and education level enrolled in this study. Participants were recruited from the departments of Endocrinology and Ophthalmology of the VU University Medical Center, Amsterdam, the Netherlands (50 patients), the department of Internal Medicine, Groene Hart Hospital, Gouda, the Netherlands (10 patients), and by advertisements in diabetes magazines and a national newspaper (10 patients).

Inclusion criteria were age range 18–55 years, right-handedness, for type 1 diabetic patients a disease duration of at least 10 years, proliferative retinopathy as described below, or no signs of microvascular complications. Mastery of the Dutch language was required for all participants. Participants were excluded if they had a BMI above 35 kg/m², current use of drugs affecting cerebral functioning, alcohol abuse (more than 20 g of alcohol per day), psychiatric disorders, anemia, thyroid dysfunction, use of glucocorticoids, hepatitis, stroke, severe head trauma, epilepsy, pregnancy, or poor visual acuity below 0.3. Fundus photography (Topcon NW 100, Capelle aan den IJssel, the Netherlands) was performed to screen for retinopathy. For each eye, one photograph with the macula in the center and one with the optic disc in the center were taken (E.v.D.). Photographs were rated by an ophthalmologist (A.M.) according to the European Diabetes (EURODIAB) classification (20). Only those participants with either an EURODIAB classification score of 0 (no retinopathy) or of 4 or 5 (proliferative retinopathy or lasercoagulation) were included in this study. Twenty-four–hour urine collections were performed to assess albumin-to-creatinine ratio (ACR). The presence of peripheral neuropathy was ascertained by the physician. Type 1 diabetes⁻ patients had normoalbuminuria (ACR <2.5 mg/mmol in men and <3.5 mg/mmol in women) and no neuropathy. Type 1 diabetes⁺ patients were allowed to have microalbuminuria and/or peripheral neuropathy. Hypertension was defined as a systolic blood pressure of 140 mmHg or above, a diastolic blood pressure of 90 mmHg or above, or the use of antihypertensive drugs. The study protocol was approved by the Medical Ethics Committee of the VU University Medical Center, and written informed consent was obtained from all participants.

Before eligible participants received written information on the study, their medical records were screened using the aforementioned criteria. Those willing to participate were additionally interviewed over the telephone to collect background information including educational level, using a Dutch scoring system ranging from 1 to 8. One indicates unfinished primary school and 8 indicates a completed university study at masters level. As depression may affect cognitive performance, depressive symptoms were assessed using the Center for Epidemiological Studies scale for Depression (CES-D). Eligible participants were invited for the first visit. During this visit, which took place in the afternoon, fundus photography was performed, blood samples were collected, and MEG registration was performed. Approximately 6 weeks after the first visit participants returned for neuropsychological assessment during a morning visit. All visits took place at the VU University Medical Center.

To rule out confounding as a result of extreme blood glucose levels, levels were checked in the type 1 diabetes participants before the start of the neuropsychological assessment and MEG registration and had to be between 4 and 15 mmol/l (72–270 mg/dl). In case of hypoglycemia, participants were instructed to eat 20 g of carbohydrates. In case of hyperglycemia, patients were instructed to inject 2 units of the participants' current rapid-acting insulin analog when blood glucose levels were between 15 and 20 mmol/l (270–362 mg/dl) and 4 units when glucose levels exceeded 20 mmol/l (362 mg/dl). Glycemic status was evaluated after 30 min. If hyperglycemia (i.e., blood glucose ≥15 mmol/l) would persist, participants were instructed to inject another 2 units of insulin; if hypoglycemia would persist, participants additionally had to eat 20 g of carbohydrates. Two type 1 diabetes⁺ participants were hyperglycemic before start of the neuropsychological assessment and one before MEG acquisition. One type 1 diabetes⁻ patient indicated symptoms of hypoglycemia during neuropsychological testing. Proper glycemic status was restored for all patients after the first step of the above-mentioned protocol. Data of the one participant who experienced hypoglycemia symptoms during neuropsychological testing were included in all analyses, as this did not change analysis outcomes.

Based on earlier type 1 diabetes cognitive research (5,21), as well as clinical neuropsychological practice (22), a battery of cognitive tests was chosen to measure potential differences in five major cognitive domains: memory, information processing speed, executive functions, attention, and motor speed. Domains were based on an earlier principal component analysis using varimax rotation with Kaiser normalization in a large group of healthy subjects and adjusted according to earlier research (5,21,23).

The domain ‘memory’ was assessed by the Dutch version of the Rey Auditory Verbal Learning Test (RALVT) (24), Wechsler Adult Intelligence Scale–3rd edition revised (WAIS-III-R) Digit Span forward and backward (25), and the WAIS-III-R Symbol Substitution Incidental Learning Test (25). The domain ‘information processing speed’ was created using the WAIS-III-R Symbol Substitution Test (25), the Stroop Color-Word Test parts 1 and 2 (26), the Concept Shifting Task (CST) parts A and B (27), the Simple Auditory and Visual Reaction Time Tests (28), and the Computerized Visual Searching Task (CVST) (28). Assessment of the domain ‘executive functions’ was conducted using the Stroop Color-Word Test part 3 (26), the CST part C (27), the D2-test total errors (29), the Wisconsin Cart Sorting Test (30), and the Category Word Fluency Task (31). The domain ‘attention’ was assessed using the D2-test range, with total correct answers and total span (29). The domain ‘motor speed’ consisted of the Tapping Test (28) and the CST part 0, administered three times (27). ‘General cognitive ability’ was constructed by averaging the above mentioned five domains.

To enhance comparability and allow construction of these domains, z-scores for every test were created based on the means ± SD of the healthy control group. Higher z-scores indicate better performance.

MEG data were obtained using a 151-channel whole-head MEG system (CTF systems; Port Coquitlam, BC, Canada), while participants were in a supine position in a magnetically shielded room (Vacuumschmelze, Hanau, Germany). A third-order software gradient (32) was used with a recording passband of 0.25–125 Hz and a sample frequency of 625 Hz. Participants had to be free of any metal materials. Magnetic fields were recorded for 2 min in an eyes-open, 5 min in an eyes-closed, 10 min in a task, and then 3 min in another eyes-closed condition. Total acquisition time was 20 min. At the beginning, middle, and end of each recording, the head position relative to the coordinate system of the helmet was determined by leading small alternating currents through three head position coils attached to the left and right preauricular points and the nasion on the subjects head. Changes in head position of ∼1.5 cm during a recording condition were accepted. If head position changed more than 1.5 cm, recordings had to be repeated. This did not happen in this protocol.

From the acquired MEG data, information about functional connectivity was calculated by means of a mathematical construct, the synchronization likelihood. For a technical description of synchronization likelihood see Stam and van Dijk (33) and Montez et al. (34). In short, interactions between two neural networks, for instance the frontal (X) and temporal (Y) networks, are of interest. MEG data are recorded for all sensors surrounding the head, including those in the frontal and temporal areas. These signals represent the time series x_i and y_i from the frontal (X) and temporal (Y) areas (Fig. 1). It is assumed that X and Y more strongly interact when x_i and y_i more ‘resemble’ each other (square with continuous line in Fig. 1). It has been shown, however, that X and Y can also interact when x_i and y_i do not resemble each other (square with the dashed line in Fig. 1). This is called generalized synchronization and can be quantified by computing synchronization likelihood. Parameter settings used for computation of synchronization likelihood are based on the frequency content of the data (for parameter settings see Montez et al. [34]). Synchronization likelihood can range from P_ref (low synchronization) to one (complete synchronization). P_ref is a value close to zero (zero indicates no synchronization, which is not possible) and was set on 0.01 for all frequency bands.

For this analysis, 141 of the 151 channels were used. Ten channels were deleted as their signals were distorted in some participants. This had no influence on the calculations. Offline, recordings of the first eyes-closed condition were transformed into ASCII-files and imported into the DIGEEGXP software (C.J. Stam, VU University Medical Center, Amsterdam, the Netherlands). For each participant five artifact-free epochs (6.25 s) of 4,096 samples were selected by two authors (E.v.D. and N.S.). Total data used were 31.25 s. With DIGEEGXP synchronization likelihood was calculated (33). Synchronization likelihood was calculated between signals recorded at all possible sensor pairs for Δ (0.5–4 Hz), θ (4–8 Hz), lower α (8–10 Hz), upper α (10–13 Hz), β (13–30 Hz), lower γ (30–45 Hz), and upper γ (55–80 Hz) frequency bands. Subsequently, MEG sensors were clustered according to their anatomical location, respectively, right and left frontal, central, parietal, temporal, and occipital. Average synchronization likelihood values were then obtained for short-distance, intrahemispheric long-distance, and interhemispheric long-distance groups. Short-distance group consisted of 10 areas, left and right hemisphere central, frontal, parietal, occipital, and temporal regions. For each of these 10 areas, the synchronization likelihood between all possible pairs of sensors belonging to that area was averaged. This resulted in ten, local short-distance synchronization likelihood values. Average synchronization likelihood for long distances was obtained by averaging all pair wise synchronization likelihood values of sensors belonging to two different areas. Intrahemispheric long-distance group consisted of eight pairs, left and right hemisphere frontoparietal, frontotemporal, parietoocciptal, and temporooccipital. Interhemispheric long-distance group consisted of five pairs, left to right hemisphere central, parietal, temporal, occipital, and frontal (12,13). See Fig. 2 for anatomical locations. Now, synchronization likelihood resembles the likelihood that the data obtained from all sensors in the above-defined short-distance and long-distance groups are synchronized.

Differences between groups for demographical and medical variables were calculated using Student's t test for independent samples or one-way ANOVA with Tukey post hoc test (for continuous variables) and χ² test (for categorical variables).

To determine group differences in neuropsychological performance MANCOVA was used with all cognitive domains as dependent variables and group as independent variable. In case of a group difference on a domain, post hoc MANCOVA was used to determine which groups differed.

Synchronization likelihood data were normalized by means of a transformation LN10(x/[1-x]) (35) to allow the use of parametric statistical tests. To minimize statistical tests, for each MEG frequency band ANCOVA with repeated measures was used to determine group differences. For each frequency band, three ANCOVAs with repeated measures were performed for intra-, inter-, and local hemispheric connections. Repeated measures for intrahemispheric connections consisted of the 8 above-mentioned levels, interhemispheric connections of 5 levels, and local hemispheric connections of 10 levels. In case of significant interaction (P values of Greenhouse-Geisser correction for degrees of freedom) or main effects of group with frequency band, post hoc MANCOVA analysis was used to determine effects of group and spatial location. To determine associations between cognitive functioning and synchronization likelihood, a procedure proposed by Stoffers et al. and used in other MEG research was implemented (11,12). To reduce multiple comparisons, cognitive domains were added as covariates to the above-mentioned ANCOVA with repeated measures method. Effects of group on changing cognitive domains and MEG regions were determined using MANCOVA tests. Positive or negative associations were further calculated using regression analysis. To correct for possible confounding influences, all statistical tests were corrected for age, sex, hypertension, neuropathy, BMI, education level, and depressive symptoms.

Partial η² is reported as a proportion of the total variance explained of the determinant by the independent factor corrected for the used covariates. All statistical analyses were performed using SPSS for Windows version 15.0 (SPSS, Chicago, IL).

Table 1 summarizes group characteristics. The control group was significantly younger than the type 1 diabetes⁺ patients and had a significantly lower mean score of depressive symptoms. As expected, type 1 diabetes participants had a higher A1C than control subjects. Type 1 diabetes⁺ patients had a significantly lower age of disease onset and longer disease duration.

Compared with control subjects, both patient groups had a significantly lower information processing speed (type 1 diabetes⁺: P = 0.003, η² = 0.242; type 1 diabetes⁻: P = 0.009, η² = 0.135) and motor speed (type 1 diabetes⁺: P < 0.001, η² = 0.311; type 1 diabetes⁻: P = 0.014, η² = 0.122) as well as general cognitive ability (type 1 diabetes⁺: P = 0.007, η² = 0.198; type 1 diabetes⁻: P = 0.004, η² = 0.165). There were no statistically significant differences between the patient groups. The z scores are displayed in Fig. 3.

MEG results are shown in Table 2 (with P values and η²) and visually represented in Fig. 4. First, repeated-measures ANCOVA was performed for all frequency bands. Significant interaction effects with group were found for the θ band local (P = 0.001), lower α band local (P = 0.004), upper α band local (P = 0.008), β band interhemispheric (P = 0.043), and β band local (P = 0.020). Significant main effects of group were found for the lower α band intrahemispheric (P = 0.028), lower α band local hemispheric (P = 0.005), and upper α band local (P = 0.042). Post hoc MANCOVA analysis revealed significantly higher synchronization likelihood in the lower α right parietooccipital pathway for the type 1 diabetes⁻ compared with the healthy control subjects. Comparing type 1 diabetes⁺ patients with control subjects revealed significantly lower synchronization likelihood in the θ and lower α left and right central and parietal areas, in the upper α left frontal and right frontal, central, and parietal areas, and in the β interparietal, left parietal and right central, and parietal areas. Comparisons of type 1 diabetes⁺ with type 1 diabetes⁻ patients yielded lower synchronization likelihood scores for type 1 diabetes⁺ patients in the θ left and right central and parietal areas, in the lower α band left frontotemporal, left parietooccipital, right parietooccipital, and right temporooccipital pathways, in the left frontal and left and right central and parietal areas, upper α left parietal and right central and parietal areas and in the β-right parietal area. Anatomical locations are illustrated in Fig. 2.

To summarize, type 1 diabetes⁺ patients showed lower synchronization likelihood in both higher and lower frequency bands compared with both type 1 diabetes⁻ patients and control subjects. Type 1 diabetes⁻ patients showed increased synchronization likelihood in the lower α band compared with control subjects. No statistically significant differences were found when the combined patient groups were compared with healthy control subjects. On the contrary, a significant difference in the Δ intrahemispheric region was observed (supplementary Table 1A and B, available in an online appendix at http://diabetes.diabetesjournals.org/cgi/content/full/db09-0425/DC1).

To test for associations between functional connectivity in different MEG frequency bands and cognition, MEG statistics were repeated with cognitive domains added as additional covariates for both type 1 diabetes groups. In type 1 diabetes⁺ patients significant associations were found between the domain of executive functions and Δ intrahemispheric (P = 0.041), Δ interhemispheric (P = 0.008), and Δ local (P = 0.043) frequency band. Furthermore, Δ interhemispheric was related to the domains of memory (P = 0.030), information processing speed (P = 0.040), and motor speed (P = 0.014). The lower γ local frequency band was related to attention (P = 0.047). For executive functions only, significant relations with anatomical areas were found. In the type 1 diabetes⁻ patients significant relations were found between attention and Δ intrahemispheric (P = 0.048), upper γ intrahemispheric (P = 0.013), and interhemispheric (P = 0.043). Motor speed was associated with Δ local (P = 0.028) and θ local (P = 0.036). Lastly, β intrahemispheric was associated with information processing speed (P = 0.040). Subsequently, regression analysis was used to determine positive or negative relations between synchronization likelihood and cognition. In both groups, an increase in synchronization likelihood was associated with better performance on the related domains. Table 3 summarizes MANCOVA and regression tests.

This is, to the best of our knowledge, the first study using MEG to assess functional brain connectivity in type 1 diabetes. We showed that, compared with sex- and education-matched type 1 diabetes⁻ patients and control subjects, type 1 diabetes⁺ had decreased functional connectivity. This decrease was most profound in the left and right central and parietal areas in the θ, lower and upper α, and β frequency bands. Moreover, some intra- and interhemispheric differences were found in the lower α and β bands. In contrast, an increase in functional connectivity was found in the lower α band when comparing the type 1 diabetes⁻ patients with control subjects. Several frequency bands were positively related to cognitive domains, indicating these domains to be dependent of functional connectivity. On cognitive assessment, type 1 diabetic patients with and those without proliferative retinopathy performed significantly poorer on information processing speed, motor speed, and general cognitive ability. These deficits are in line with previous research (1,3,36). Results are in line with the hypothesis that type 1 diabetes⁺ patients show cognitive and cerebral changes. Changes found in type 1 diabetes⁻ patients were, however, not hypothesized. When both patient groups were pooled and compared with control subjects, cognitive results did not change; however, all differences in MEG measurements, but those in the Δ intrahemispheric band, lost statistical significance (supplementary Table 1A and B, available in an online appendix). This indicates that type 1 diabetes⁻ patients largely resemble the control subjects regarding functional connectivity.

Given the novelty of the technique used, caution is required when it comes to the interpretation of our connectivity findings. As there is currently only one study available that uses our method in metabolic disease, it may be difficult to put the current findings into a broader perspective. In female obese adolescents, significant increases in functional connectivity in Δ and β frequency bands were found (13), possibly because of an increase in white matter, which has been reported in MRI studies of obese adolescents. This is contrary to the decrease we found in our group. In contrast, in multiple sclerosis a decrease in functional connectivity was reported (37). This might be related to a loss of white matter, one of the core features of the disease. Following this line of reasoning, our results for the type 1 diabetes⁺ patients could be a consequence of white matter loss, as has been found earlier in type 1 diabetes⁺ patients than type 1 diabetes⁻ patients and healthy control subjects (5). Conversely, the increase in functional connectivity in type 1 diabetes⁻ patients might reflect a compensatory mechanism, which fails in type 1 diabetes⁺ patients. Because of large interindividual differences in cognitive performance, this compensation was not observed in cognitive functioning. Therefore, larger samples are needed to enable the detection of significant differences between type 1 diabetes⁺ and type 1 diabetes⁻ patients.

Importantly, we found a significant positive relationship between cognitive functioning and functional connectivity, suggesting changes in functional connectivity are involved in cognitive dysfunction in type 1 diabetic patients.

Our data support the hypothesis that cerebral changes are related to the presence of microvascular complications. However, we also found indications that cerebral and cognitive changes may be present before microvascular complications become apparent. It could be hypothesized that chronic hyperglycemia, even in the absence of clinically detectable microvascular complications, can negatively affect cognitive functioning and cerebral communication. It is known that hyperglycemia as such leads to a cascade of changes, including increases in formation of reactive oxygen species and advanced glycation end products (38,39), as well as activation of the hypothalamic-pituitary-adrenal axis (40), which could have an effect on the brain. Effects of hypoglycemia, particularly of asymptomatic hypoglycemic episodes because of hypoglycemia unawareness, cannot be completely ruled out as a contributor. Subsequently, we corrected for severe hypoglycemic events. This did not change the results (supplementary Table 2A–C, available in an online appendix).

Another possible contributor to the altered functional connectivity and cognitive functions observed in type 1 diabetic patients may be cerebrovascular, that is, macrovascular disease, the subclinical form of which can be estimated by ultrasound measurements of the carotid intima media thickness, which is known to be increased in asymptomatic type 1 diabetic patients relative to their nondiabetic peers (41). Furthermore, early diabetes onset, defined as disease onset before 7 years of age, has been shown previously to negatively affect intellectual and cognitive performance and cerebral structure (42,43).

Some limitations of this study should be mentioned. First, MEG provides a high temporal resolution. Unfortunately, the spatial resolution is lower. For a higher spatial resolution fMRI is currently the method of choice. Furthermore, the cognitive tests chosen had the highest sensitivity to detect cognitive changes. However, these tests may not be sufficiently sensitive to detect very subtle changes in type 1 diabetes–associated cognitive decline. Second, age of disease onset and subsequently disease duration differ between the patient groups. Although the percentages of early disease onset do not statistically differ between groups, for the type 1 diabetes⁺ patients age of onset reflects childhood and early adolescence, whereas for the type 1 diabetes⁻ patients it reflects late adolescence and early adulthood. In a subsequent analysis we compared both patient groups, additionally correcting for age of diabetes onset and diabetes duration (see supplementary Table 3A–C, available in an online appendix). Taking into account the risk of emerging power problems when adding two more covariates to this small sample, these adjustments did not essentially change the conclusions regarding cognitive functioning and functional connectivity. Disease duration might be less of a confounder here as the type 1 diabetes⁻ patients had an average disease duration of almost 20 years, which is sufficient for complications to develop. As stated before, an increase in functional connectivity was found for the lower α band in type 1 diabetes⁻ patients. Although the possibility of a statistical aberration cannot be completely ruled out, synchronization likelihood tended to be higher for 20 of the 23 anatomical locations of the lower α band in type 1 diabetes⁻ patients than control subjects (supplementary Table 4, available in an online appendix). Moreover, we additionally repeated the post hoc MANCOVA for the lower α band correcting for multiple comparisons. This did not change the statistical significance of the parietooccipital pathway. Thus, these findings suggest that the significant increase detected is valid. As both age and depressive symptoms differed between groups, we statistically corrected for both variables in all analyses. Supplementary Table 5, available in an online appendix, summarizes the relation between age and depressive symptoms and cognitive and MEG variables. Third, A1C values over the time period preceding MEG recording and neuropsychological assessment unfortunately were not available.

In conclusion, this study showed that changes in functional brain connectivity and cognitive function exist in type 1 diabetic patients with and without microvascular complications. Furthermore, interactions between functional connectivity and cognitive domains have been demonstrated. Whether these abnormalities have clinical implications remains to be determined. Longitudinal research in larger samples is required to determine the predictive clinical value of our findings for the progression of cognitive deterioration in type 1 diabetic patients as well as its potential to serve as clinical outcome parameter.

This research was supported by Grant 2005.00.06 of the Dutch Diabetes Research Foundation.

No potential conflicts of interest relevant to this article were reported.