Resting-State Brain Networks in Type 1 Diabetic Patients
Fifty-one T1DM MA patients and 53 patients with no clinically significant complications (MA) were studied, as were 51 healthy nondiabetic subjects. Because of technical problems, scans were not available for 6 participants, leaving 149 for the actual analysis. Outpatients were recruited from the departments of Internal Medicine and Ophthalmology of the VU University Medical Center and affiliated hospitals and through advertisements in a national newspaper and diabetes patient magazine.
Subjects were eligible if they were 18–56 years of age, right-handed, and proficient in Dutch and, for T1DM patients, had disease duration of at least 10 years. Exclusion criteria were a BMI >35 kg/m, history of or current cardio- and cerebrovascular events or epilepsy, pregnancy, contraindication for magnetic resonance imaging (MRI), centrally acting medication, drug or alcohol abuse, psychiatric disorder, thyroid dysfunction, anemia, or insufficient visual acuity to perform the neuropsychological assessment. MA patients were selected based on the presence of proliferative retinopathy, but other microvascular complications, such as microalbuminuria, could also be present. MA patients were free of all clinically detectable complications. (See below.) During the study, blood glucose levels of patients had to range between 4 and 15 mmol/L (72 and 270 mg/dL). Glucose levels were checked regularly during the study and corrected if necessary. No subject had glucose levels out of range before fMRI scanning. During neuropsychological testing, one participant experienced mild hypoglycemia; testing was postponed until blood glucose had normalized for 30 min following ingestion of an equivalent of 20 g carbohydrates. This study was conducted in accordance with the Declaration of Helsinki and approved by the medical ethics committee of the VU University Medical Center. Written informed consent was obtained from all participants.
Anthropometric measures were collected using a standardized questionnaire. Blood and 24-h urine sampling was performed for routine measures (hemoglobin, creatinine, lipid profile, liver enzymes, A1C, thyrotropin, and urine creatinine and albumin) on the study day. Retinopathy was assessed by fundus photography, which was rated according to the EURODIAB classification. Only patients with score 0 (no retinopathy) or scores 4 and 5 (proliferative retinopathy or laser coagulation) were included. Microalbuminuria was defined by an albumin-to-creatinine ratio (ACR) >2.5 mg/mmol for men and >3.5 mg/mmol for women. Peripheral neuropathy was based on the results of the annual checkup that patients receive and is incorporated into patients' medical record. This checkup consists of the assessment of vibration perception using a 128-Hz tuning fork and tactile perception with the 10-g Semmes-Weinstein monofilament. These data were used whenever patients were recruited from our and affiliated outpatient clinics (n = 89), whereas patients recruited from other hospitals were asked about the results of the examination (n = 12). Severe hypoglycemic events, defined according to DCCT guidelines, were self-reported across the lifetime. Because depressive symptomatology may be elevated in T1DM and may confound the study results, the Center of Epidemiological Studies scale for Depression was completed by all participants.
All participants underwent a detailed neuropsychological assessment covering the domains of general cognitive ability, memory, information-processing speed, executive functions, attention, and motor and psychomotor speed. The tests have previously been described. Raw scores were transformed into z scores based on the mean and SD values from control subjects. Higher z scores indicate better performance.
MRI scanning was performed on a 1.5T whole-body magnetic resonance system (Siemens Sonata, Erlangen, Germany) using an eight-channel phased-array head coil. Scans included a T1-based magnetization-prepared rapid-acquisition gradient echo (T1-MPRAGE) (repetition time 2,700 ms, echo time 5.17 ms, inversion time 950 ms, flip angle 8°, 248 × 330 mm field of view, 1.0 × 1.0 × 1.5 mm voxel size, and 160 contiguous coronal partitions) for registration purposes as well as 10 min of fMRI sequence (10 min, 202 volumes of echo-planar images, repetition time 2,850 ms, echo time 60 ms, flip angle 90°, 384 × 384 mm field of view, isotropic 3.3-mm voxels, and 36 axial slices). Scanning occurred in a darkened room, and subjects were asked to keep their eyes closed and not think of anything particular or fall asleep.
The Software Library (FSL 4.1) of the Functional MRI of the Brain (FMRIB) (http://www.fmrib.ox.ac.uk/fsl) was used for these analyses. The following preprocessing steps were applied to all images. After discarding the first two volumes to allow for occurrence of a steady state, the remaining 200 volumes were motion corrected, brain extracted, and smoothed using a Gaussian kernel of 5 mm. High-pass filtering was applied using a cutoff of 150.0 s. Each scan was first registered to each subject's high-resolution T1-MPRAGE scan using an affine registration (6 df) and afterward nonlinearly registered to standard space (MNI152) using a warp resolution of 10 mm. Finally, the registered fMRI sequences were temporally concatenated into a single four-dimensional dataset. This dataset was then analyzed using ICA to identify large-scale patterns of connectivity across the entire study population. For checking of errors in preprocessing for ICA analysis, all steps were monitored manually; i.e., scan coverage, brain extraction, and excessive motion were checked for. No scans met criteria for excessive motion; all data were used. After this analysis, dual-regression analysis (part of FSL4.1) was performed, the aim of which was to create personalized maps of each network for every subject. The first step of the dual-regression is creating the average time course within each network for each subject, which is done using a linear model fit of each group-based network map onto each subject's fMRI dataset (spatial regression). After this, the personalized time course is regressed back onto that subject's fMRI dataset to create personal spatial maps for each network after variance normalization using another linear model fit (temporal regression). A single four-dimensional file of all components was created for each participant. Because of the normalization of the variance of the time series used in the final regression, these spatial maps reflect both amplitude of spontaneous fluctuation in a network and its coherence (correlated BOLD signals) across space. The individual values in these maps, therefore, represent connectivity in a more sophisticated way than just a coherence measure (which implies independence of amplitude).
Group differences were tested using nonparametric permutation testing (5,000 permutations), corrected for age, sex, systolic blood pressure, and depressive symptoms. Data were corrected for multiple comparison using threshold-free cluster enhancement, which allows the identification of clusters of significant voxels without having to define them in a binary way, as well as family-wise error, using a final corrected threshold of P < 0.05. We investigated four different contrasts: healthy control subjects compared with 1) all T1DM patients, 2) only MA patients, 3) MA patients and 4) MA contrasted with MA patients. Each contrast was tested in two directions separately (i.e., increases or decreases), resulting in eight final contrasts.
For each resting-state network, the mean connectivity z value per subject was extracted for further inspection, using a threshold of z > 3.9, corresponding to P < 0.0001. In networks where group differences were significant, an additional mean z score of significant voxels was calculated per subject.
Demographic, medical, and anthropometric measures were analyzed using one-way ANOVA with Bonferroni correction, Student t test, or χ, where appropriate. Cognitive domains were compared using a multivariate ANCOVA, corrected for age, sex, systolic blood pressure, and depressive symptoms. In case of a significant multivariate F test, post hoc individual tests were checked for significance. Group differences were Bonferroni corrected.
For determination of associations between demographic, medical, and cognitive variables and changed functional connectivity, regression analyses were performed. For each network, all demographic (age, sex, BMI, depressive symptoms, and systolic blood pressure) and medical (disease duration, onset age, severe hypoglycemic events, microangiopathy, ACR, and A1C) or cognitive domains, corrected for age, sex, systolic blood pressure, and depressive symptoms, were entered in one block as independent variables. A forward regression was used to determine variables that were significantly associated with resting-state networks. This resulted in 10 regression analyses, with 2 for each network.
All fMRI statistics were performed in FSL4.1. All other statistical analyses were performed in SPSS 15.0 (SPSS, Chicago, IL). A P value <0.05 was considered statistically significant.
Research Design and Methods
Fifty-one T1DM MA patients and 53 patients with no clinically significant complications (MA) were studied, as were 51 healthy nondiabetic subjects. Because of technical problems, scans were not available for 6 participants, leaving 149 for the actual analysis. Outpatients were recruited from the departments of Internal Medicine and Ophthalmology of the VU University Medical Center and affiliated hospitals and through advertisements in a national newspaper and diabetes patient magazine.
Subjects were eligible if they were 18–56 years of age, right-handed, and proficient in Dutch and, for T1DM patients, had disease duration of at least 10 years. Exclusion criteria were a BMI >35 kg/m, history of or current cardio- and cerebrovascular events or epilepsy, pregnancy, contraindication for magnetic resonance imaging (MRI), centrally acting medication, drug or alcohol abuse, psychiatric disorder, thyroid dysfunction, anemia, or insufficient visual acuity to perform the neuropsychological assessment. MA patients were selected based on the presence of proliferative retinopathy, but other microvascular complications, such as microalbuminuria, could also be present. MA patients were free of all clinically detectable complications. (See below.) During the study, blood glucose levels of patients had to range between 4 and 15 mmol/L (72 and 270 mg/dL). Glucose levels were checked regularly during the study and corrected if necessary. No subject had glucose levels out of range before fMRI scanning. During neuropsychological testing, one participant experienced mild hypoglycemia; testing was postponed until blood glucose had normalized for 30 min following ingestion of an equivalent of 20 g carbohydrates. This study was conducted in accordance with the Declaration of Helsinki and approved by the medical ethics committee of the VU University Medical Center. Written informed consent was obtained from all participants.
Biomedical, Anthropometric, and Psychological Measures
Anthropometric measures were collected using a standardized questionnaire. Blood and 24-h urine sampling was performed for routine measures (hemoglobin, creatinine, lipid profile, liver enzymes, A1C, thyrotropin, and urine creatinine and albumin) on the study day. Retinopathy was assessed by fundus photography, which was rated according to the EURODIAB classification. Only patients with score 0 (no retinopathy) or scores 4 and 5 (proliferative retinopathy or laser coagulation) were included. Microalbuminuria was defined by an albumin-to-creatinine ratio (ACR) >2.5 mg/mmol for men and >3.5 mg/mmol for women. Peripheral neuropathy was based on the results of the annual checkup that patients receive and is incorporated into patients' medical record. This checkup consists of the assessment of vibration perception using a 128-Hz tuning fork and tactile perception with the 10-g Semmes-Weinstein monofilament. These data were used whenever patients were recruited from our and affiliated outpatient clinics (n = 89), whereas patients recruited from other hospitals were asked about the results of the examination (n = 12). Severe hypoglycemic events, defined according to DCCT guidelines, were self-reported across the lifetime. Because depressive symptomatology may be elevated in T1DM and may confound the study results, the Center of Epidemiological Studies scale for Depression was completed by all participants.
Neuropsychological Testing
All participants underwent a detailed neuropsychological assessment covering the domains of general cognitive ability, memory, information-processing speed, executive functions, attention, and motor and psychomotor speed. The tests have previously been described. Raw scores were transformed into z scores based on the mean and SD values from control subjects. Higher z scores indicate better performance.
MRI Procedure
MRI scanning was performed on a 1.5T whole-body magnetic resonance system (Siemens Sonata, Erlangen, Germany) using an eight-channel phased-array head coil. Scans included a T1-based magnetization-prepared rapid-acquisition gradient echo (T1-MPRAGE) (repetition time 2,700 ms, echo time 5.17 ms, inversion time 950 ms, flip angle 8°, 248 × 330 mm field of view, 1.0 × 1.0 × 1.5 mm voxel size, and 160 contiguous coronal partitions) for registration purposes as well as 10 min of fMRI sequence (10 min, 202 volumes of echo-planar images, repetition time 2,850 ms, echo time 60 ms, flip angle 90°, 384 × 384 mm field of view, isotropic 3.3-mm voxels, and 36 axial slices). Scanning occurred in a darkened room, and subjects were asked to keep their eyes closed and not think of anything particular or fall asleep.
fMRI Resting-state Analysis
The Software Library (FSL 4.1) of the Functional MRI of the Brain (FMRIB) (http://www.fmrib.ox.ac.uk/fsl) was used for these analyses. The following preprocessing steps were applied to all images. After discarding the first two volumes to allow for occurrence of a steady state, the remaining 200 volumes were motion corrected, brain extracted, and smoothed using a Gaussian kernel of 5 mm. High-pass filtering was applied using a cutoff of 150.0 s. Each scan was first registered to each subject's high-resolution T1-MPRAGE scan using an affine registration (6 df) and afterward nonlinearly registered to standard space (MNI152) using a warp resolution of 10 mm. Finally, the registered fMRI sequences were temporally concatenated into a single four-dimensional dataset. This dataset was then analyzed using ICA to identify large-scale patterns of connectivity across the entire study population. For checking of errors in preprocessing for ICA analysis, all steps were monitored manually; i.e., scan coverage, brain extraction, and excessive motion were checked for. No scans met criteria for excessive motion; all data were used. After this analysis, dual-regression analysis (part of FSL4.1) was performed, the aim of which was to create personalized maps of each network for every subject. The first step of the dual-regression is creating the average time course within each network for each subject, which is done using a linear model fit of each group-based network map onto each subject's fMRI dataset (spatial regression). After this, the personalized time course is regressed back onto that subject's fMRI dataset to create personal spatial maps for each network after variance normalization using another linear model fit (temporal regression). A single four-dimensional file of all components was created for each participant. Because of the normalization of the variance of the time series used in the final regression, these spatial maps reflect both amplitude of spontaneous fluctuation in a network and its coherence (correlated BOLD signals) across space. The individual values in these maps, therefore, represent connectivity in a more sophisticated way than just a coherence measure (which implies independence of amplitude).
Group differences were tested using nonparametric permutation testing (5,000 permutations), corrected for age, sex, systolic blood pressure, and depressive symptoms. Data were corrected for multiple comparison using threshold-free cluster enhancement, which allows the identification of clusters of significant voxels without having to define them in a binary way, as well as family-wise error, using a final corrected threshold of P < 0.05. We investigated four different contrasts: healthy control subjects compared with 1) all T1DM patients, 2) only MA patients, 3) MA patients and 4) MA contrasted with MA patients. Each contrast was tested in two directions separately (i.e., increases or decreases), resulting in eight final contrasts.
For each resting-state network, the mean connectivity z value per subject was extracted for further inspection, using a threshold of z > 3.9, corresponding to P < 0.0001. In networks where group differences were significant, an additional mean z score of significant voxels was calculated per subject.
Statistical Analysis
Demographic, medical, and anthropometric measures were analyzed using one-way ANOVA with Bonferroni correction, Student t test, or χ, where appropriate. Cognitive domains were compared using a multivariate ANCOVA, corrected for age, sex, systolic blood pressure, and depressive symptoms. In case of a significant multivariate F test, post hoc individual tests were checked for significance. Group differences were Bonferroni corrected.
For determination of associations between demographic, medical, and cognitive variables and changed functional connectivity, regression analyses were performed. For each network, all demographic (age, sex, BMI, depressive symptoms, and systolic blood pressure) and medical (disease duration, onset age, severe hypoglycemic events, microangiopathy, ACR, and A1C) or cognitive domains, corrected for age, sex, systolic blood pressure, and depressive symptoms, were entered in one block as independent variables. A forward regression was used to determine variables that were significantly associated with resting-state networks. This resulted in 10 regression analyses, with 2 for each network.
All fMRI statistics were performed in FSL4.1. All other statistical analyses were performed in SPSS 15.0 (SPSS, Chicago, IL). A P value <0.05 was considered statistically significant.