Central Nervous System Function in Youth With Type 1 Diabetes 12 Years After Disease Onset

Consecutive patients admitted to the Royal Children's Hospital, Melbourne, with newly diagnosed type 1 diabetes between 1990 and 1992 (n = 133) together with healthy control subjects (n = 126), stratified for age and sex. A history of CNS disease or trauma was an exclusion criterion. Twelve years later, all participants who could be located (125 type 1 diabetic patients and 93 control subjects: 94 and 74%, respectively) using the study database, Royal Children's Hospital, Melbourne and adult diabetes clinics, private endocrinologists, and the Health Insurance Commission were invited to take part in the current study. Of those located, 106 type 1 diabetic and 75 control subjects agreed to participate (rates of 85 and 81%, respectively).

All participants underwent neurocognitive testing. Participants were consecutively invited to undergo neuroimaging until available funding was exhausted. Blood glucose levels (BGLs) of diabetic participants were determined before assessment by capillary sample to ensure that subjects had a reading >4 mmol/l. This study was approved by the Human Ethics Research Committee of the Victorian Government Department of Human Services.

The Wechsler Abbreviated Scale of General Intelligence (13) is a standardized, brief measure of intelligence providing full scale (FSIQ), verbal (VIQ), and performance (PIQ) IQ scores (mean ± SD 100 ± 15).

Imaging was carried out on a 3-T scanner (GE Healthcare, Milwaukee, WI).

Bilateral single-voxel spectra were acquired using a standard PRESS sequence to provide a metabolite profile from three brain regions of interest (ROIs): basal ganglia and frontal and temporal lobes. The ROIs were positioned using T1-weighted scout images. Basal ganglia ROIs were positioned over the lentiform nuclei, temporal lobe ROIs were positioned to include hippocampus and mesial temporal structures, and frontal lobe ROIs were placed above the third ventricle sufficiently posterior to avoid interference from curvature of the skull. Acquisition parameters were TE/TR = 30/3000 ms. For frontal and temporal spectra, isotropic 2-cm voxels were used, and for basal ganglia voxel size was 2 × 2 × 1.5 cm. Data were analyzed using software packages, SAGE/IDL (GE Healthcare; WI/RSI, Boulder, CO) and LCModel (14), with a basis set of 15 metabolites acquired on the same spectrometer. Data were included in analyses if the Cramer-Rao lower bound was <30% as determined by LCModel. Results are presented in institutional units approximating millimoles per liter. In vivo proton magnetic resonance spectroscopy (MRS) of brain provides information on the tissue content of total N-acetylaspartate (NAA) (including N-acetylaspartylglutamate), creatine + creatine phosphate, myoinositol, trimethylamines or choline-containing metabolites (Cho), and glutamine + glutamate.

A fast spoiled gradient-recalled echo at steady-state sequence was used (TR/TE/TI = 13.8/2.7/500 ms, voxel size 0.48 × 0.48 × 2 mm). For T2 relaxometry, a modified, optimized Carr-Purcell-Meiboom-Gill multiecho sequence (14) was used (eight echoes, TE = 28.9–231 ms, TR = 6.24 s, 24 slices, 5-mm slice thickness, in-plane voxel size 0.94 × 1.88 mm). The slice plane was perpendicular to the long axis of the hippocampus. T2 maps were generated by fitting to a monoexponential model with the inclusion of a baseline that minimizes the contribution of long T2 components (mainly cerebrospinal fluid) to the fit. For analyses, images were warped to standard space in which they could be compared and smoothed (with a 10-mm kernel). All analyses were performed using SPM2 (http://www.fil.ion.ucl.ac.uk/spm/software/spm2). Separate gray matter and white matter volumetry analyses were performed using optimized voxel-based morphometry (VBM) (15). Voxel-wise T2 changes were assessed using the approach of voxel-based relaxometry (16).

Participants with diabetes diagnosed at age ≤5 years were classified as having early-onset disease. Participants reported any episodes from diagnosis of hypoglycemia associated with seizure/coma, and these were corroborated through scrutiny of medical records. The sample was dichotomized into those who reported no events and those with a history of ≥1 event. A1C measurements from diagnosis were obtained for each patient (range 9–57, median 37) from hospital and clinic databases. The percentage of total time from diagnosis that A1C was ≥9.0% was calculated to estimate overall metabolic control.

SPSS (version 14; SPSS, Chicago, IL) was used for statistical analyses of demographic, IQ, and MRS data.

Group differences were examined using ANCOVA, covarying for socioeconomic status (SES), FSIQ_baseline, and time between baseline and current assessment.

Mean bilateral metabolite concentrations in each ROI were used in analyses. Group differences were examined using ANCOVA with age as a covariate.

ANCOVA was used to examine group differences in volume and T2, covarying for age. Volume analyses also included total brain volume and sex as covariates.

Multiple linear regression was used to predict IQ from illness variables (age of disease onset, hypoglycemia, and metabolic control), SES, FSIQ_baseline, and time between baseline and current assessment. Age of disease onset was highly correlated with age; hence, age was used as a predictor for regression analyses of MRS/magnetic resonance imaging (MRI) data, together with hypoglycemia and metabolic control. MRI regression analyses were restricted to anatomical areas shown to differ in the initial type 1 diabetic and control comparisons.

Sample characteristics of participants are presented in Table 1. Type 1 diabetic and control participants did not differ on FSIQ_baseline, age, sex ratio, SES, or psychiatric symptoms. Time between baseline and current assessment was shorter in type 1 diabetic subjects (95% CI 0.37–1.0 years). Type 1 diabetic and control participants who underwent neuroimaging did not differ significantly on age or sex ratio, and there were no differences between type 1 diabetic subjects who had neuroimaging and those who did not on age of disease onset, hypoglycemia, or metabolic control. Group mean scores ± SD on IQ and MRS and MRI variables are presented in Tables 2 and 3.

Type 1 diabetic subjects had lower VIQ (P = 0.03) and FSIQ (P = 0.05) than control subjects.

The largest mean differences were in myoinositol and NAA, with myoinositol higher and NAA lower in type 1 diabetic participants. Smaller statistically significant mean differences (type 1 diabetic participants higher) were found in Cho in all locations.

Type 1 diabetic subjects, relative to control subjects, had decreased gray matter volume in bilateral thalami, right parahippocampal gyrus, and right insular cortex (supplemental Figure A1, available in online appendix at http://dx.doi.org/10.2337/dc08-57). Mean white matter volume was decreased in bilateral mesial temporal lobes (parahippocampal region), in other areas of the left temporal lobe, and in the left-middle frontal area. T2 relaxation times in type 1 diabetic participants were increased in the left superior temporal gyrus and decreased in bilateral lentiform nuclei, caudate nuclei and thalami, and the right insular area (all P < 0.0005 uncorrected).

Regression analyses of illness-related predictors of IQ, MRS, and MRI findings are presented in supplemental Table A1, available in an online appendix.

In each regression, SES and FSIQ_baseline contributed significantly to the model. In addition, hypoglycemia predicted lower VIQ (R² change = 0.032, P = 0.01) and early disease onset predicted lower PIQ (R² change = 0.135, P < 0.001) and FSIQ (R² change = 0.064, P < 0.001). The verbal IQ of the hypoglycemia subgroup was nearly ⅓ SD below that for type 1 diabetic patients with no hypoglycemia (adjusted mean ± SEM 93.8 ± 1.3 vs. 98.2 ± 1.1). Early-onset participants had a PIQ more than 1/2 SD and an FSIQ ⅓ SD below that for those with later-onset type 1 diabetes (101.6 ± 2.5 vs. 109.2 ± 1.7 and 97.8 ± 2.1 vs. 103.4 ± 1.4, respectively).

Poor metabolic control predicted higher levels of myoinositol in basal ganglia (R² change = 0.055, P = 0.04), and older age predicted lower levels of glutamine + glutamate in frontal lobes (R² change = 0.116, P < 0.01) and NAA in temporal lobes (R² change = 0.133, P < 0.01).

Regression analyses of MRI data in areas implicated in the previous group comparisons revealed the dominating influence of age in explaining variation in volume and T2: R² change = 0.247, P < 0.001 (gray matter volume lentiform); R² change = 0.060, P < 0.02 (gray matter volume thalamus); R² change = 0.578, P < 0.001 (T2 lentiform); and R² change = 0.418, P < 0.001 (T2 thalamus). Hypoglycemia predicted reduced gray matter volume in the thalamus (R² change = 0.045, P < 0.03), and poor metabolic control was associated with decreased T2 in the thalamus (R² change = 0.021, P < 0.05).

In view of the strong association between older age and volume and T2 reduction in type 1 diabetic subjects, we investigated how group and age predicted each of the MRI variables. The models included an interaction term of group by age, which was statistically significant in all analyses. For volume, the models predicted small positive changes with age for control subjects and relatively larger negative changes for type 1 diabetic subjects (lentiform, P_interaction = 0.002, age slopes: type 1 diabetic subjects −0.0038, control subjects 0.0001; thalamus, P_interaction = 0.046, age slopes: type 1 diabetic subjects −0.0025, control subjects 0.0012). For T2, the models predicted relatively larger negative changes for type 1 diabetic subjects than for control subjects (lentiform, P_interaction = 0.013, age slopes: type 1 diabetic subjects −5.11, control subjects −3.17; thalamus, P_interaction = 0.017, age slopes: type 1 diabetic subjects = −5.25, control subjects = −2.85).

Correlations between BGL before neuropsychological assessment and IQ scores and between BGL before scanning and metabolite profiles were calculated and ranged from r = 0.001 to r = −0.242 (all P >0.05). To further examine the possible confounding of intercurrent hyperglycemia on study findings, data from type 1 diabetic participants with BGLs >15 mmol/l at the time of testing (n = 34) were removed and the data were reanalyzed. Group differences on VIQ and FSIQ became trends only (P < 0.06 and P < 0.07, respectively). The significance of the group difference on basal ganglia NAA was also reduced (P < 0.08) after removal of type 1 diabetic participants with BGL >15 (n = 19) at the time of scanning. All other group differences and illness-related predictors of IQ and metabolites were unchanged.

This study is the first to document CNS effects 12 years after diagnosis in youth with type 1 diabetes whose neurocognitive profile at disease onset was not different from that of control subjects. Metabolites and IQ results were largely unchanged after removal of data from type 1 diabetic participants with high BGLs at the time of assessment, suggesting that study findings reflect stable changes in CNS function. Group differences on IQ were marginal and appear to reflect the selective impact of specific illness risk factors. Lower VIQ was associated with a positive history of hypoglycemia, whereas early-onset disease predicted lower PIQ and FSIQ. The association between early-onset disease and lower IQ, particularly PIQ, is a consistent finding (7,8). Hypoglycemia-related effects on VIQ have also been reported previously in the pediatric literature (8,12) but were not found in the DCCT (11), suggesting that the developing CNS may be especially vulnerable to hypoglycemia. It is important to note that illness-related effects on IQ, although statistically significant, are small, and individuals with type 1 diabetes function well within the average range. However, even mild decrements in ability may have functional significance during childhood when new knowledge is being acquired. Of participants who had reached school-leaving age at follow-up (type 1 diabetic subjects, n = 76; control subjects, n = 55), 17% fewer type 1 diabetic subjects than control subjects (68 vs. 85%, P < 0.01) completed year 12, the pretertiary year of education in Australia.

This study showed lower mean NAA and higher mean myoinositol and Cho in type 1 diabetic subjects than in control subjects. The magnitudes of these differences were in the range of 10–15% and are similar to those reported for diseases such as dementia, epilepsy, and Parkinson's disease (17,18). There are also similarities between current findings and previous reports in diabetes (4,5,10), although direct comparison is difficult because methodology has varied, with different brain regions assessed and some samples including both type 1 and type 2 diabetic participants. With one exception (10), study populations have been older than the current sample. NAA is a marker of neuronal density or viability, with lower levels indicative of neuronal death and/or decreased neuronal metabolism (17). Lower mean NAA is consistent with animal models of diabetes suggesting that chronic hyperglycemia reduces neuronal number and impedes myelination (19). Although we were not able to relate NAA directly to metabolic control in the current study, NAA levels correlated inversely with lifetime glycemic exposure in another report (5).

Myoinositol is a marker of changes in osmolarity, and increased levels are associated with both gliosis and demyelination (17). In the current study, poorer metabolic control was associated with higher levels of myoinositol in basal ganglia which, together with other reports of elevated myoinositol in patients recovering from diabetic ketoacidosis (20), suggests that higher levels may represent a homeostatic response of the brain to prolonged hyperglycemia. Cho levels are increased with demyelination and other forms of cell membrane breakdown, including gliosis (17). In diabetes, increases in both myoinositol and Cho are consistent with glial proliferation due to tissue hypoxia in the context of chronically elevated blood glucose levels (21). As with a previous report (10), metabolite profiles were not related to history of hypoglycemia.

Volumetric measurements can identify regionally specific macroscopic atrophy associated with neuronal degeneration. Musen et al. (2), using VBM, reported reduced gray matter densities in temporal brain regions and left thalamus. Our finding of decreased mean gray matter in thalamus is similar, but, in addition, we found reduced mean gray matter volume in insular cortex and frontal precentral regions and a reduction in mean white matter volume in mesial temporal areas. Previous reports have linked VBM changes to a previous history of either chronic hyperglycemia (2,3,8) and/or severe hypoglycemia (2,8). In the current study, hypoglycemia was associated with volume reduction in the thalamus, poor control predicted decreased T2 in the thalamus, and there were strong associations between older age and reduced volume and T2 in anterior and temporal brain regions as well as thalamus, caudate, and lentiform nuclei.

The volume group differences found in this study reach levels of ∼10%, comparable to changes in temporal lobe epilepsy in which volume reduction is ∼10–15% (22). T2 group differences of 3% in the current study are smaller than those of 5–20% reported in other CNS diseases (23) and may in part reflect an osmotic influence of acutely elevated glucose on tissue water distribution as well as or instead of any longer-term impact of the illness. In the current study, older age was the strongest predictor of volume and T2 decrease in type 1 diabetic subjects. Older age is a surrogate for later disease onset in this cohort, as duration of illness was controlled; hence, these findings are counterintuitive, given the consistent association between early disease onset and neurocognitive deficits reported previously (7,8,11). It is now recognized, though, that CNS maturation continues into the third decade of life (24), and it is possible that findings reflect an interaction between type 1 diabetes and the final stages of neurodevelopment, with the early-onset subgroup yet to experience this disruption. The suggestion by Biessels et al. (25) that diabetes is a model for “accelerated [CNS] aging” provides an alternative explanation for our findings. That is, rather than reflecting disrupted neurodevelopment, the greater volume loss in our older and later-onset participants may represent the earliest stages of cerebral microvascular disease or some other neurodegenerative process.

The current findings are suggestive of a number of possible illness-related neuropathological processes, including gliosis, demyelination, and changes in osmolarity and neural cell type/viability. Anterior and temporal brain regions have particularly high glucose demands and increased sensitivity to glucose disruption (1); thus, it not surprising that these brain regions show reduced brain volumes, altered T2 relaxation times, and metabolite changes. Relationships between specific illness variables and CNS effects are more difficult to disentangle. There were strong associations between age and IQ and structural brain changes but in the opposite direction. Early-onset diabetes (and younger age) was associated with lower PIQ and FSIQ, whereas older age predicted lower levels of NAA, reduced brain volumes, and altered T2. Metabolite profiles are more suggestive of hyperglycemia than of hypoglycemia-mediated neurotoxicity in the developing CNS. It is possible that specific illness variables exert different effects on the CNS, but inconsistent associations may also reflect difficulties in obtaining reliable and comprehensive metabolic control histories, including documentation of diabetes complications. Further studies, involving large multicenter cohorts, meticulous collection of metabolic control histories, documentation of diabetes complications, and tight control over the glycemic level of participants at the time of assessment, are needed to fully understand the pathogenesis of CNS changes in childhood-onset diabetes.

This research was supported by Juvenile Diabetes Foundation International (grant 1-2003-135). E.A.N. and F.J.C. received the Australian Diabetes Society-Servier National Action Plan grant, the beyondblue Victorian Centre of Excellence in Depression grant, and National Health and Medical Research Council Project Grant 334354. E.A.N. also received grant support from Australian Rotary Health Research Fund.

E.A.N. received lecture fees from Novo Nordisk and research grant support from Eli Lilly. G.A.W. received consulting fees from IPSEN, has equity in Antisense Therapeutics Limited (Australia) and Neuren (Australia), and is currently receiving research support from Novo Nordisk and Sandoz. F.J.C. received lecture fees and/or consultancy fees from Novo Nordisk, Eli Lilly, and Medtronic and research grants from Eli Lilly and Medtronic. No other potential conflicts of interest relevant to this article were reported.

We acknowledge the contributions of Deb Boyce, SN, in locating and recruiting participants and Professor Graeme Jackson of the Brain Research Institute, Austin Health, for his support for the project.