Dr. Lyoo has received research support from Lundbeck, AstraZeneca, GSK, and Boryung Pharmaceutical. Dr. Renshaw has been a consultant to Roche, Novartis, Kyowa Hakko, and Repligen, and he has received research support from GlaxoSmithKline and Roche. Other authors have no conflicts of interest in relation to the subject of this study. This does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials.
Conceived and designed the experiments: IKL PFR GM AMJ. Performed the experiments: IKL PFR GM NB KW AMJ. Analyzed the data: IKL SY PFR JH SB GM JEK HSJ SHL KW JJJ YC AMJ. Contributed reagents/materials/analysis tools: IKL PFR AMJ. Wrote the paper: IKL SY PFR JH SB GM JEK NB HSJ DCS SHL KW JJJ CMR YC AMJ. Supervised study: IKL PFR AMJ. Critically revised the manuscript: IKL SY PFR JH SB GM JEK NB HSJ DCS SHL KW JJJ CMR YC AMJ.
Type 1 diabetes mellitus (T1DM) usually begins in childhood and adolescence and causes lifelong damage to several major organs including the brain. Despite increasing evidence of T1DM-induced structural deficits in cortical regions implicated in higher cognitive and emotional functions, little is known whether and how the structural connectivity between these regions is altered in the T1DM brain. Using inter-regional covariance of cortical thickness measurements from high-resolution T1-weighted magnetic resonance data, we examined the topological organizations of cortical structural networks in 81 T1DM patients and 38 healthy subjects. We found a relative absence of hierarchically high-level hubs in the prefrontal lobe of T1DM patients, which suggests ineffective top-down control of the prefrontal cortex in T1DM. Furthermore, inter-network connections between the strategic/executive control system and systems subserving other cortical functions including language and mnemonic/emotional processing were also less integrated in T1DM patients than in healthy individuals. The current results provide structural evidence for T1DM-related dysfunctional cortical organization, which specifically underlie the top-down cognitive control of language, memory, and emotion.
The central nervous system (CNS) serves as one of the representative end-organ targets of metabolic insult related to type 1 diabetes mellitus (T1DM)
To render the human brain capable of exerting its characteristic functions, it is essential not only that individual neuroanatomical areas responsible for specific function are operational but also that the complex and hierarchal networks between corresponding areas are comparably efficient
The study protocol was approved by the institutional committee on human subjects of the Joslin Diabetes Center, Boston, MA, USA, the Brain Imaging Center of the McLean Hospital, Boston, MA, USA, and the Seoul National University Hospital, Seoul, South Korea. All participants were given a complete explanation of the study details and then provided written informed consent to participate in the study.
Adult T1DM patients between the ages of 25 and 40 and age- and sex-matched healthy control subjects were recruited from the Joslin Diabetes Center. The sample has been described in detail elsewhere
Acquisition of high-resolution of MR images was performed at the Brain Imaging Center of the McLean Hospital. Subjects were excluded if they had major neurological or medical disorders, if they had ever been diagnosed with psychosis, schizophrenia, bipolar disorder, attention-deficit hyperactivity disorder, or cocaine, heroin, or alcohol dependence, as assessed using the Structural Clinical Interview for DSM-IV
Characteristics | T1DM patients (n = 81) | Control subjects (n = 38) | |
Mean age (SD), |
32.5 (4.6) | 30.8 (5.1) | 0.08 |
Female, |
41 (50.6) | 19 (50.0) | 0.95 |
Right handedness, |
79 (97.5) | 35 (92.1) | 0.33 |
Duration of illness (SD), |
19.8 (3.5) | NA | NA |
Onset age (SD), |
12.8 (5.1) | NA | NA |
Lifetime average HbA1C (SD) |
7.99 (1.19) | NA | NA |
No. of hypoglycemic episodes (SD) |
6.41 (14.7) | NA | NA |
Current HbA1C (SD), |
7.73 (1.43) | 5.08 (0.33) | <0.001 |
Group differences were tested by independent t-tests or χ2 tests appropriately.
Average value of HbA1C, grouped and time-weighted every 4 years for the duration of illness.
A severe hypoglycemic episode was defined as an event that leads to a coma, seizure, or unconsciousness according to the Diabetes Control and Complications Trial Research Group Criteria.
Abbreviations: T1DM, type 1 diabetes mellitus; SD, standard deviation; HbA1C, hemoglobin A1C; NA, not available or not applicable.
Brain MR images were obtained using a 1.5 Tesla GE whole body imaging system (Horizon LX, GE Medical systems, Milwaukee, Wisconsin, USA) at the Brain Imaging Center of the McLean Hospital. Coronal T1 weighted images were produced using a three–dimensional spoiled gradient echo pulse sequence (124 slices, slice thickness = 1.5 mm, echo time [TE] = 5 ms, repetition time [TR] = 35 ms, matrix = 256×192; field of view [FOV] = 24 cm, flip angle [FA] = 45°, number of excitations [NEX] = 1). Axial T2 weighted images (TE = 80 ms, TR = 3,000 ms, 256×192 matrix; FOV = 20 cm, FA = 90°, NEX = 0.5, slice thickness/gap = 3/0 mm) and Fluid Attenuated Inversion Recovery axial images (TE = 133 ms, TR = 9,000 ms, inversion time = 2,200 ms, matrix = 256×192, FOV = 20 cm, FA = 90°, NEX = 1, slice thickness/gap = 5/2 mm) were obtained to screen for potential brain structural abnormalities.
Cortical thickness measurements were conducted using the FreeSurfer Tools (
Cerebral cortices on MR images were parcellated into 32 gyral-based regions for each hemisphere, using an automated labeling system that is distributed in the FreeSurfer Tool (Figure S1 in
Parcellated regions were further categorized into five segregated intrinsic structural network systems subserving different neurobehavioral functions including strategic/executive control, language, mnemonic/emotional processing, sensorimotor, and visual functions based on prior shared knowledge on structural and functional clusters (Figure S1 in
Partial correlation analyses of cortical thickness in each possible pair of 64 parcellated regions after adjusting for age and sex were used for the measurement of structural associations between regions. Interregional connection matrices were made for each of the T1DM and control groups by calculating the partial correlation coefficients for all subjects in each study group, representing the specific anatomical connections between each of the 2016 [ = (64×63)/2] possible pairs of parcellated regions. Each connection matrix was thresholded into a binarized matrix with a fixed sparsity (
Segregation and integration of whole-brain structural network reflect the efficiency of connections within densely interconnected regions and throughout remote brain regions, respectively
The length of path reflecting potential routes of information between regions could represent an estimate for efficiency of network integration
Based on evidence of the prefrontal regions' vital role in top-down control for other brain function
Among various measures of centrality
As supplementary analyses, modularity of the structural cortical network (
Group differences in demographic characteristics were analyzed using the independent t-test or chi-square test.
Differences in topological parameters for structural cortical network between the T1DM and control groups were examined using a non-parametric permutation test
We first assessed the global and local efficiencies of whole-brain structural networks, and compare the results between the T1DM and control groups. The global efficiency values (
The graphs showed the differences in network parameters between the T1DM and control groups (blue line). The mean values and 95% of confidence interval of the null distribution of between-group differences obtained from 1000 permutation tests at each sparsity level were represented as gray circles and error bars, respectively. Asterisks indicate significant differences in parameters between the T1DM and control groups at
In order to examine whether the hierarchical cognitive control of the prefrontal cortex was altered in T1DM patients, we investigated inter-network efficiencies between structural network systems for strategic/executive control and other major brain functions including language, mnemonic/emotional processing, and sensorimotor function.
Average inverse shortest path length (
Connections of cortical structural network for strategic/executive control with other networks mediating language (A), mnemonic/emotional processing (B), and sensorimotor function (C) and between-group differences in inter-network efficiencies are presented in the left and right panels, respectively. Brain templates in figures demonstrate cortical parcellated regions for corresponding intrinsic cortical structural sub-network systems and inter-network connections at the sparsity threshold of 0.23. Red arrows in graphs indicate the sparsity of 0.23 that whole-brain structural networks of both control and T1DM groups included all 64 connected brain regions. Hub regions shown in
Sixty-four gyral-based parcellated regions were defined to generate a correlation matrix (Figure S1 in
As shown in
Regions (brain templates of panels A for control and B T1DM subjects) in orange, green, blue, yellow, and light purple colors represent each intrinsic cortical structural sub-network system subserving strategic/executive control, language, mnemonic/emotional processing, sensorimotor, and visual functions, respectively. Figures depict hub regions and significant inter-hub structural connections of control (A) and T1DM (B) subjects at the sparsity threshold of 0.23. A given region was identified as a hub of whole-brain structural networks if its normalized betweenness-centrality (
Modular organization of structural cortical network in the T1DM and control groups was presented in Figure S4 in
Furthermore, the β values, which indicate the level of hierarchy in whole-brain structural networks, were also found to be larger in control subjects than in T1DM patients in a range of the sparsity (0.13≤
Our data provide novel evidence of T1DM-related alterations in the topological organization of structural brain networks, which was assessed by using cortical thickness data from high-resolution T1-weighted MR imaging. We found that hierarchically high-level prefrontal hubs were relatively absent in T1DM patients, and that the levels of the hierarchical organization of whole-brain structural networks were also lower in T1DM patients than in control subjects. By utilizing the division of brain regions into segregated intrinsic sub-network systems with functional significance, we also found that the structural connections between the prefrontal regions exerting top-down cognitive controls and other regions subserving complex brain functions including language and mnemonic/emotional processing were likely to be less integrated in T1DM patients relative to control subjects.
Previous studies have reported that the normal human brain demonstrates significant levels of hierarchical organization, as it is characterized by optimal connections for top-down control with minimum wiring costs
The absence of hierarchically organized hubs in key regions, such as the prefrontal regions, suggests that the contribution of these regions to the network efficiency may be reduced in T1DM patients. Our inter-network analysis results (
A recent functional study, which assessed resting-state brain networks in young adults with T1DM, has demonstrated similar findings to ours
Although sample characteristics (old age vs. young age) and measurements for network construction (white matter connectivity vs. cortical thickness) might be different from ours, it should be noted that the structural white matter network for the type 2 diabetes mellitus (T2DM) brain was also altered in terms of decreased global and local efficiency
As T1DM often starts during childhood and adolescence
Considering a recent longitudinal observation suggesting minimal age-related brain volume loss in adolescents with T1DM
It is unclear why the structural connections subserving prefrontal controls on sensorimotor cortices were likely to be retained relative to those on other functions. Subclinical involvements of the retina or peripheral nerves in T1DM, which could occur even in the early stage of disease
As suggested by previous diffusion tensor imaging studies that could provide anatomical information on the connections between brain regions, white matter connectivity in a wide range of brain areas may be affected in T1DM patients relative to control subjects
T1DM may lead to several neurobehavioral abnormalities, as the CNS manifestations of T1DM include impaired performances on several cognitive domains as well as a wide range of emotional problems such as depression and anxiety
The current findings suggest that altered inter-network connectivity between these key regions may play an important role in T1DM-associated neurobehavioral abnormalities, in addition to T1DM-related volume or thickness reductions with a similar regional pattern. In contrast, network efficiencies between brain regions implicated in the earliest developed functions such as sensorimotor function were preserved in T1DM patients.
One might assume that the metabolic load including history of glycemic control might affect the extent of alterations of structural cortical network in T1DM patients as shown in dose-dependent effects of hyperglycemia on neurochemical disturbances in the prefrontal cortex of T1DM patients
Several limitations should be taken into account in interpreting the results. First, it remains to be determined whether the inter-regional correlation patterns of cortical thickness could reflect the alterations in intrinsic network systems for specific brain functions. However, growing evidence suggests that shared morphological organization may occur in functionally synchronous brain regions due to mutually trophic or neuroplastic influences
It should also be noted that anatomical connection matrices based on cortical thickness data that were used in the present study, were constructed by measuring inter-regional correlation coefficients over a group of subjects. In contrast to individual levels of connection matrices yielded from functional imaging data, we could not determine the network information at individual levels as well as the effects of inter-individual clinical variances on network organization. Furthermore, future studies using different cortical parcellation maps would be needed to replicate the current results based on Desikan-Killiany cortical atlas
Although evidence based on histological and neuroimaging studies has indicated that cortical thickness may reflect composite estimations of size, density, and arrangement of cortical neurons, glia, or nerve fibers
An Increasing body of literature has provided evidence that topological changes in brain networks may reflect disease progression or plastic changes in response to external stimuli
(DOC)
(DOC)