Resting-State Brain Networks in Type 1 Diabetic Patients
This is the first study to show that in T1DM patients, resting-state cerebral connectivity networks differ from those in healthy comparison subjects and that in patients, the direction of the effects vary according to complication status. The dual-regression analysis identified five networks that were affected in T1DM. MA patients showed decreased connectivity in all five networks. For MA patients versus control subjects, increased connectivity was found in the sensorimotor and secondary visual networks. In the other three networks (ventral attention, auditory and language-processing, and left frontoparietal networks), a significant dose-response relationship was evident, indicating that connectivity is lowest in MA patients, intermediate in MA patients, and highest in control subjects (Fig. 2). Changes in network integrity were associated not only with microangiopathy but also with age, sex, and BMI. Moreover, in at least one network (secondary visual), increased connectivity was associated with higher general cognitive ability scores and faster information-processing speed. The observed connectivity results in these patients are consistent with our earlier magnetoencephalography findings.
At first glance, increased connectivity seems counterintuitive, but it is also found in early multiple sclerosis and mild cognitive impairment. This phenomenon may be explained by loss of local inhibitory (inter)neurons that results in increased firing of long-distance neurons and that in turn leads to increased connectivity. This increase could also be a reaction to loss of connectivity in other networks that inhibit connectivity in visual and sensorimotor networks. Another explanation is that increased connectivity is a sign of functional reorganization, occurring in response to early subtle brain damage. Early cerebral compromise has been demonstrated in T1DM youth and adults with no or only minimal microvascular complications. It has been shown in multiple sclerosis that with increasing disease severity, connectivity decreases, possibly as a consequence of failing functional reorganization. This may also explain the decreased connectivity in our MA patients. Why increased connectivity in our MA patients is limited to the primary and secondary visual and sensorimotor networks is unknown (Figs. 2 and 4). It might be that increased connectivity can be longer retained in networks comprising brain regions situated close together. Alternately, these three networks receive direct neuronal input from retinal and peripheral nerve fibers. Already in MA patients, subclinical retinal and peripheral nerve changes have been found, which have been related to increased brain activity measured with electroencephalography.
(Enlarge Image)
Figure 4.
Bar graph with SDs of the mean connectivity value of each network resulting from the ICA. Networks left of the dotted line are composed of brain areas relatively close to each other and are thought to involve lower-order cognitive functions, whereas networks right of the dotted line represent networks with a more complex topography and involvement in higher-order cognitive functions. White bars, control subjects; gray bars, MA patients; black bars, MA patients.
We dichotomized patients with respect to microangiopathy presence as a proxy for overall glycemic burden. Our fMRI and cognitive results, however, suggest more of a dose-response relationship of glycemic burden. Accumulating hyperglycemia, in MA patients, may lead to increased connectivity but decreased cognitive performance. With the presence of clinical microangiopathy, i.e., increased glycemic burden, connectivity decreased and cognitive functions further decreased. We did not observe an effect of severe hypoglycemic events on our results. Although this may have been due to the difficulty in accurately determining these events, our results are in line with other studies in adults. In addition, neuropathy and nephropathy may contribute to the results. Given the study design, we were not able to tease out their relative contribution, which should be determined in future studies.
The connectivity results may be confounded by several factors. For factors like age, sex, BMI, IQ, blood pressure, depressive symptoms, and severe hypoglycemic events, groups were matched or we statistically corrected for those variables. Disease duration and onset age may also be important confounders. However, results remained similar when we analyzed groups matched for these variables. This may indicate that microangiopathy, as a surrogate marker of hyperglycemic burden, may be the most prominent diabetes-related variable associated with connectivity and cognition changes. Besides these demographic and medical factors, technical factors may also influence BOLD signal and thereby our results. First, BOLD signal is influenced by the cardiac and respiratory systems, cerebrospinal fluid, blood flow, and head movement. Given the strong temporal coherence of these systems, ICA analysis will identify these as independent components. Second, cerebral microvascular changes may lead to altered hemodynamic responses, as well as to changes in neurovascular coupling (i.e., process by which neuronal activity influences the surrounding vasculature), which may affect BOLD signal. Third, hypoperfusion can also lead to changes in BOLD signal. Perfusion has been mostly studied in relation to hypoglycemia and found to increase. One study found disturbed increased perfusion after a dilatory stimulus in T1DM versus control subjects, related to the presence of microangiopathy. Although it is difficult to exclude the influence of these factors, as we used ICA and given our magnetoencephalography data, which are less influenced by these technical factors, we may conclude that our results, at least in part, represent true connectivity changes.
Although one might expect to find a strong relationship between functional connectivity and cognition, there have been many negative findings, including results from the present set of analyses. The absence of an association between cognition and functional circuitry could be explained in several ways. On the one hand, it could be a limitation of the methods used. We correlated a single functional connectivity value (comprised of multiple values summed across multiple brain regions) with each cognitive domain (comprised of multiple cognitive test scores). A voxel-wise correlation analysis between specific cognitive test scores and the individual voxels of a particular network may be more sensitive and more strongly correlated to cognition. On the other hand, there is growing evidence to suggest that complex cognitive processes are mediated by multiple interacting neural networks and may not be limited to activity in the discrete networks identified by resting-state fMRI.
One potential limitation is the large number of statistical comparisons performed in the dual-regression analysis. Although all components are statistically independent, and therefore the sets of statistical tests are also independent, many statistical comparisons were performed, and there is a possibility that our results capitalized on chance associations. However, given the novelty of the analysis and the thought that cognition is mediated by multiple neuronal networks, we chose this approach.
In conclusion, we showed that functional connectivity of neuronal circuits is compromised in MA patients, whereas increased resting-state connectivity was observed to some extent in MA patients. This altered resting-state connectivity may, in part, contribute to the impairments in cognitive performance observed in T1DM patients, particularly in those with microangiopathy.
Discussion
This is the first study to show that in T1DM patients, resting-state cerebral connectivity networks differ from those in healthy comparison subjects and that in patients, the direction of the effects vary according to complication status. The dual-regression analysis identified five networks that were affected in T1DM. MA patients showed decreased connectivity in all five networks. For MA patients versus control subjects, increased connectivity was found in the sensorimotor and secondary visual networks. In the other three networks (ventral attention, auditory and language-processing, and left frontoparietal networks), a significant dose-response relationship was evident, indicating that connectivity is lowest in MA patients, intermediate in MA patients, and highest in control subjects (Fig. 2). Changes in network integrity were associated not only with microangiopathy but also with age, sex, and BMI. Moreover, in at least one network (secondary visual), increased connectivity was associated with higher general cognitive ability scores and faster information-processing speed. The observed connectivity results in these patients are consistent with our earlier magnetoencephalography findings.
At first glance, increased connectivity seems counterintuitive, but it is also found in early multiple sclerosis and mild cognitive impairment. This phenomenon may be explained by loss of local inhibitory (inter)neurons that results in increased firing of long-distance neurons and that in turn leads to increased connectivity. This increase could also be a reaction to loss of connectivity in other networks that inhibit connectivity in visual and sensorimotor networks. Another explanation is that increased connectivity is a sign of functional reorganization, occurring in response to early subtle brain damage. Early cerebral compromise has been demonstrated in T1DM youth and adults with no or only minimal microvascular complications. It has been shown in multiple sclerosis that with increasing disease severity, connectivity decreases, possibly as a consequence of failing functional reorganization. This may also explain the decreased connectivity in our MA patients. Why increased connectivity in our MA patients is limited to the primary and secondary visual and sensorimotor networks is unknown (Figs. 2 and 4). It might be that increased connectivity can be longer retained in networks comprising brain regions situated close together. Alternately, these three networks receive direct neuronal input from retinal and peripheral nerve fibers. Already in MA patients, subclinical retinal and peripheral nerve changes have been found, which have been related to increased brain activity measured with electroencephalography.
(Enlarge Image)
Figure 4.
Bar graph with SDs of the mean connectivity value of each network resulting from the ICA. Networks left of the dotted line are composed of brain areas relatively close to each other and are thought to involve lower-order cognitive functions, whereas networks right of the dotted line represent networks with a more complex topography and involvement in higher-order cognitive functions. White bars, control subjects; gray bars, MA patients; black bars, MA patients.
We dichotomized patients with respect to microangiopathy presence as a proxy for overall glycemic burden. Our fMRI and cognitive results, however, suggest more of a dose-response relationship of glycemic burden. Accumulating hyperglycemia, in MA patients, may lead to increased connectivity but decreased cognitive performance. With the presence of clinical microangiopathy, i.e., increased glycemic burden, connectivity decreased and cognitive functions further decreased. We did not observe an effect of severe hypoglycemic events on our results. Although this may have been due to the difficulty in accurately determining these events, our results are in line with other studies in adults. In addition, neuropathy and nephropathy may contribute to the results. Given the study design, we were not able to tease out their relative contribution, which should be determined in future studies.
The connectivity results may be confounded by several factors. For factors like age, sex, BMI, IQ, blood pressure, depressive symptoms, and severe hypoglycemic events, groups were matched or we statistically corrected for those variables. Disease duration and onset age may also be important confounders. However, results remained similar when we analyzed groups matched for these variables. This may indicate that microangiopathy, as a surrogate marker of hyperglycemic burden, may be the most prominent diabetes-related variable associated with connectivity and cognition changes. Besides these demographic and medical factors, technical factors may also influence BOLD signal and thereby our results. First, BOLD signal is influenced by the cardiac and respiratory systems, cerebrospinal fluid, blood flow, and head movement. Given the strong temporal coherence of these systems, ICA analysis will identify these as independent components. Second, cerebral microvascular changes may lead to altered hemodynamic responses, as well as to changes in neurovascular coupling (i.e., process by which neuronal activity influences the surrounding vasculature), which may affect BOLD signal. Third, hypoperfusion can also lead to changes in BOLD signal. Perfusion has been mostly studied in relation to hypoglycemia and found to increase. One study found disturbed increased perfusion after a dilatory stimulus in T1DM versus control subjects, related to the presence of microangiopathy. Although it is difficult to exclude the influence of these factors, as we used ICA and given our magnetoencephalography data, which are less influenced by these technical factors, we may conclude that our results, at least in part, represent true connectivity changes.
Although one might expect to find a strong relationship between functional connectivity and cognition, there have been many negative findings, including results from the present set of analyses. The absence of an association between cognition and functional circuitry could be explained in several ways. On the one hand, it could be a limitation of the methods used. We correlated a single functional connectivity value (comprised of multiple values summed across multiple brain regions) with each cognitive domain (comprised of multiple cognitive test scores). A voxel-wise correlation analysis between specific cognitive test scores and the individual voxels of a particular network may be more sensitive and more strongly correlated to cognition. On the other hand, there is growing evidence to suggest that complex cognitive processes are mediated by multiple interacting neural networks and may not be limited to activity in the discrete networks identified by resting-state fMRI.
One potential limitation is the large number of statistical comparisons performed in the dual-regression analysis. Although all components are statistically independent, and therefore the sets of statistical tests are also independent, many statistical comparisons were performed, and there is a possibility that our results capitalized on chance associations. However, given the novelty of the analysis and the thought that cognition is mediated by multiple neuronal networks, we chose this approach.
In conclusion, we showed that functional connectivity of neuronal circuits is compromised in MA patients, whereas increased resting-state connectivity was observed to some extent in MA patients. This altered resting-state connectivity may, in part, contribute to the impairments in cognitive performance observed in T1DM patients, particularly in those with microangiopathy.
SHARE
