Brain alterations in regions associated with end‐organ diabetic microvascular disease in diabetes mellitus: A UK Biobank study

Diabetes mellitus (DM) is associated with structural grey matter alterations in the brain, including changes in the somatosensory and pain processing regions seen in association with diabetic peripheral neuropathy. In this case‐controlled biobank study, we aimed to ascertain differences in grey and white matter anatomy in people with DM compared with non‐diabetic controls (NDC).


| INTRODUCTION
The global prevalence of diabetes is estimated to increase by 46% from ~537 million people in 2021 to >783 million people by 2045. 1 The economic cost of diabetes increased by 26% from 2012 to 2017, primarily due to the growing prevalence of complications in an ageing population. 2 Long term microvascular complications of diabetes include kidney disease, retinopathy, and peripheral neuropathy. 3The central nervous system has also demonstrated end-organ damage. 4,5cently published longitudinal data have shown accelerated cognitive decline in a large cohort of deeply phenotyped participants with diabetes. 6Cognitive deficits in psychomotor and mental efficiency domains equivalent to an additional 9•4 years of ageing are present in late-middle age people with diabetes. 6Indeed, there is strong evidence of the association of cognitive decrements with diabetes-related cerebral small vessel disease and cerebral atrophy. 7Longitudinal studies have also demonstrated global grey matter reductions accompanied by the expansion of the lateral ventricles synonymous with a neurodegenerative process. 8In both type 1 (T1D) and type 2 (T2D) diabetes, regional reductions in volume have been identified in cortical brain regions of the frontal, posterior and temporal lobes, and also sub-cortical structures. 9e microvascular complications of diabetes are associated with structural alterations in the central nervous system, notably affecting the functional connectivity of key brain networks.These include the default-mode network, a group of brain regions active during selfreferential thoughts and internal monologue; the visual network, responsible for processing visual information; and the salience network, which is implicated in the integration of sensory and emotional stimuli as well as internal and externally directed cognition. 10The presence and severity of retinopathy in people with T2D is associated with decreased grey matter volume and a higher incidence of white matter abnormalities. 11Sink and colleagues 12 demonstrated a relationship of T2D with diabetic kidney disease to cerebral atrophic changes and a greater white matter lesion volume.
Further, volumetric grey matter reductions occur in regions associated with sensorimotor function in diabetic peripheral neuropathy. 13evious data have demonstrated significant anatomic and functional changes in the somatosensory cortex, including reductions in cortical thickness and expansion of the homuncular area of the lower limb to include regions which represent the face and lip. 4 These alterations in regions associated with sensory and motor functions have implications for the understanding of the natural history of microvascular complications.
Over the last decade, studies using diffusion magnetic resonance imaging (MRI) in people with diabetes have shown decreased white matter integrity and network disorganisation in white matter pathways. 14Alterations in fractional anisotropy (FA), a scalar metric of diffusion tensor imaging (DTI), indicate microstructural abnormalities in people with diabetes, frequently found in the frontal, temporal, and parietal lobes, corpus callosum, cingulum, uncinate fasciculus, corona radiata, and internal and external capsules. 15These alterations are particularly prominent in white matter tracts associated with sensorimotor functions in people with diabetic peripheral neuropathy. 13To our knowledge a large community-based cohort of people with diabetes has not yet been analysed for differences in grey matter volume and white matter tracts.This study aims to investigate grey matter volumetric and microstructural white matter alterations in a large prospective community-based cohort of comprehensively phenotyped people with diabetes, in comparison to a control cohort without diabetes.

| Study design
Between 2006 and 2010, 502,490 participants aged between 40 and 69 years, from a variety of settings, attended one of 22 UK assessment centres. 16The initial visit comprised of self-reported outcomes from touch-screen questionnaires, physical measures, and blood tests.Of the 502,490 participants who attended the baseline visit, 221,320 were invited to take part in a follow-up imaging visit.
Excluding those who did not respond to the invitation and those contraindicated for MRI, 48,712 participants attended an imaging visit at the time of data release. 17The UK Biobank is a national epidemiological resource, with the protocol available here: https:// www.ukbiobank.ac.uk/media/gnkeyh2q/study-rationale.pdf.This biobank study obtained data release from all 502,490 UK Biobank participants.Within this dataset, we identified 26,402 participants with ICD-10 codes for Diabetes mellitus (DM) (Figure 1).
Participants with ICD-10 codes for DM must also have answered "Yes" to having had diabetes diagnosed by a doctor.Before exclusions, there were 19,227 (15,748 type two and 3479 type one) eligible participants with diabetes.Five participants with diabetes who withdrew their consent were removed before matching.Control participants without diabetes were selected from the study population if they had no ICD-10 codes associated with their unique electronic identifier and had answered "No" to diabetes having been diagnosed by a doctor (n = 89,448).Optmatch 18 was used to identify unique participant identifiers of non-diabetic controls that were most similar (partial matching) using the default parameters of the 'pairmatch' argument to individuals with diabetes matched by age and sex at baseline at a 1:1 ratio, giving a combined total of 38,444 participants.
Unique participant identifiers were tracked against ICD-10 codes, and participants with ineligible diagnoses were excluded (Table S1).The ICD-10 codes for participants with T1D and T2D, which encode diabetes associated microvascular complications (e.g., "E102", "E103", "E112"), were tabulated for all participants at baseline and at follow-up.The ICD-10 codes pertaining to diabetesassociated retinopathy, nephropathy, and hypoglycaemia were consolidated to calculate the overall frequency and percentage of each condition.There were subsequently 601 participants with diabetes with MRI-derived grey matter volume data.Control participants without diabetes with complete MRI-derived grey matter data were matched using Optmatch by age and sex at a ratio of three-toone (n = 1803).The ICD-10 codes of the included participants were listed by frequency at the imaging visit (Table S2).Participants pro-

| Cognitive tests
Data from the following cognitive tests were extracted: Trail-Making, Symbol Digit Substitution Pair Matching, Fluid Intelligence and Numeric Memory.A summary of each test is detailed in the supplementary appendix.The baseline cognitive data were collected using an unsupervised touchscreen interface.

| Pain measures
Participants remotely completed a questionnaire on the location, nature and impact of their pain.We utilised replies to questions about 'General pain in the last 3 months' and 'Pain types experienced in the last month'.The pain questionnaire is available here: https:// biobank.ndph.ox.ac.uk/showcase/ukb/docs/pain_questionnaire.pdf.

| Brain imaging and analysis
All brain MRI data from 2015 onwards were acquired using a Siemens Skyra 3 T scanner with a Siemens 32-channel head coil.

BURGESS ET AL.
Volumetric whole brain and regional MRI data were generated by the UK Biobank Oxford Biomedical Research Computing team.For cortical and subcortical grey matter volume analysis, Freesurfer software (version 6.0; https://surfer.nmr.mgh.harvard.edu/)was used, utilising both the ASEG Atlas and a2009s Destrieux Atlas for detailed brain parcellations.Automated white matter tract segmentation was performed using TractSeg 19 (https://github.com/MIC-DKFZ/TractSeg), which segments the original fibre orientation distribution function peaks into tract orientation maps. 20These maps allow for 'along-tract' comparisons, assessing FA at 100 equidistant points within each segmented white matter tract.This enabled a detailed analysis of FA variations along each tract, which contrasts with other methods such as tract-based spatial statistics (TBSS), as it facilitates the localised examination of FA along tracts.
Visual representations of diffusion models for each tract were generated to highlight significant group differences.Further details on brain imaging acquisition and analysis are available in the supplementary appendix.Continuous variables within the dataset are summarised as group mean and standard deviation.Categorical variables are presented as frequencies and percentage.Prior to statistical analysis, continuous volumetric variables were assessed for normality by relevant plots and statistical tests.For the volumetric analysis, separate individual multiple linear regression models were fitted for each brain region with age, sex, and intra-cranial volume as co-variates to identify regions of interest which were significantly different between participants with diabetes and non-diabetic controls.In this study, we opted for multiple linear regression over the more typical analysis of covariance to directly discern the direction and magnitude of volumetric changes in specific brain regions.The threshold of significance for each individual test was calculated as 0•05 divided by the number of included brain regions (n = 188).Therefore, a threshold of significance was applied of <0•00026 after Bonferroni correction.A less conservative approach was also applied using a threshold of significance of <0•05 after false discovery rate correction (equivalent to the Benjamini-Hochberg procedure 21 ) to mitigate the likelihood of type one error.The mean percentage difference was calculated, for every region, by subtracting the uncorrected mean brain volume between the two groups and dividing by the first group's mean value (people with diabetes) and expressing this value as a percentage.
Tractseg 19 was used for tractometry analysis of DTI data.The alpha family wise error measurement represents the p-threshold after accounting for multiple along-tract comparisons with age and sex as covariates.

| Baseline demographic, anthropometric and clinical measures
Of the 502,490 participants who took part in the UK Biobank project, 38,444 participants were included in the group summary of cognitive function and pain measures as detailed in Figure 1.The total number of complete data entries for each demographic and metabolic variable at baseline is shown in table S3.The demographic and metabolic measures are summarised in Table 1.Similar summary values in age, sex, or blood pressure were observed between the two groups.
Participants with diabetes had a greater waist circumference and increased body mass index (BMI), triglycerides, glucose, and glycated haemoglobin (HbA1c) but lower high-density lipoprotein (HDL) and low-density lipoprotein (LDL).In the diabetic group, the mean (� standard deviation) duration of known diabetes was 11 � 14 years at baseline.Participants with DM at baseline were associated with ICD-10 codes for renal and ophthalmic complications in 2.6% and 13.2% of cases, respectively, and episodes of hypoglycaemia were recorded in 6%.The demographic composition differed between the groups; there were fewer White individuals in the DM group compared to the non-diabetic controls (NDC) group, and higher proportions of Asian T A B L E 1 Baseline anthropometric, clinical and demographic measures of non-diabetic control participants with people with diabetes.or Asian British, Black, Black British, Caribbean, or African, Mixed or Multiple Ethnic Groups, and Other Ethnic Groups in the DM group.

| Baseline visit cognitive measures
There were differences between the cognitive scores of participants with diabetes and controls without diabetes, as seen in Table 2.
Participants with diabetes took relatively longer on both the numeric and alphanumeric trail exercises, although there were no differences seen in the number of errors made.Furthermore, the symbol digit, numeric memory and fluid intelligence scores were marginally lower in participants with diabetes relative to controls.No differences were likely in the pairs matching exercise in either time to complete the round or the number of incorrect matches.The total number of complete data entries for each of the cognitive measure is shown table S4.

| Baseline visit pain measures
There was a greater proportion of participants with diabetes who reported general pain in the last 3 months, as shown in

| Imaging visit cognitive measures and pain measures
No differences were seen in any of the cognitive measures, as shown in Table 4.There tended to be a greater proportion of participants with diabetes who reported pain in the last month since attending the research visit, particularly due to headache, and locations such as the back, stomach, hip, and knee in the last month.Furthermore, a greater proportion of participants with diabetes reported pain all over the body in the last month relative to non-diabetic controls.
There were no differences found in either cognitive or pain measures between T1D and T2D, as shown in Tables S9 and S10.

| Grey matter volumetric analysis
There was a reduction in the whole brain, subcortical grey matter, and total grey matter volume in people with diabetes (all p < 0•0002) relative to controls, as seen in Table 5. Figure 2 illustrates an increase in the volume of the third, fourth and lateral ventricles bilaterally using the subcortical segmentation analysis (all p < 0•0002).diabetic controls (all p < 0•0002).Decreased volume was found in regions associated with the visual, sensorimotor and default mode networks (all p < 0•0002) (Figure 2, Figure 3 (left)).When p-value correction was adjusted using false discovery rate correction, additional cortical and sub-regions were identified as statistically different between participants with diabetes relative to non-diabetic controls (Table S11).There were no significant differences in grey matter volume between participants with T1D and T2D.

| Mean percentage difference
There was a negative mean volume percentage difference in the whole

| Tractometry using fractional anisotropy
There were significant reductions in FA along the length of the tha-  sensorimotor functions in people with diabetes are shown in Figure 3 (right).In addition, reduced FA was observed in white matter tracts associated with co-ordinating movements and visual information synthesis, such as the middle cerebellar peduncle, right superior cerebellar peduncle, parieto-occipital pontine, right fronto-pontine tract and bilateral inferior occipitofrontal fascicle.

| DISCUSSION
This is the largest volumetric and tractometric study to examine structural grey and white matter changes in a prospective cohort of people with diabetes.Volumetric analysis showed lower grey matter volume in participants with adequately controlled glycaemia and a  low prevalence of microvascular complications compared to nondiabetic controls.Moreover, both volumetric grey matter and microstructural white matter changes were found in regions associated with the visual and sensorimotor networks.A pathophysiological relationship appears to exist between regions involved in sensory processing and diabetic peripheral neuropathy severity, 22 neuropathic pain 4 and worsening visual impairment in diabetic retinopathy. 23evious studies reported structural alterations in the central nervous system in T1D 24 and T2D. 25 26 This was demonstrated in a large cohort of deeply phenotyped participants with diabetes, where mild cognitive impairment prevalence was comparable to age-adjusted rates of the general population despite accelerated incipient cognitive decline. 6 the current study, participants with diabetes reported more general and widespread pain, although the neuropathic pain subtype which occurs in diabetic peripheral neuropathy is not routinely captured in the self-reported outcomes.Interestingly, regional reductions in brain volume associated with chronic pain are seen in diabetes, such as the insula, sensorimotor areas, the cerebellum and deep basal nuclei. 13Our analysis found reduced brain volume in regions associated with sensorimotor processing.However, these observations (despite exhibiting excess pain) occurred despite a lack of a formal diagnosis of diabetic peripheral neuropathy or any detailed pain phenotyping.Indeed, if we were to advance the hypothesis of a neurodegenerative process perhaps driven by diabetes-related endorgan damage, the symptom burden of diabetic peripheral neuropathy could be potentiated by these early changes in the central nervous system.
In diabetic peripheral neuropathy, significant peripheral grey matter decrements are reported, affecting the sensorimotor cortex, somatosensory area, and supplementary sensorimotor area. 4,13,22,27dataset and meta-analysis using functional MRI identified significant reductions in brain activity areas associated with sensorimotor and visual functions in people with T2D. 30 Furthermore, differences are reported in both volume and vascularity of the thalamus in people with painless compared to painful diabetic peripheral neuropathy. 31I has potential as a clinical biomarker in diabetes and diabetic peripheral neuropathy.Furthermore, multimodal MRI studies have suggested the potential underlying pathogenic mechanisms of diabetic peripheral neuropathy and neuropathic pain. 4ltiplexed MRI studies, which include DTI measures such as FA, predict white matter hyper-intensities in cerebral small vessel disease prior to their occurrence. 32Indeed, white matter microstructural alterations in sensorimotor tracts are associated with worsening motor performance. 33DTI measures such as FA are a measure of the integrity of white matter tracts; a reduction in FA can broadly indicate microstructural disruption, but they have also been used to infer neuronal loss and/or tract demyelination. 34cently published data from Muthulingam and colleagues 35 report reduced FA of the thalamic radiations, longitudinal fasciculi, internal capsule and corticospinal tract in participants with T1D with diabetic peripheral neuropathy and a high prevalence of retinopathy.
The aforementioned microstructural alterations are positively associated with peripheral nerve function (sensory nerve conduction amplitude) and diabetes duration.Indeed, widespread microstructural alterations in white matter tracts that occur in the absence of clinically defined microvascular complications are markedly worse in the presence of diabetic angiopathy. 35,36Similarly, we have identified significant differences in FA in tracts with associated visual and sensorimotor functions, including the corticospinal tract, corpus callosum, thalamo-premotor tract and thalamic radiations.These white matter tract alterations shown in the present study are in agreement with a recently published systematic review, albeit with the majority of included studies representing cohorts with microvascular complications. 34e presence of microvascular complications are associated with microstructural abnormalities in white matter tracts with associated sensorimotor, visual and cognitive functions. 34Indeed, significant global alterations in subcortical white matter are demonstrated in diabetic peripheral neuropathy, with alterations in sensory ascending tracts such as the corticospinal and spinothalamic tract, and thalamocortical projecting fibres. 13Motor impairment has been associated with regional reductions in primary motor areas in diabetic peripheral neuropathy. 37Additionally, recently published data indicate marked disruptions of functional connectivity of the thalamus, and increased connectivity of the default mode network in T1D with painful diabetic peripheral neuropathy. 38 the present study, univariate analyses were employed to identify group differences; however, this method does not consider the possible multicollinearity of adjacent brain regions.Furthermore, it is possible that intra-cranial volume may have served better as part of the matching process as opposed to being used as an adjustment within our univariate regression models.The volumetric analysis had many comparisons; however, efforts were made to mitigate false positives using co-variates for adjustment, strict significance thresholds and p-value corrections.Moreover, "healthy volunteer" selection bias is evidenced within the UK Biobank dataset 39 and should be considered when interpreting these data.A key strength of this analysis is that it is based on a large prospective populationbased study with well phenotyped participants, with many participants also undergoing with brain MRI.Furthermore, the use of the novel TractSeg tool for tractometry may provide a more sensitive means of identifying microstructural group differences in white matter compared to TBSS methodology.We acknowledge that the UK Biobank is not a dedicated resource for diabetes research and
vided full informed consent to participate in the UK Biobank.The UK Biobank received a favourable ethical opinion 17 June 2011 from the National Health Service Research Ethics Service (Ref 11/NW/0382).

F I G U R E 1
Flow chart showing the process of participant selection and inclusion for each step of participant selection, matching by age and sex, and participant numbers within each analysis.Created with BioRender.com.All data analyses were conducted by R software (R Foundation, R version 4•0•2).Descriptive analysis is provided for the baseline and imaging visit demographic, clinical, cognitive and pain variables.
lamocortical radiations, thalamostriatal projections and sensory ascending tracts including the corticospinal tract and commissural somatosensory associated tracts of the corpus callosum in participants with diabetes relative to non-diabetic controls (Figure 4; all p < 0•001).The tracts with reduced FA and typical connections to areas with decreased grey matter regions associated with T A B L E 4 Cognitive function and pain outcomes comparing non-diabetic control participants with people with diabetes who attended the imaging visit.Time to complete round (dS)

F I G U R E 2
Cortical and subcortical regions.The t value statistic indicates either an increase or decrease in volume compared to healthy volunteers, where red indicates an increase in volume, whilst blue indicates a decrease.Sub-cortical regions: An increase in volume is shown in the ventricles.A decrease in volume is seen bilaterally in the ventral diencephalon, basal ganglia, brain stem, cerebellar cortex and central corpus callosum.Cortical Regions: A decrease in volume is shown in regions associated with the visual, sensorimotor and default mode networks.The areas highlighted in green are implicated in sensorimotor function and pain processing.Created with BioRender.com.BURGESS ET AL.

F I G U R E 3
The cortical and sub-cortical regions of grey matter with decreased volume (shown in blue) associated with sensorimotor functions and the white matter streamlines with reduced fractional anisotropy (FA) (shown in red) which typically connect them.ATR-Left, Left anterior thalamic radiation; CC5, Corpus callosum (Primary Somatosensory); CC6, Corpus callosum (isthmus); CST-Right, Right corticospinal tract; STR-Left, Left superior thalamic radiation; STR-Right, Right superior thalamic radiation; TPAR-Right, Right thalamo-parietal tract; TPREM-Right, Right thalamo-premotor tract.Created with BioRender.com.
diabetic complications, and as such ICD-10 coding was used as a proxy for concomitant complications.Indeed, further comprehensive observational longitudinal studies are required in people with new or early onset diabetes with microvascular risk factor phenotyping.The primary objective of these studies should be to characterise the contribution of changes in the central and peripheral nervous systems in the presence of microvascular complications and cardiovascular risk factors towards the development of cognitive dysfunction or sensory deficits.6| CONCLUSIONThis analysis demonstrates volumetric and microstructural white matter differences between participants with DM and individuals without diabetes in the central nervous system.Differences are present in areas associated with sensorimotor and visual functions in a cohort with DM, good metabolic control, and minimal microvascular complications.Structural alterations and neural remodelling of the central nervous system are likely to be involved early in the development and maintenance of incipient diabetes-related microvascular complications.

Table 2
Baseline cognitive function and pain outcomes comparing non-diabetic control participants with people with diabetes.
reported pain all over the body in the last month relative to nondiabetic controls.The total number of complete data entries of the pain questionnaire is shown in tableS5.T A B L E 2 a Mean (SD); n (%).Abbreviations: dS, decisecond; n, number; S, second.4.1.4|Demographic,anthropometricandclinical measures in the magnetic resonance imaging subgroup Of the included 19,222 participants with DM, 601 (T1D = 94; T2D = 507) had complete volumetric data from structural MRI scans, with Freesurfer segmentations as shown in Figure 1.Volumetric data from non-diabetic controls were matched at a ratio of 3:1 to participants with DM and complete volumetric data (n = 1803).The total number of complete data entries for each demographic and metabolic variable at the imaging visit is shown in TableS6.The cumulative demographic and metabolic measures of the participants who completed the MRI visit are summarised in Table3.As expected by design, there was no difference in age or sex between the two groups.The mean (� standard deviation) duration of known diabetes was 18 � 11 years at the imaging visit.Compared to the control group, participants with diabetes had differences in BMI, waist circumference, triglycerides, cholesterol, HDL, LDL, glucose, Cophthalmic complications in 0.2% and 5.8% of cases, respectively, and episodes of hypoglycaemia were recorded in 2.3%.The demographic compositions differed; the NDC group had a higher proportion of White individuals compared with the DM group.The DM group had higher proportions of Asian or Asian British, Black, Black British, Caribbean or African, and Other Ethnic Groups, with both groups having an equal representation of Mixed or Multiple Ethnic Groups.
People with diabetes versus non-diabetic controls: Significant global, cortical and sub-cortical regions after bonferroni correction for multiple comparisons.