Structural Brain Alterations in Key Somatosensory and Nociceptive Regions in Diabetic Peripheral Neuropathy

Diabetes peripheral neuropathy (DPN) has been considered a disease of the peripheral nervous system only; however, a substantial body of research has now demonstrated alterations within the central nervous system (CNS) (1–9). These alterations are most pronounced in key somatosensory and nociceptive processing cortical regions (10). Alterations in the spinal cord were first demonstrated in postmortem and somatosensory evoked potential studies in patients with advanced DPN (11–13). Subsequent investigation into the structural brain alterations found reduced peripheral gray matter volume in DPN (4). More recent studies have localized this volume loss to the primary somatosensory cortex (S1), supramarginal gyrus, cingulate cortex, insular cortex, and primary motor cortex (M1) (4,8,9). However, the key unanswered question is whether there are differences in brain structure between painful and painless DPN or indeed between the different clinical phenotypes of painful DPN (7). This question is important because delineating the CNS correlates of DPN may afford us a better understanding of the disease mechanisms and enable the development of new biomarkers and potential targets for treatment. Previous studies are limited by relatively small sample sizes and the absence of detailed phenotyping of painful DPN, precluding a robust mechanistic exploration of cortical changes in DPN. The aim of this study, which to our knowledge is the largest neuroimaging study in DPN to date, was to 1) define the cerebral morphological alterations in carefully phenotyped patients with painful and painless DPN and 2) explore the relationships between brain morphology and clinical/neurophysiological measures of DPN.

We performed a cross-sectional, observational, case-control cohort study in 283 right-handed individuals (217 with diabetes and 66 healthy volunteers [HVs]) recruited from outpatient diabetes clinics at the Royal Hallamshire Hospital (Sheffield, U.K.) between 2009 and 2019. Inclusion criteria were right-handedness, age 18–85 years, and type 1 or type 2 diabetes diagnosed >6 months previous. Exclusion criteria were pregnancy, insufficient command of the English language or mental capacity to provide informed consent, concurrent severe psychological/psychiatric conditions, moderate to severe pain from causes other than DPN, nondiabetic neuropathies (e.g., thyroid disease; vitamin B12 or folate deficiencies; drug-induced or toxic neuropathy; inflammatory, autoimmune, or genetic neuropathy), other diabetic neuropathies (e.g., lumbosacral plexopathy, mononeuropathies), history of alcohol consumption >20 units/week (1 unit equivalent to one glass of wine or one measure of spirits), recurrent severe hypoglycemia, neurological disorders that may confound radiological or clinical assessments (e.g., cerebrovascular disease, epilepsy, dementia, multiple sclerosis), or contraindications to MRI (e.g., pacemaker, claustrophobia). Per protocol, cross-sectional analyses included all participants with complete outcome data (six participants excluded because of claustrophobia, inability to fit in the scanner, and unusable MRI scans). All participants gave written informed consent before participating in the study, which had prior ethics approval by the National Health Service Health Research Authority (Sheffield, U.K.) review board.

Study group assignment was based on a detailed clinical history, physical examination, neurophysiological assessment, and blood testing. Nerve conduction studies were performed at a stable skin temperature of 31°C and a room temperature of 24°C using a Medelec electrophysiological system (Synergy Oxford Instruments, Oxford, U.K.). The following nerve attributes were measured: 1) sural sensory nerve action potentials and conduction velocities; 2) common peroneal distal latency, compound muscle action potential, and conduction velocity; and 3) tibial motor nerve distal latency. The American Academy of Neurology minimum case definition criterion to confirm the presence of DPN was used, i.e., an abnormality (>99th or <1st percentile) of any attribute of nerve conduction in two separate nerves, one of which must be the sural nerve (14). Painful DPN was defined according to the International Association for the Study of Pain definition of neuropathic pain (15) and a Douleur Neuropathique 4 (DN4) score ≥4 (16).

Participants with diabetes were divided into three groups: 1) no DPN (patients without neuropathic symptoms and with normal clinical and neurophysiological assessments), 2) painless DPN (patients with painless neuropathy, abnormal clinical findings, and at least two nerve conduction abnormalities), and 3) painful DPN (patients with chronic painful neuropathic symptoms for at least 6 months with clinical and nerve conduction abnormalities). Patients with <6 months of pain duration or who did not have abnormal clinical/nerve conduction abnormalities were excluded from the study.

Quantitative sensory testing (QST) was performed using the German Research Network on Neuropathic Pain (DFNS) protocol (17). QST consists of 1) cold and warm detection threshold and thermal sensory limen; 2) thermal pain thresholds for cold and hot stimuli; 3) mechanical detection thresholds for touch and vibration; and 4) mechanical pain sensitivity, including thresholds for pinprick and blunt pressure, stimulus/response functions for pinprick sensitivity and dynamic mechanical allodynia, and pain summation to repetitive pinprick stimuli (wind-up–like pain).

On the basis of QST findings, participants with painful DPN were divided into two subgroups using methods described previously (17). The irritable (IR) nociceptor phenotype includes the presence of either dynamic mechanical allodynia, reduced mechanical or pressure threshold, increased mechanical pain sensitivity, lower cold or heat pain threshold, or any combination of these signs of hyperexcitability. The nonirritable (NIR) nociceptor phenotype includes sensory loss with no signs of hyperexcitability. Supplementary Fig. 1 presents the Strengthening the Reporting of Observational Studies in Epidemiology diagram of participant flow.

Magnetic resonance brain scan was performed using a three-dimensional T1-weighted magnetization-prepared rapid echo sequence (further details provided in the Supplementary Material). Analysis of the MRI data sets was conducted while blinded to group allocation.

Cortical thickness and global and deep brain nuclei quantification were performed using FreeSurfer software (https://surfer.nmr.mgh.harvard.edu). Seven regions of interest (ROIs) were chosen in well-recognized brain regions involved with somatomotor function: combined (right and left) S1, M1, insular cortex, anterior cingulate gyrus, and thalamus, with the lateral occipital cortex chosen as a control region. ROI volumetric data from each hemisphere were combined prior to analysis being performed.

Voxel-based morphometric (VBM) analysis was performed using the FMRIB Software Library (FSL) version 6.0.30. The Harvard-Oxford Cortical and Subcortical Structural Atlases were used to create the following masks for our ROIs: anterior cingulate cortex (ACC), insular cortex, postcentral gyrus, precentral gyrus, thalamus left, and thalamus right, with an intensity threshold of 25 at 2-mm resolution. The ROI masks were used for group comparisons using general linear modeling applied in Randomize (18). Our group comparison investigated differences between the painful DPN group versus the HV, painless DPN, and no DPN groups and the painless DPN group versus the HV, painful DPN, and no DPN groups. Voxelwise general linear modeling was then applied using permutation-based (5,000 permutations) nonparametric testing. Clusters of significance were identified using the threshold-free cluster enhancement method (19), taking family-wise error rate–corrected P < 0.05. The family-wise error multiple comparison correction is based on Bonferroni method and controls the likelihood of false positive findings in analysis. Probabilistic anatomical descriptors were determined using FSL atlasquery avoiding labeling bias, and cluster peak information was extracted using the FSL cluster tool for voxels that had survived multiple comparison correction. Further details of brain morphological quantification and analyses can be found in the Supplementary Material.

Continuous baseline characteristics are described as mean (SD) unless otherwise indicated. Group differences on demographic characteristics and psychophysical experiments were compared using one-way ANOVA (normally distributed), χ² test (categorical variables), and Kruskal-Wallis rank test (nonnormally distributed) as appropriate. Global measures of segmented volume (e.g., subcortical gray matter volume, cortical white matter volume) were analyzed using ANCOVA, with age as a covariate and group as a fixed factor. A linear mixed model with restricted maximum likelihood estimation and an interaction term, group-by-ROI differences, with controls for mean ROI volumetric data were used, followed by formal post hoc analyses when a significant interaction was present. Inferential analyses were adjusted for age and sex unless otherwise stated. Student t test was used for pairwise group analysis. The relationship between ROIs and neurological/neurophysiological assessments (e.g., nerve conductions studies, DN4 and numeric rating scale [NRS] for pain) were analyzed in more detail using Pearson correlation coefficients. Based on our preliminary study (4), a sample size of at least 269 will provide 0.80 power (α = 0.05) to determine differences via ANOVA in postcentral cortical thickness among the four study groups with two covariates for a 0.25 (medium) effect size. All statistical analyses were completed using SPSS version 24 software.

The data sets generated and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Table 1 shows the demographic, clinical, and neurophysiological assessments for all groups. There were no significant post hoc group differences apart from participants with painless DPN being older than HVs (ANOVA P = 0.03; post hoc P = 0.003; 95% CI 1.74, 8.55), and there were fewer females in the painful DPN group compared with the HV (χ²P = 0.04) and no DPN (P < 0.01) groups. Participants with painless DPN had a longer duration of diabetes compared with the no DPN (ANOVA P = 0.001; post hoc P < 0.001; 95% CI 3.85, 12.99) and painful DPN (P < 0.01; 95% CI 2.01, 10.44) groups. BMI was higher in the painful DPN group compared with the HV group (ANOVA P = 0.008; post hoc P = 0.001; 95% CI 1.58, 5.74). There were no significant group differences in estimated glomerular filtration rate or retinopathy status; however, there was a greater percentage of participants with proliferative diabetic retinopathy in the painless and painful DPN groups compared with the no DPN group. Peroneal motor nerve conduction velocity and sural nerve action potential were both lower in the no DPN (ANOVA P < 0.001; post hoc P = 0.02; 95% CI −4.05, −0.35) compared with the HV group (ANOVA P = 0.002; 95% CI 1.35, 6.22). In addition, tibial motor nerve distal latency was prolonged in participants with painful DPN compared with painless DPN (post hoc P = 0.01; 95% CI 0.28, 1.91). The DN4 score and NRS were significantly higher in the painful DPN group compared with all other groups (all ANOVA P < 0.001).

Global measures of segmented brain volumes were significantly lower in participants with diabetes compared with HVs (total gray volume mean [SD]: HV 571.3 [61.8] vs. diabetes 549.5 [61.2] mL [t test P = 0.012; 95% CI 4.80, 38.8]; subcortical gray volume: HV 54.7 [5.4] vs. diabetes 52.6 [5.4] mL [P = 0.005; 95% CI 0.68, 3.69]; cortical volume: HV 420.0 [47.8] vs. diabetes 404.1 [47.3] mL [P = 0.018; 95% CI 2.69, 29.02]; total brain volume: HV 1.06 [0.11] vs. diabetes 1.03 [0.12] mL [P = not significant]; white matter volume: HV 459.8 [52.5] vs. diabetes 449.5 [63.8] mL [P = not significant]). Apart from a marginal reduction in subcortical gray matter volume in participants with painless DPN (ANOVA P = 0.04), there were no statistically significant subgroup differences in global measures of segmented brain volumes among study cohorts (Table 2).

We tested whether groups differed in postcentral cortical thickness versus control region by including a group-by-region interaction term in a linear mixed model adjusted for age and sex, with a significant result (P = 0.02; F[3,275] = 3.36) (Table 2 and Fig. 1). Post hoc comparisons confirmed that age- and sex-adjusted bilateral postcentral cortical thicknesses (mm) were significantly lower in the painless and painful DPN groups compared with the HV group and between the painless DPN and no DPN groups (Table 2 and Fig. 1). There were no significant differences between either the painless and painful DPN groups (P = 0.17; 95% CI −0.01, 0.05) or the no DPN and HV groups (P = 0.46; 95% CI −0.02, 0.05). Similarly, precentral cortical thickness (mm) (P = 0.01; F[3,275] = 3.80) (Fig. 1) and insula cortical thickness (mm) (P = 0.04; F[3,275] = 2.72) (Table 2) were also significantly lower in the painful and painless DPN groups compared with the HV and no DPN groups. There were no significant differences in thalamic volumes (P = 0.53; F[3,275] = 0.74) and ACC thickness (P = 0.06; F(3,275) = 2.50) among the study groups. Participants with evidence of greater impairment of neurophysiological assessments had a greater reduction in both precentral and postcentral cortical thickness (Supplementary Table 1); however, the strength of these correlations was weak.

The largest area (volume) of gray matter structural change in participants with painless DPN was centered in the postcentral cortex compared with HVs (p-MAX voxel coordinates: x = 13, y = 60, z = 43 [cluster size K_E = 4,897 voxels, threshold-free cluster enhancement (TFCE) P < 0.001] and x = 75, y = 60, z = 48 [cluster size K_E = 4,119 voxels; TFCE P < 0.001]) and participants with no DPN (x = 18, y = 59, z = 50 [cluster size K_E = 265 voxels; TFCE P = 0.02] and x = 40, y = 46, and z = 64 [cluster size K_E = 188 voxels; TFCE P = 0.01]). It encompassed the anatomical representation of the lower limb (Fig. 2). Similarly, in participants with painful DPN, the largest area of consistent gray matter structural change was in the postcentral cortex compared with HVs (x = 67, y = 56, z = 52 [cluster size K_E = 4,699 voxels; TFCE P < 0.001] and x = 36, y = 46, z = 57 [cluster size K_E = 4,177; TFCE P < 0.001]) and participants with no DPN (x = 40, y = 47, z = 63 [cluster size K_E = 132; TFCE P = 0.01]). There was another large cluster of gray matter structural change in precentral cortex in the painless DPN and painful DPN groups compared with the HV and no DPN groups (Supplementary Table 2 and Fig. 2). There were also significant clusters located in the insula cortex and cingulate cortex, albeit with smaller volumes (Supplementary Table 2). In the subcortical nuclei, thalamic structural changes in the painful and painless DPN groups were localized to the ventroposterior lateral nuclei compared with the no DPN group (painful DPN: x = 57, y = 46, z = 35 [cluster size K_E = 62; TFCE P = 0.02]; painless DPN: x = 57, y = 48, z = 32 [cluster size K_E = 171; TFCE P = 0.01]) and HV group (painful DPN: x = 57, y = 46, z = 32 [cluster size K_E = 275; TFCE P = 0.001]; painless DPN: x = 56, y = 46, z = 34 [cluster size K_E = 313; TFCE P < 0.001]). When the painful and painless DPN groups were compared, there was a significant reduction in gray matter volume in posterior cingulate cortex (PCC) in participants with painless DPN (x = 48, y = 36, z = 52 [cluster size K_E = 87; TFCE P = 0.02]) (Fig. 2) and in the ACC in participants with painful DPN (x = 45, y = 79, z = 28 [cluster size K_E = 103; TFCE P = 0.02]). There were no significant regions of cortical gray matter increase, with cluster size K_E >100 found in both neuropathy groups compared with the control groups.

Participants with painful DPN were divided into IR and NIR clinical pain phenotypes using detailed QST based on the DFNS protocol. There were no statistically significant differences in age and sex distribution, duration of diabetes, pain intensity score, or HbA_1c between the subgroups. There was a higher proportion of participants with type 2 diabetes (IR 47% vs. NIR 76.8%), with neurophysiological markers indicative of more severe DPN (sural amplitude mean [SD]: IR 6.9 [9.9] vs. NIR 2.4 [4.6] [P = 0.02]; peroneal latency: IR 5.6 [1.1] vs. NIR 7.3 [3.4] [P = 0.02]) in the NIR cohort (Supplementary Table 3). There were between-group differences in cold pain threshold (P = 0.012), pressure pain threshold (P < 0.001), mechanical pain sensitivity (P = 0.001), and dynamic mechanical allodynia (P = 0.045) suggestive of neuronal hypersensitivity/hyperalgesia in the IR nociceptor group (Supplementary Table 3). Mean combined primary somatosensory cortical surface area was significantly higher in the IR nociceptor phenotype compared with the NIR phenotype (IR 11.6 [1.4] vs. NIR 10.8 [1.3] cm²; P = 0.02; 95% CI 0.1, 1.4) (Supplementary Table 3). On VBM analysis (Fig. 2), participants with the IR nociceptor phenotype had significantly higher PCC (x = 48, y = 36, z = 52 [cluster size K_E = 87; TFCE P = 0.02]), right thalamic (x = 42, y = 51, z = 44 [cluster size K_E = 29; TFCE P = 0.03]), and left thalamic (x = 46, y = 56, z = 46 [cluster size K_E = 16; TFCE P = 0.03]) volumes compared with those with the NIR phenotype. Conversely, participants with the IR nociceptor phenotype had a significant reduction in mean combined ACC thickness compared with the NIR phenotype (IR 4.71 [0.4] vs. NIR 5.13 [0.4] mm; P < 0.001; 95% CI −0.6, −0.22) (Supplementary Fig. 2). There were no statistically significant differences in precentral and insular cortical volume measurements between the IR and NIR nociceptor phenotypes.

Significant reductions in sensorimotor cortical and ventrobasal thalamic nuclei volume in both the painful and painless DPN cohorts was a key finding of this study. Furthermore, participants with more severe neuropathy had greater reductions in the S1 and M1 regions, suggesting that the neuropathic process plays an important role in driving cortical changes in the brain. There was also a painful DPN phenotype-specific gray matter reduction in S1, PCC, and thalamic volumes in the NIR nociceptor phenotype and in ACC volume in the IR nociceptor phenotype. To our knowledge, this study is the largest to date providing a quantitative evaluation of gray matter alterations across carefully phenotyped participants with painful (including IR and NIR) and painless DPN. These findings confirm and extend the findings of previous studies.

In keeping with what was hypothesized based on prior research, there are statistically significant structural brain correlates overlapping across painful and painless DPN. The primary somatosensory and motor cortices have shown reduced volumes in several relatively small DPN studies, including in participants with painful and painless DPN (4,6,8), thus supporting its presence as a transdiagnostic marker for CNS involvement in DPN. Participants with painless DPN had the lowest S1 thickness compared with those with painful DPN. Furthermore, participants with DPN with evidence of more severe neuropathy had the greatest reduction in both S1 and M1 regions, albeit the strength of these correlations was relatively weak. Nevertheless, the strongest correlations were observed in neurophysiological assessments of large-fiber function indicative of a deafferentation or a dying back axonopathy leading to disuse atrophy. Although this explanation may be the most likely, reduction in regional brain volume may represent altered connectivity of sensory functions in participants with DPN. When participants with painful DPN were divided into IR and NIR nociceptor phenotypes, the latter group demonstrated the greatest reduction in S1 cortical volume. We have previously reported reduction in S1 cortical volume in participants with painful DPN with the painful insensate phenotype (20). One common trait shared between NIR and the insensate phenotype is the absence of hyperalgesia or evoked/contact hypersensitivity (e.g., allodynia). Taken together, our findings indicate that an absence of a peripheral sensory input, regardless of whether it is nociceptive, leads to a greater reduction in S1 cortical volume in DPN. Reduction in S1 cortical volume is in turn accompanied by functional changes that have been previously demonstrated, i.e., reorganization of lower-limb somatotropic functional representation (20) and alterations in somatosensory cortical functional connectivity (6).

There was also a significant reduction in M1, insular cortical, and thalamic volume in both the painful and painless DPN groups. These findings are consistent with previous small cohort studies in DPN demonstrating a reduction in M1 (20) and the insular cortex (8). As in previous studies, focal reduction in thalamic gray matter volume in DPN was only demonstrated using the VBM technique (8,21), which has greater sensitivity to detect regional changes in thalamic subcortical nuclei compared with whole thalamic volume measurements performed using the FreeSurfer technique. The ventrobasal nucleus is the key somatotrophic thalamic region, which receives projections from ascending somatosensory/nociceptive pathways (22). It does not merely act as a sensory relay station but modulates/processes sensory inputs before transmitting the information to the cerebral cortex. The ventrobasal thalamic nuclei complex is mainly involved in the sensory discriminative component of sensory processing and projects primarily to the S1. Using different modalities of magnetic resonance neuroimaging, we have previously demonstrated that preservation of thalamic neuronal function and increased thalamic blood flow were neuroimaging features that differentiated painful from painless DPN (3). Taken together with the subgroup differences in thalamic volume in this study, these findings suggest a key pathogenic role of supraspinal plasticity involving the thalamus following peripheral nerve injury in diabetes that is associated with the different clinical presentations of DPN.

Reduction in M1 volume may simply be regarded as a consequence of disuse atrophy from movement-related pain or anticipated movement-related pain (e.g., kinesiophobia); however, the similar reduction in M1 volume between painless and painful DPN suggests that the interactions between neuropathy and motor control are much more complex. It is likely that the interaction is bidirectional, i.e., the neuropathic process may have an effect on motor cortex activity (e.g., lack of somatosensory input, disuse of limb, loss of muscle targets in more advanced DPN), but motor-cortex activity may also have an impact on neuropathy (e.g., maladaptive plasticity/reorganization of the motor cortex is related to development/maintenance of neuropathic pain). More in-depth knowledge about these interactions is necessary to understand the physiology of the motor and sensory/nociceptive systems in patients with painful and painless DPN, as future rehabilitation strategies might be informed by these interactions or take them into account. Overall, it can be concluded based on this and previous studies that there is a high concordance of changes in the S1 and M1 systems in painless DPN. Overall, participants with painful DPN also demonstrated similar reductions in S1 and M1; however, regional differences began to emerge when this cohort was divided into IR and NIR nociceptor phenotypes.

In addition to alterations involving key somatosensory brain regions (S1 and thalamus) in participants with DPN, we report significant differences in cingulate cortical volume between the painful DPN IR and NIR nociceptor phenotypes. Participants with the IR nociceptor phenotype demonstrated a greater reduction in ACC volume, whereas those with the NIR nociceptor phenotype had a greater reduction in PCC volume. Although both painful DPN subtypes had similar pain intensity scores, ACC thickness was lower in participants with higher mechanical pain sensitivity scores, a measure of neuronal hypersensitivity. These findings suggest a link between alterations in cingulate cortical morphometry and clinical pain phenotype in DPN. Zhang et al. (8) also demonstrated a reduction in ACC thickness in participants with painful DPN, but this study did not examine the different subtypes of painful DPN separately. The ACC has been highlighted as a key area for nociceptive processing of both acute and chronic pain in humans (23). This region is activated in response to noxious pain and is involved in the descending modulation of pain and the emotional aspects of pain (24). The ACC also shows heightened activation in chronic pain states (25), with a number of studies demonstrating reduction in ACC volume in several chronic pain states (26). The reason for the reduction in ACC thickness in the IR nociceptor phenotype is uncertain but may be due to neuronal loss related to excitotoxicity (27). Indeed, increased ACC activation has been shown in participants with painful DPN, and this predicted successful treatment with duloxetine (28). Subsequent cluster analysis of the Combination vs. Monotherapy of Pregabalin and Duloxetine in Diabetic Neuropathy (COMBO-DN) study showed that optimizing neuropathic pain treatment with duloxetine was more beneficial in patients with predominant symptoms of paraesthesia/dysesthesia, i.e., clinical features most closely related to the IR nociceptor phenotype (29).

The strengths of the study include the large sample size, detailed neuropathy phenotyping of study participants, and the use of robust/validated MRI acquisition and analysis methodology. Moreover, the wide age range of our cohort (18–85 years old) ensured wider representation of people with diabetes. Most studies exclude older adults (70–85 years old), despite an increased prevalence of diabetes in an aging population (30). However, as age-related neuropathy could coexist with DPN, we conducted a sensitivity analysis by excluding participants from this older age range and found similar results to the primary analysis of the whole data set.

Study limitations include the cross-sectional nature of this study, which precludes a causal association to be reached. There were significant subgroup differences in age and sex, but the effects persisted in models with adjustments for both covariates and the use of normalization procedures to control for global shape and brain size differences. Nevertheless, future studies should examine an exact-matched sample of participants for age and sex. Likewise, although participants were carefully characterized using detailed clinical and neurophysiological assessments, certain clinical characteristics, such as coexisting mood disorders, were not considered. In this regard, it should be noted that brain volume analysis was focused on selected ROIs mainly involved with somatosensory and nociceptive processing. Future studies should also examine other brain regions involved with wider pain processing, such as the dorsal lateral prefrontal cortex (24). We used the DFNS QST protocol to classify participants into IR and NIR nociceptor pain phenotypes. This approach predominantly assesses superficial sensory function, with limited assessment of deep sensory function (apart from pressure pain threshold). Moreover, there is likely to be some overlap between these two clinical pain subsets. However, at present, this is the best way to phenotype patients with painful DPN, as several studies have demonstrated an association between clinical pain phenotypes and treatment response (5,6) and underlying disease mechanisms (31). There was also a higher-than-usual proportion of participants with painful DPN with the IR nociceptor phenotype, which may be explained by the rigorous examination over the whole surface of both feet rather than a small, fixed location in one foot where the features of the IR phenotype can be missed. Nevertheless, these results should be taken with appropriate caveats, and these additional factors should be considered in future studies.

In conclusion, this study points to specific alterations in cortical geometry in painless DPN and painful DPN, including the IR and NIR nociceptor phenotypes. There was a reduction in key somatosensory (S1, thalamus) and motor cortical brain regions in participants with DPN, which is related to neurophysiological assessments of neuropathy, indicating that the neuropathic process is driving these alterations. Compared with participants with painless DPN, those with painful DPN were uniquely characterized by alterations in the cingulate cortex; specifically, ACC volume reduction in the IR nociceptor phenotype and PCC volume reduction in the NIR phenotype. These findings indicate that cerebral structural alterations are involved in determining the clinical phenotype of painful DPN, which opens new lines of research to investigate whether it can be used to stratify patients for mechanism-based treatments. Finally, to truly elucidate the causality of the observed relation between DPN and brain structure, more whole-brain longitudinal studies are needed.

Acknowledgments. The authors acknowledge the hard work, skills, and contributions of the radiographers at the University of Sheffield Magnetic Resonance Imaging Department. The authors also greatly appreciate the study volunteers who spent considerable time participating in this study. Dr. Iain D. Wilkinson, who substantially contributed to this research, died on 22 October 2020 before publication of this work. Dr. Wilkinson will be fondly remembered and sadly missed.

Funding. This study was supported by the European Foundation for the Study of Diabetes European Research Programme in Microvascular Complications of Diabetes supported by Novartis; the Knowledge Exchange Support Fund, University of Sheffield (X/162218); and the Efficacy and Mechanism Evaluation Programme (NIHR129921), a Medical Research Council and National Institute for Health and Care Research partnership.

The views expressed in this publication are those of the authors and not necessarily those of the Medical Research Council, National Institute for Health and Care Research, or the Department of Health and Social Care. The study sponsor/funder was not involved in the design of the study; collection, analysis, and interpretation of data or writing of the report and did not impose any restrictions regarding the publication of the report.

Duality of Interest. S.T. has received lecture honoraria from Wörwag Pharma, Pfizer, Novo Nordisk, Merck, EVA Pharma, Hikma Pharmaceuticals, Grünenthal, Abbott Laboratories, AstraZeneca, and Trigocare International and is on the medical advisory board of Bayer. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. D.S. and G.S. recruited participants, undertook clinical and neurophysiological assessments, researched and analyzed clinical and magnetic resonance data, and wrote the manuscript. G.S., F.H.-G., M.A., and S.P. are certified in DFNS QST assessments and undertook clinical and neurophysiological assessments, drafted the manuscript, and gave final approval of the version to be published. K.T. and I.D.W. contributed to the design; acquisition, analysis, and interpretation of data; and drafting and final approval of the manuscript. A.K. collected and collated data on clinical and neurophysiological assessments, drafted the manuscript, and gave final approval of the version to be published. S.T. contributed to the conception and design, review and revision of the manuscript, and final approval of the version to be published. D.S. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.