Causal Central Network Remodeling in Diabetic Neuropathy: An Integrate

Introduction

Diabetic peripheral neuropathy (DPN), a prevalent and debilitating chronic complication of diabetes mellitus, affects 67.6% of patients, with 57.2% experiencing painful DPN (PDPN).¹ Clinical manifestations range from asymptomatic numbness to intractable pain, frequently causing sleep disturbances, mood disorders, and reduced quality of life.^2,3 Notably, 25–50% of DPN patients develop neuropsychological impairments such as anxiety or depression, correlating with diminished psychomotor efficiency on cognitive assessments.⁴

Diabetes-associated central nervous system (CNS) pathology was initially documented through spinal cord atrophy and demyelination in 1960s autopsies,⁵ while contemporary neuroimaging confirms structural CNS damage in DPN, including: spinal cord atrophy, somatosensory cortex grey matter loss, thalamic metabolic dysfunction, functional reorganization,^6,7 and altered neural activation patterns are evident across thalamic, insular, cingulate, and prefrontal regions.^8,9

Additionally, DPN patients frequently exhibit visual impairments.¹⁰ Subjects with DPN but without diabetic retinopathy (DR) exhibit increased macular thickness, reduced retinal nerve fiber layer (RNFL) thickness, and significantly impaired visual functions, suggesting optic nerve involvement via CNS mechanisms.¹¹ Animal models reveal CNS-mediated excitotoxicity and neurovascular changes contributing to visual pathway dysfunction,^12,13 while clinical studies indicate metformin improves retinal-brain axis abnormalities in diabetic retinopathy,¹⁴ collectively establishing CNS-visual impairment associations.

Traditional observational studies examining brain-DPN relationships face limitations from confounding and reverse causation. Mendelian randomization (MR) overcomes these by leveraging genetic variants as instrumental variables, effectively simulating randomized controlled trials to infer causality.¹⁵

Although electrophysiological techniques (eg, nerve conduction studies, NCS) are considered diagnostic gold standards for peripheral neuropathy, their sensitivity to central pathway impairment remains suboptimal.¹⁶ Advanced neuroimaging techniques provide complementary insights: magnetic resonance spectroscopy (MRS) detects regional metabolite concentrations,¹⁷ voxel-based morphometry (VBM) quantifies gray matter volume, and diffusion tensor imaging (DTI) assesses white matter integrity.¹⁸ Resting-state fMRI (rs-fMRI) decodes blood-oxygen-level-dependent (BOLD) fluctuations to dynamically map functional networks, with the Amplitude of Low-Frequency Fluctuations (ALFF) serves as a robust indicator of spontaneous neural activity within specific brain regions,¹⁹ Functional Connectivity (FC) analysis quantifies the temporal coherence between distinct neural regions.²⁰ The integration of ALFF and FC analyses in rs-fMRI analysis provides a comprehensive framework for understanding disease-related disruptions in brain function across both local and global scales. By integrating bidirectional MR with rs-fMRI network analyses, our study establishes neuroimaging signatures for preclinical DPN detection and redefines therapeutic targets, bridging molecular diabetology and systems neuroscience.

Materials and Methods

GWASs of Brain rs-fMRI and DPN in MR Analysis

This study used rs-fMRI data from Reference,²¹ which analyzed associations between 1,777 intrinsic brain activity phenotypes and 9,026,427 genetic variants in the UK Biobank (n = 34,691).^22–24 Of these, 191 phenotypes significantly influenced by genetic variation (p < 5×10⁻⁸) were selected, including amplitude features, functional connectivity pairs, and global connectivity measures. Details can be found in the original publication and Supplementary Table 1. GWAS summary statistics of DPN were obtained from FinnGen Biobank (R12 release: 3862 DPN cases:92,612 controls; https://risteys.finngen.fi/endpoints/DM_NEUROPATHY). Significant SNPs (p < 5×10⁻⁶) were pruned using a linkage disequilibrium threshold (R²> 0.001) and filtered to exclude confounding traits. Exposure and outcome data were standardized, and palindromic SNPs were removed. F-statistics (>10) were calculated to ensure robust instrument strength.²⁵

Two-Sample MR Analysis Procedure

We performed bidirectional two-sample MR analyses to explore the causal relationships between DPN and rs-fMRI phenotypes. Brain rs-fMRI served as the exposure and DPN as the outcome in forward MR analysis, while DPN were used as exposure and brain rs-fMRI were used as outcome in the reverse MR analyses. The inverse-variance-weighted (IVW) model as the primary method for causal evaluation.²⁶ Additional methods, including weighted mode, weighted median, MR-Egger regression, and simple mode, were applied for result validation. Odds ratios (ORs) with 95% confidence intervals (CIs) were used to present effect estimates.²⁷ To reduce horizontal pleiotropy, MR-PRESSO identified outlier instruments, and heterogeneity tests using IVW and MR-Egger regression quantified heterogeneity with Cochran’s Q statistics (P < 0.05). A leave-one-out analysis assessed result stability by sequentially excluding each SNP.²⁸ A detailed overview of the process is illustrated in Figure 1.

Study Population in Clinical rs-fMRI Study

We recruited 16 patients with DPN (age 55.88±9.44 years, 5 females), 24 NDPN patients (age 37.88±9.08 years, 5 females), and 20 healthy controls (age 32.25±7.66 years, 9 females) from Renmin Hospital of Wuhan University outpatient clinics between February 2023 and September 2024. This study was approved by the Ethics Committee of Institutional Review Board of Renmin Hospital of Wuhan University (Ethics No.WDRY2024-K220). All participants provided written informed consent, and the research adhered to the principles of the Declaration of Helsinki.

All patients met the 1999 WHO T2DM diagnostic criteria, while DPN diagnosis adhered to the Toronto consensus criteria. Inclusion criteria for patients included age 30–70 years, right-handedness, and at least a junior high school education. NDPN patients had confirmed T2DM but no neuropathic symptoms, normal nerve conduction studies, and exclusion of other neuropathy causes. Healthy controls were aged 30–70 years, right-handed, had no history of diabetes (HbA1c 4–6%), and were free from severe medical, psychiatric, or neurological disorders.

Exclusion criteria encompassed left-handedness, other causes of neuropathy, brain trauma or tumors, acute T2DM complications, cerebrovascular or psychiatric conditions, alcohol or drug abuse, and MRI contraindications.

Clinical Assessment and Questionnaire

First, general data, medical history, and blood biochemical tests were collected from all participants. Then, neurological, psychological, and cognitive assessments were conducted, with the Montreal Cognitive Assessment (MoCA) Beijing version used for a comprehensive evaluation of cognitive function. Next, the Douleur Neuropathique 4(DN4) was used to quantitatively assess the severity of pain. The Visual Function Index-14 (VF-14) questionnaire was used to assess the visual function of patients. All tests were administered by trained and qualified assessors.

rs-fMRI Data Acquisition and Preprocessing

All rs-fMRI scans were obtained on the same GE Discovery MR750w 3.0TMR scanner. The rs-fMRI data were acquired with an echo-planar imaging sequence that included the following parameters—repetition time: TR 2000 ms, TE 25 ms, flip angle 90°, field of view 240 mm×240 mm, number of layers 40, layer thickness 3.5 mm, layer spacing 0.6 mm, matrix 64×64, scanning time about 8 min. During the scan acquisition, the participants were asked to keep their eyes closed, not to think of anything and not to fall asleep.

Spatial preprocessing of rs-fMRI images was conducted using Statistical Parametric Mapping (SPM12, https://www.fil.ion.ucl.ac.uk/spm/software/spm12/). To minimize acquisition delays and head movement artifacts,²⁹ rs-fMRI images were corrected accordingly. Subjects with head movements exceeding 3 mm or rotations greater than 3°during scanning were excluded from further analysis. The rs-fMRI data were normalized to the standard SPM12 echo planar imaging (EPI) template and resampled to a voxel size of 3 × 3×3 mm³.³⁰ To reduce the impact of low-frequency drifts and physiological high-frequency noise, the data were temporally band-pass filtered (0.01–0.08 Hz). Additionally, linear trends were removed, and head movement parameters were regressed out.³¹

ALFF and FC Analysis

The rs-fMRI Data Analysis Toolkit was used to conduct ALFF analysis. Using fast Fourier transform (FFT), the filtered time series was transformed into the frequency domain. The power spectrum was obtained by applying a squared FFT and averaging across the 0.01–0.08 Hz frequency range for each voxel.¹⁹ The square root of the mean power was defined as the ALFF value. A mask based on the Montreal Neurological Institute (MNI) template was created to exclude non-cerebral tissues, ensuring alignment with the normalization procedure. To facilitate inter-subject comparison, ALFF values were normalized by dividing each voxel’s ALFF by the whole-brain mean ALFF. Finally, a zALFF map was generated by standardizing ALFF values, subtracting the mean, and dividing by the standard deviation.²⁰

Brain regions with altered activity identified through ALFF analysis were selected as seeds for whole-brain FC analysis. Pearson correlation coefficients between each seed and the whole brain were computed, followed by Fisher’s Z-transformation for normalization.

rs-fMRI Statistical Analysis

Group differences were assessed using DPABI. A one-way ANCOVA compared ALFF values among the three groups, controlling for age, sex, disease duration, and mean FD. Post hoc analyses with Scheffé correction were performed, and adjusted p-values were converted to z-scores. Significant clusters were identified using FDR correction (p < 0.05) and localized with the Automated Anatomical Labeling map. Regions showing significant ALFF differences were selected as seeds for FC analysis.³² Group-level FC comparisons used ANCOVA, adjusting for age, sex, mean FD, disease duration and cognitive competence, with significant clusters identified using GRF correction (voxel-wise p < 0.001, cluster-wise p < 0.05).^33,34 Correlations between ALFF of significant brain area and clinical/cognitive variables were also analyzed.

Receiver Operating Characteristic Curve Analysis

ALFF values from the LSFG were extracted after post hoc testing, and receiver operating characteristic (ROC) curves were used to evaluate their potential as neuroimaging biomarkers for distinguishing DPN from NDPN patients. The area under the curve (AUC),³⁵ maximum Youden index, optimal cutoff point, sensitivity, specificity, and 95% confidence intervals (CIs) were calculated.³⁶ Statistical comparisons of the AUCs for each mean value were then performed.

Results

Clinical and Neuropsychological Data

Compared to the HC and NDPN groups, the DPN group was older (p < 0.001), demonstrated lower cognitive function as assessed by the MoCA (p < 0.001), had more severe pain perception (p < 0.001), had a longer duration of the disease (p < 0.001). Additionally, the DPN group exhibited higher levels of triglycerides (TG) compared to the NDPN group (p < 0.001). The demographic details and clinical characteristics of each group are shown in Table 1.

Table 1 Demographic and Clinical Characteristics of DPN, NDPN and HC (X±s/ M(P25,P75))

Bidirectional Influences Between Brain Network and DPN in MR Analysis

The full MR analysis results are available in Supplementary Tables S2, S3 and Figure 2. In forward analysis, we identified that a one standard deviation decrease in functional connectivity of default mode network (DMN) and visual network casually corresponds to a 39% increased risk of DPN (IVW OR = 0.61, 95% CI: 0.43–0.86, P = 0.035). In reverse analysis, higher risk of DPN increased the connectivity of the subcortical-cerebellar network (IVW OR = 1.05, 95% CI: 1.01–1.09, P = 0.010). No heterogeneity or horizontal pleiotropy was found (Figures 3 and 4). Additional analyses, including forest plots, scatter plots, funnel plots, and leave-one-out analyses, are shown in Supplementary Figures S1 and S2.

Figure 2 Ring heat map for the MR analysis. (A) Forward MR Analysis (rs-fMRI for exposure, DPN for outcome). (B) Reverse MR Analysis (DPN for exposure, rs-fMRI for outcome). The causal relationships between 191 rs-fMRI and DPN are illustrated. Each ring represents a different analytical method, and the association between each phenotype and DPN is visualized through variations in color intensity and position.

Figure 3 Causalities in the forward MR analysis. (A) Forest plots demonstrate significant causal relationships estimated using the five MR Methods. The OR represents the magnitude of the effect on the risk of DPN per 1 standard deviation change in the mean rs-fMRI phenotype, and the error line represents the 95% confidence interval. P values were derived from IVW analyses. Right shows the brain anatomical regions of the corresponding rs-fMRI phenotype. (B) Heterogeneity and horizontal pleiotropy analysis of MR analysis.

Figure 4 Causalities in the Reverse MR analysis. (A) Forest plots demonstrate significant causal relationships estimated using the five MR Methods. The OR represents the magnitude of the effect on the risk of rs-fMRI per 1 standard deviation change in the mean DPN phenotype, and the error line represents the 95% confidence interval. P values were derived from IVW analyses. Right shows the brain anatomical regions of the corresponding rs-fMRI phenotype. (B) Heterogeneity and horizontal pleiotropy analysis of MR analysis.

Altered Brain Activity Among Groups

After removing age, sex, mean FD, disease duration and cognitive competence as covariates, compared to the NDPN group, the DPN group exhibited a significant increase in the ALFF in left superior frontal gyrus (LSFG, modulates pain and integrates emotions). No significant differences observed between DPN and NDPN groups, NDPN and HC groups. For further details, please refer to Table 2 and Figure 5.

Table 2 ALFF Values Were Significantly Different Between Groups (P <0.05 After FDR Correction Using ANCOVA and Sheffe Post Hoc Tests)

Figure 5 Clusters of significant changes in ALFF values among groups (analyzed using analysis of covariance (ANCOVA) and Scheffé post hoc tests, with FDR correction, P <0.05). Red regions (LSFG, left superior frontal gyrus) indicate that the DPN group has higher ALFF values compared to the HC group.

Correlation Analysis

We conducted a correlation analysis between the ALFF values of the LSFG and clinical indicators as well as scale scores. The ALFF of LSFG was positively correlated with durations of DM (r = 0.55 p =0.0022) and negatively correlated with DN-4 (r = −0.47, p = 0.0024). The details are shown in Supplementary Figure S3.

ROC Curve Analysis Results

According to the ROC curves, the mean ALFF values of the LSFG can distinguish DPN from patients with NDPN, which all achieved a significance level of P<0.05 for AUC. An AUC of 0.79 indicates a moderately performing classifier, demonstrating a certain level of predictive value. The details are shown in Figure 6.

Figure 6 ROC curves for ALFF that distinguish patients with NDPN from patients with DPN and controls. The AUC of “ALFF_LSFG” index was 0.792, which had moderate diagnostic performance. Abbreviations: ALFF, amplitude of low-frequency fluctuation; AUC, area under the curve; LSFG, left superior frontal gyrus.

Functional Connectivity Analysis

Compared to the HC group, the DPN group exhibited decreased FC between LSFG and right lingual (LING.R, basic visual feature processing). Conversely, FC between the LSFG and right precuneus (PCUN.R, complex visuospatial processing, visual input integration, and episodic memory within the DMN) increased. No significant changes were observed in DPN and NDPN groups, NDPN and HC groups. Shown in Table 3 and Figure 7.

Table 3 FC Values Were Significantly Different Between Groups, Seed as LSFG

Figure 7 Brain regions with significantly different FC using LSFG as seed points. LSFG has an increased connection to PCUN.R and a decreased connection to LING.R. Abbreviations: LSFG, left superior frontal gyrus; PCUN.R, right precuneus; LING.R, right lingual gyrus.

Discussion

This study identified altered intrinsic brain activity in DPN patients, characterized by abnormal ALFF and distinct FC patterns in emotional and visual processing regions. Notably, elevated ALFF in the LSFG emerged as a potential neuroimaging biomarker for DPN. To our knowledge, this represents the first integration of Mendelian randomization with clinical resting-state fMRI, enabling comprehensive exploration of regional brain alterations and the “metabolism-neurodegeneration-compensation imbalance” cycle in DPN.

Significant age differences between groups reflected longer disease duration and more frequent neurological screening in elderly patients.³⁷ The DPN group showed higher DN4 scores and lower MoCA scores, indicating greater pain severity and cognitive impairment that may indirectly contribute to negative emotional states.³⁸ To address this demographic heterogeneity, we included age and cognitive function as a covariate to remove in subsequent analyses. Elevated triglyceride (TG) levels in DPN patients align with preclinical evidence linking TG overload to impaired nerve metabolism and mitochondrial dysfunction.³⁹ These findings align with results from a Danish cohort study that demonstrated a significant association between elevated triglycerides and increased DPN risk.⁴⁰

Forward MR demonstrated that reduced DMN-visual connectivity increases DPN risk. The DMN supports environmental monitoring, emotional processing, and self-referential cognition,^41,42 while the visual network coordinates visual perception.⁴³ Notably, Both networks contain NMDAR-GLUT1 co-expressing neurons highly vulnerable to glycoxidative stress-induced mitochondrial damage.⁴⁴ This connectivity disruption may impair visual processing, spatial orientation, and attentional allocation – consistent with findings in other chronic pain conditions.^45,46 Our results suggest network-specific dysconnectivity acts as a causal driver rather than mere consequence of DPN progression.

Reverse MR revealed that genetic DPN risk enhances subcortical-cerebellar connectivity. Peripheral sensory deafferentation likely triggers cerebellar compensation for degraded proprioceptive inputs.⁴⁷ Subcortical-cerebellar circuits also exhibit abnormal connectivity in psychosis, potentially associated with cognitive dysfunction, positive symptoms (e.g.hallucinations, delusions) and negative symptoms (eg social withdrawal, lack of motivation).⁴⁸ Thus, this compensatory mechanism may paradoxically amplifies error prediction signals and ultimately leads to maladaptive cortical reorganization, even mental illness.

The DPN group demonstrated significantly elevated ALFF values in the LSFG, a region integral to top-down pain inhibition modulation,⁴⁹ previously implicated in both chronic neuropathic pain conditions and severe ocular pain manifestations.⁵⁰ The LSFG serves as a critical hub for integrating emotional inputs with environmental stimuli.^51,52 An Italian cohort study, which identified pain unpredictability and reduced pain control as primary contributors to depression in DPN patients.⁴ Additionally, DPN patients commonly experience anxiety related to disease progression, potential injury, and perceived social stigma.⁴³ Thus, the observed heightened LSFG activity may represent a dual mechanism: enhanced nociceptive processing coupled with an adaptive response for managing the emotional and cognitive challenges associated with the condition.⁵³ The positive ALFF-duration correlation suggests progressive exacerbation of negative emotions. Importantly, ROC analysis confirmed LSFG ALFF effectively discriminates DPN from non-DPN diabetics (AUC=0.79), supporting its utility as a non-invasive neuroplasticity biomarker.

Subsequently, using the LSFG as a seed region, we assessed whole-brain functional connectivity. Compared to the HC group, the DPN group exhibited decreased FC between the LSFG and the LING.R. The lingual gyrus primarily supports basic visual feature processing, including shape, color, and letter recognition.⁵⁴ Previous research has documented reduced brain activity in the lingual gyrus among T2DM patients, with cognitive impairment notably preceding the onset of visual dysfunction,⁵⁵ which is consistent with our MR Study in the previous part. We hypothesize that this decline may result from attentional reallocation, where the brain prioritizes pain regulation and emotional processing, thereby reducing connectivity with visual areas.⁵⁶ Additionally, this may represent an adaptive protective segregation mechanism, in which the brain actively downregulates synchronization with non-essential visual input to minimize interference with critical emotional and pain regulatory processes.

In contrast, increased FC between the LSFG and the PCUN.R was observed in DPN patients. Notably, the precuneus has traditionally been considered a visual region involved in complex visuospatial processing and integrating visual inputs.⁵⁷ Additionally, it serves as a key hub of the DMN, associated with episodic memory and visuospatial imagery.⁵⁸ Enhanced inter-DMN functional connectivity in painful DPN has been already demonstrated,⁵⁹ and alterations in the connectivity of the precuneus have been observed in depression and bipolar disorder.^60,61 We propose that this enhanced FC may represent a compensatory mechanism in DPN patients with visual deficits. These patients may rely more heavily on available contextual cues, such as facial expressions and social interactions,⁶² and potentially employ mental imagery to infer and compensate for visual impairments.⁶³ However, research has shown that prolonged negative self-consciousness activates the precuneus,⁶⁴ which is associated with the onset of anxiety and depression. Therefore, the chronic enhancement of connectivity between these regions may potentially increasing psychiatric vulnerability.

Combined MR analysis and rs-fMRI, we propose the following hypothesis: hyperglycemia genetically impairs metabolically sensitive DMN-visual network connections, disrupting the integration of basic visual processing (as evidenced by forward MR results) and higher-order spatial conformation. During the early compensatory phase, loss of peripheral sensory neurons triggers enhanced FC between the cerebellum and thalamus via the spinocerebellar tract (as shown by reverse MR results), which rapidly recalibrates motor output. Concurrently, increased ALFF in the LSFG may reflect cognitive inhibition of aberrant sensory inputs by the prefrontal cortex (eg, pain modulation). In the late maladaptive phase, chronic enhancement of hyperconnectivity between the superior frontal gyrus and precuneus may lead to DMN encroachment, resulting in excessive precuneus involvement in visual scenario simulation (as a compensatory substitute for impaired real-time visual processing), may interferes with sensory-motor integration efficiency, perpetuating a vicious cycle of “metabolic dysfunction-neurodegeneration-dysregulated compensation.”

This study has several limitations. First, demographic differences among groups in gender ratio, age, and disease duration, while controlled for as covariates, suggest the need for better-matched participants. Second, the small sample size and cross-sectional design limit our understanding of dynamic brain activity changes in DPN, necessitating larger, longitudinally studied cohorts. Third, medication effects on neural activity warrant future studies focusing on drug-naïve patients. Fourth, the predominantly European GWAS dataset requires validation in other populations. Finally, while genetic variations parallel environmental factors, they may not fully replicate environmental changes, potentially biasing MR analysis results and requiring further clinical validation.

Conclusions

This study integrates MR with rs-fMRI to systematically elucidate the cross-scale network reorganization mechanisms underlying DPN. Key findings include that changes in LSFG may serve as a potential neuroimaging biomarker for DPN diagnosis, reflecting abnormalities in pain and emotional processing. The reduction in DMN-visual network functional connectivity is identified as a causal driver of DPN onset, while DPN genetic risk achieves central compensation by enhancing subcortical-cerebellar functional connectivity. Resting-state functional imaging reveals dynamic network remodeling in DPN patients: increased spontaneous neural activity (ALFF) in the superior frontal gyrus and decreased functional connectivity with the lingual gyrus reflect metabolic attacks on primary visual pathways, while elevated connectivity with the precuneus suggests higher-order cognitive networks recruit spatial memory to compensate for impaired visual input. This “metabolic injury-compensatory remodeling” pattern reveals DPN’s pathological evolution from central to peripheral systems and provides clinical research with novel biomarkers and intervention targets.