Diabetes mellitus (DM) is a prevalent metabolic disorder associated with various microvascular complications (Xu, 2015; Bahtiyar et al., 2016; Bansal et al., 2023), among these, DR is a significant microvascular complication affecting the retina in diabetic individuals and stands as a leading cause of global blindness (Wong et al., 2016). A survey conducted in 2020 revealed a worldwide DR prevalence of 22.27% in diabetic populations, impacting approximately 103.12 million individuals (Teo et al., 2021). Risk factors for DR include hypertension, dyslipidemia, obesity, pregnancy, and prolonged diabetes duration (Lopes de Faria et al., 2011; Ting et al., 2016; Wondmeneh and Mohammed, 2024). The primary pathological manifestations of DR encompass capillary non-perfusion, vascular leakage, and degeneration, progressing to proliferative retinal detachment and eventual vision loss. Notably, retinal vessels share similarities in anatomy, physiology, and embryology with cerebral vasculature (Kwa et al., 2002; van de Kreeke et al., 2018). Notably, retinal vessels share similarities in anatomy, physiology, and embryology with cerebral vasculature (Hägg et al., 2013; Sanahuja et al., 2016). Individuals with DR face heightened susceptibility to dementia and Alzheimer’s disease (Ciudin et al., 2017; Biessels and Despa, 2018), and potential cognitive impairment (Crosby-Nwaobi et al., 2013; Chai et al., 2024). While evidence indicates brain alterations and abnormal central nervous system (CNS) activity in DR patients, the underlying neurophysiological mechanisms necessitate further exploration.

Resting-state functional magnetic resonance imaging (rs-fMRI), a non-invasive imaging technique, enables the examination of spontaneous intrinsic brain activity by measuring fluctuations in blood oxygen level-dependent (BOLD) signals without specific tasks or stimuli (Fox and Raichle, 2007; Barkhof et al., 2014). Studies using resting-state functional MRI have revealed diverse patterns of spontaneous brain activity alterations in DR patients, including amplitude of low-frequency fluctuations (ALFF), dynamic ALFF (dALFF), regional homogeneity (ReHo), and functional network connectivity (FNC), predominantly observed in brain regions like the middle occipital gyrus, calcarine, anterior cingulate gyrus, posterior cerebellar lobe, and networks associated with visual and cognitive functions (e.g., visual network, default mode network, executive control network, and salience network). These findings shed light on the neural mechanisms underlying visual and central nervous system abnormalities in DR (Huang et al., 2019, 2020, 2021a; Liao X.-L et al., 2019; Qi et al., 2020; Shi et al., 2022). Functional connectivity density (FCD) analysis, an extension of the FC method, assesses the importance of network nodes in functional integration by measuring their total connectivity strength with all other nodes in the brain network (Lv et al., 2018; Tomasi and Volkow, 2010). This approach, without requiring prior assumptions, is suited for exploratory analyses (Tomasi et al., 2016; Pang et al., 2017; Zhang et al., 2017). Moreover, the human brain is a complex dynamic system with non-stationary neural activity and constantly changing neural interactions (Hutchison et al., 2013; Calhoun et al., 2014; Liao W. et al., 2019), implying the importance of investigating dynamic neural communication across different brain regions in DR patients utilizing the dFCD approach. However, the specific alterations in dFCD among DR patients remain unclear.

DR has a hereditary component, with estimated heritability ranging from 6 to 33% (Kim et al., 2022). Multiple studies have demonstrated familial clustering of DR (Zhang et al., 2010; Cho and Sobrin, 2014), with notably high concordance rates among family members; 65% in twins with type 1 diabetes and 95% in twins with type 2 diabetes (Leslie and Pyke, 1982). These findings imply the genetic regulation of DR onset and severity. Genome-wide association studies (GWAS) have the potential to identify genetic risk loci associated with DR. A comprehensive meta-analysis has revealed several gene loci linked to DR that are crucial for cell survival, insulin signaling, and angiogenesis (Grassi et al., 2011). Moreover, various studies have highlighted the influence of genetic factors on resting-state brain function (Fornito et al., 2011). Nevertheless, the genetic underpinnings of neuroimaging changes in DR patients are not effectively captured by GWAS. This challenge is compounded by the fact that most loci identified by GWAS are situated in regulatory, rather than coding, regions of the genome, posing significant challenges for their biological interpretation (Edwards et al., 2013; Tam et al., 2019). Spatial correlation analyses integrating transcriptomics and neuroimaging data from the Allen Human Brain Atlas (AHBA), known for its case-control imaging discrepancies, were conducted to explore the genetic basis of macroscopic neuroimaging alterations (Arnatkeviciute et al., 2022; Xue et al., 2022). This method is widely used to unveil gene expression patterns associated with structural and functional variations in patients’ brains, although investigations on the genetic mechanisms underlying dFCD changes in DR are lacking.

Therefore, In our study, we initially assessed and compared dFCD differences between DR patients and controls. Subsequently, utilizing gene expression data from the AHBA database integrated with dFCD maps, we conducted spatial correlation analyses to identify genes whose expression levels correlated with dFCD changes in DR. Finally, gene enrichment, specific expression, and protein-protein interaction analyses were carried out on the identified genes to elucidate their functions and understand the molecular mechanisms behind dFCD alterations in DR.

This study represents an innovative approach integrating functional magnetic resonance imaging with transcriptome-neuroimaging spatial correlation to explore the genetic underpinnings of dFCD variability in diabetic retinopathy (DR). Our investigation revealed a significant reduction in dFCD variability within the left anterior cingulum, left superior occipital gyrus, and right postcentral gyrus in the DR group compared to the healthy control (HC) group. Transcriptome-neuroimaging correlation analyses identified two groups of DR-related genes (PLS1 + and PLS1-) that exhibited substantial enrichment in biological functions and pathways related to ion channels, synaptic function, neuronal systems, and cellular junctions. Specific expression analyses indicated the unique expression of these genes in Purkinje neurons, the cortex, and striatum brain regions, with no developmental period-specific expression observed. Additionally, genes linked to altered brain function were capable of forming a protein-protein interaction (PPI) network supported by hub genes. These insights suggest that the disrupted brain function in DR patients arises from intricate interactions among multiple genes with diverse functional characteristics, offering novel perspectives on the molecular mechanisms linked to dFCD alterations in DR patients.

Significantly reduced dFCD variability was observed in the left anterior cingulum, left superior occipital gyrus, and right postcentral gyrus in the diabetic retinopathy (DR) group compared to the healthy control (HC) group. The anterior cingulate gyrus, a component of the salience network, plays a role in higher functions like attention allocation (Pardo et al., 1990), reward anticipation, decision making, and emotion (Jackson et al., 2006). A study by Liao et al showed significantly reduced ReHo values in the right anterior cingulate gyrus in patients with DR and hypothesized that this reflected dysfunction of the anterior cingulate gyrus (Liao X.-L et al., 2019). Similar to their results, we found significantly reduced dFCD variability in the left anterior cingulate gyrus in DR patients, which may imply anterior cingulate gyrus dysfunction and lead to corresponding cognitive impairment in DR patients. The superior occipital gyrus is located in the occipital lobe and is mainly responsible for the processing of visual information. It has been found that functional connectivity from the bilateral lingual gyrus to the left superior occipital gyrus was reduced in the DR group compared to healthy controls (Huang et al., 2021b), which suggests disrupted connectivity of the visual network and the dorsal visual pathway, and may reflect impaired top-down modulation and visuospatial function in patients with DR. In another study, gray matter density (GMD) in the right occipital lobe of DR patients was significantly lower than that of healthy controls (Wessels et al., 2006). In addition, there is a significant decrease in intra-network functional connectivity (FC) of the visual network (VN) in DR patients (Huang et al., 2020). Combining these studies, we suggest that the reduced dFCD variability in the left superior occipital gyrus may reflect impaired visual function in DR patients. The postcentral gyrus is located in sensorimotor network (SMN) (Chenji et al., 2016), which plays an important role in motion control. One study found reduced local clustering in the postcentral gyrus in DR patients, which correlates with motor speed (van Duinkerken et al., 2016). Postcentral gyrus is closely associated with PVA-related (primary visual areas) spontaneous brain activity (Wang et al., 2008), and their connection affects the processing of spatial visual information (Ringach et al., 1996). We hypothesize that reduced dFCD variability in the postcentral gyrus of DR patients may cause abnormalities in motor control, such as abnormal eye movements.

Transcription-neuroimaging spatial association analysis is an effective method to study the genetic mechanisms underlying neuroimaging phenotypes in the patient’s brain (Zhang et al., 2022), and can identify genes associated with neuroimaging phenotypes (Fornito et al., 2019). Using transcriptional-neuroimaging spatial association analysis, we identified two groups of genes, PLS + and PLS-, which correlate with changes in dFCD in DR patients. Enrichment analyses showed that these genes were mainly enriched for biological functions and pathways such as ion channels, synapses and cellular junctions. It was shown that ion channels are closely related to neuronal growth and differentiation, membrane potential generation, signal transduction and neurotransmitter release (Moody and Bosma, 2005; Smith and Walsh, 2020; Kumar et al., 2016; Hull and Isom, 2018). Alterations in the expression levels of somatostatin (SRIF) or specific SRIF receptors affect the morphology of retinal cell types (i.e., optic rod cells) and may influence transmitter release in photoreceptors by modulating voltage-gated K + and Ca2 + currents, making SRIF a potentially promising drug SRIF for the treatment of DR (Casini et al., 2005). 11-cis-retinal intervention improves aberrant outer retinal ion channel closure, thereby alleviate diabetes-induced outer retinal layer dysfunction (Berkowitz et al., 2012). This implies that ion channels may be an important therapeutic target for DR. Inter-neuronal communication is largely dependent on synaptic function (Pereda, 2014), and in hyperglycemia, both neurochemical (synaptic) and electrical (gap junction) modes of communication between retinal cells are affected (Eleftheriou et al., 2020). Diabetes causes a cumulative loss of neural dendrites and synapses in the inner retina, which results in significant thinning of the inner plexiform layer (IPL) (Barber et al., 1998). Hyperphosphorylated tau proteins contribute to diabetic retinal neurodegeneration by disrupting synaptic and mitochondrial function in retinal ganglion cells (RGCs) (Zhu et al., 2018). This suggests the importance of synaptic dysfunction in the pathophysiological process of DR. In DR, the blood-retinal barrier (BRB) is disrupted (Xu et al., 2011). Inflammatory cytokine levels are elevated in the vitreous fluid of patients with DR (Daruich et al., 2018), whereas pro-inflammatory cytokines have been found to cause down-regulation of tight junctions in bovine retinal endothelial cells through the NF-κB pathway, and, increase retinal endothelial cell permeability (Aveleira et al., 2010). In the mouse retina, hyperglycemic memory (HGM) induces sustained oxidative stress, mitochondrial membrane potential collapse and fission, adhesion junction disassembly and subsequent vascular leakage after normalization of blood glucose (Lee et al., 2022). The association between disruption of the blood-retinal barrier (BRB) and the down-regulation or disintegration of cellular connections is evident. Conversely, the overexpression of endothelial CYP2J2 preserves the distribution of endothelial tight junctions and adherens junctions in an ANXA1-dependent manner, thus safeguarding the integrity of the BRB and shielding retinal ganglion cells from loss following ischemia-reperfusion injury (Zhao et al., 2022).

The PLS + genes exhibited specific expression mainly in the cortex and striatum, while the PLS- genes showed predominant expression in the striatum, indicating a focus of diabetic retinopathy (DR) effects on the cerebral cortex. Moreover, the PLS + gene demonstrated particular expression in Purkinje neurons (Cb.Pcp2) located in the cerebellar cortex (Hirano, 2018), pivotal for integrating cerebellar inputs and outputs. The regular firing activity of Purkinje neurons is maintained through precise ion channel control (Raman and Bean, 1999), and ion channel dysregulation leads to perturbations in the intrinsic membrane excitability of Purkinje neurons, causing abnormal firing in Purkinje neurons (Bushart and Shakkottai, 2019; Cook et al., 2021; Huang and Shakkottai, 2023), which corresponds to the enrichment of these genes in ion channels. We did not find significant specific expression of PLS- genes at the cellular level. Furthermore, time-specific expression analyses, using predefined thresholds, did not reveal specific expression of dFCD-related genes at distinct developmental stages.

PPI analyses have revealed that dFCD-related genes can form a PPI network supported by functionally significant hub genes crucial for various biological processes. For instance, the RBX1 gene encodes RBX1 (RING box protein 1), essential for developmental processes. RBX1 suppression triggers the DNA damage response (DDR), resulting in G2-M arrest, senescence, and apoptosis (Jia et al., 2011). Another example is DDB2 (damage-specific DNA-binding protein 2), a component of the group E gene associated with dyschromic dry dermatosis, participating in the initiation of molecular ubiquitination within the ubiquitin ligase complex involving DDB1 and CUL4A (cullin 4A) (Stoyanova et al., 2009; Yan et al., 2011). DDB2 directly interacts with progestin and adipoQ receptor 3 (PAQR3), facilitating PAQR3 ubiquitination, consequently enhancing PI3K/AKT axis signaling to ameliorate insulin resistance and mitigate retinopathy in diabetic mice (Xiao et al., 2022).

Several limitations are evident in this study. Firstly, the gene expression data originated from the brains of six postmortem donors, while the neuroimaging data were obtained from 93 living participants (Zeng et al., 2012; Hawrylycz et al., 2015). To address the differential sources, we focused on genes displaying more conserved expression profiles by setting a DS threshold, facilitating our transcriptome-neuroimaging spatial correlation analysis. Secondly, due to limited gene expression data in the right hemisphere and notable variations between cortical and subcortical regions, only tissue samples from the left cerebral cortex were included, potentially introducing bias. Thirdly, oversampling in spatial correlation analyses is a common concern, which may inflate the rate of false positives. Lastly, gene identification related to dynamic functional connectivity density (dFCD) was achieved indirectly through spatial correlation analysis of transcriptional neuroimaging, lacking direct experimental validation. Further exploration and validation through animal experiments are recommended for future studies.

Our study identified altered dFCD variability in brain regions linked to visual and cognitive functions in DR patients. This was manifested by significantly lower dFCD variability in the left anterior cingulate gyrus, left supraoccipital gyrus, and right postcentral gyrus in DR group. Moreover, transcriptome-neuroimaging correlation analyses revealed a spatial association between these dFCD changes and the genes with unique functional profiles. These genes were enriched in biologically significant functions and pathways, specific to certain cells and brain areas. Our research offers novel understandings of the genetic mechanisms influencing dFCD alterations in DR. The findings are also important for the diagnosis of DR and the discovery of potential therapeutic targets for DR.