Endothelial dysfunction in human diabetic vascular complications: translating single-cell transcriptomics into therapeutic opportunities

Abstract

Type 2 diabetes mellitus (T2DM) is increasingly recognized as a systemic vascular disease in which endothelial cell (EC) dysfunction is a central driver of diabetic vascular complications (DVCs), which account for most diabetes-related morbidity and mortality. Although conventional therapies improve glycemic control and reduce some cardiovascular risk, they do not fully prevent microvascular and macrovascular injury, highlighting the need for more precise vascular-targeted strategies. Recent advances in single-cell RNA sequencing have transformed our understanding of diabetic endothelial biology by resolving EC heterogeneity at unprecedented resolution across human tissues, yet insights from human diabetic tissues remain fragmented. Here, we provide a cross-tissue synthesis of currently available human single-cell studies profiling ECs across diabetic vasculature, focusing on diabetic arteries, diabetic retinopathy, diabetic nephropathy, and diabetic foot ulcers to define mechanisms underlying dysfunctional EC states in different DVCs. We propose a unifying framework in which DVCs are driven by tissue-specific endothelial-state transitions arising from combinations of epigenetic, transcriptional, post-translational, and intercellular signaling programs, rather than by isolated pathway abnormalities. These include a pro-inflammatory, pro-fibrotic, and anti-angiogenic state in diabetic arteries; a pathological angiogenic and barrier-disruptive inflammatory state in diabetic retinopathy; a pro-fibrotic and maladaptive angiogenic/proliferative state in diabetic nephropathy; and an inflammatory, anti-angiogenic state in diabetic foot ulcers. This framework also provides a mechanistic explanation for the limited efficacy of current single-pathway therapies, including VEGF-centered approaches, and uncovers VEGF-independent mechanisms that may be therapeutically actionable for these DVCs. Importantly, this review not only highlights the need for but also proposes combination therapeutic strategies that target multiple regulatory layers within tissue-specific endothelial states. Overall, this review supports a paradigm shift toward vascular bed-specific, combinatorial, and state-directed therapeutic strategies to reprogram endothelial dysfunction in DVCs.

Graphical Abstract

Introduction

Type 2 diabetes mellitus (T2DM) is a major global health burden and is increasingly recognized as a systemic vascular disease rather than a disorder of glucose metabolism alone (Yapislar and Gurler 2024; Genitsaridi et al. 2026). Diabetic vascular complications (DVCs), encompassing both microvascular and macrovascular pathologies, are the primary contributors to morbidity and mortality in patients with diabetes (Cole and Florez 2020; Zheng et al. 2018). As global prevalence continues to rise, it places an increasing burden on healthcare systems, underscoring the urgent need for more effective therapeutic strategies. Despite the successful development of T2DM therapies, including glucagon-like peptide-1 (GLP-1) receptor agonists and sodium-glucose cotransporter-2 inhibitors (SGLT2i), the search for novel therapeutic options that can further reduce associated DVCs is a continuous effort (Nauck et al. 2021; Kunutsor et al. 2024).

Recently, endothelial cell (EC) dysfunction has received increasing attention as a key mediator in the pathogenesis of DVCs (Sloten et al. 2014). The endothelium, a single layer of ECs lining blood vessels, lymphatic vessels, and the chambers of the heart, regulates homeostasis, vasoregulation, inflammation, and thrombosis through dynamic interactions with surrounding cells (Goncharov et al. 2017). In diabetes, hyperglycemia elevates intracellular glucose-derived metabolites in ECs (Rask-Madsen and King 2007). These disturbances promote a dysfunctional endothelial phenotype characterized by NADPH oxidase activation, upregulation of adhesion molecules, uncoupling of endothelial nitric oxide synthase, increased production of endothelin-1, and an altered balance between vasodilatory and vasoconstrictive prostanoids. Collectively, these changes drive vascular inflammation, impaired perfusion, and progressive tissue injury. Despite these shared features of dysfunction, ECs are highly heterogeneous, differing in structure, gene expression, functional specialization, and proliferative capacity across vascular beds within and between organs (Zhong et al. 2025; Becker et al. 2023). This heterogeneity is maintained and accentuated in diabetes, where ECs exhibit distinct organ-specific transcriptional and functional responses to hyperglycemia (Wang et al. 2024). Resolving this heterogeneity is therefore essential to understanding tissue-specific vascular dysfunction and to developing targeted therapies.

Traditional bulk RNA sequencing approaches have provided important insights into endothelial dysfunction but are limited by signal averaging, which masks changes in rare or functionally distinct EC subpopulations (Li and Wang 2021). Single-cell RNA sequencing (scRNA-seq) and single-nucleus RNA-seq (snRNA-seq) overcome these limitations, enabling high-resolution profiling of individual EC transcriptomes, revealing rare subtypes, mapping cell–cell communication networks through computational ligand-receptor analyses, and reconstructing differentiation and state-transition trajectories, which infer how cells transition between functional states (Tang et al. 2009; Kim et al. 2023; Zhang et al. 2025). Studies in multiple organs have already uncovered organ-specific endothelial transcriptional signatures, emphasizing the power of single-cell approaches to resolve vascular heterogeneity (Kalucka et al. 2020; Paik et al. 2020; Niu et al. 2024).

Despite advances in single-cell profiling, to our knowledge, studies of human diabetic endothelial dysfunction remain largely tissue-specific, with each study focusing on a single organ or diabetic complication in isolation. While these studies have provided important insights into local mechanisms of disease, they do not permit direct comparison of endothelial responses across the diabetic vasculature. Consequently, it remains unclear which endothelial programs are shared across DVCs and which represent tissue-specific adaptations. This gap limits understanding of the common and tissue-specific mechanisms underlying endothelial dysfunction in diabetes and constrains the development of vascular bed-specific therapeutic strategies. In this review, we compile and integrate currently available scRNA-seq datasets profiling ECs from diabetic human tissues (Table 1). Through this synthesis, we define tissue-specific endothelial states, signaling pathways, and cell–cell communication programs associated with diabetic EC dysfunction across different DVCs. We show how these studies redefine diabetic endothelial dysfunction as a set of tissue-adapted endothelial states driven by multiple interacting regulatory layers (Fig. 1). We further propose an integrated cross-tissue framework that elucidates the limitations of current single-pathway-centered therapies and supports a shift toward vascular bed-specific, combinatorial, and state-directed therapeutic strategies. Through this lens, the review provides a mechanistic and translational framework for developing next-generation combinatorial therapies to target endothelial dysfunction in DVCs.

Table 1.

Human single-cell and related transcriptomic studies profiling endothelial cells in diabetes

Tissue	Diabetic complication	Modality	Platform (dataset accession)	Cohort and tissue source	Endothelial subpopulations identified†	Key endothelial findings in human diabetes	Ref
Artery	T2DM	scRNA-seq	10 × Genomics Chromium (GSE156341)	Post-mortem mesenteric arteries from T2D (n = 2) and non-diabetic (n = 2) donors	NA	↑ LINC00607	Calandrelli et al. (2020); Sriram et al. (2022)
	T2DM	scRNA-seq	10 × Genomics Chromium (GSE156341)	Post-mortem mesenteric arteries from T2D (n = 2) and a non-diabetic (n = 1) donors	NA	↓ WWP2	You et al. (2023)
	T2DM	scRNA-seq	10 × Genomics Chromium (GSE156341)	Post-mortem mesenteric arteries from T2D (n = 5) and non-diabetic (n = 5) donors	EC1, EC2	EC expression of TREM2 ligands (CLU, APP, APOE)	Malhi et al. (2024)
Retina	PDR	scRNA-seq	10 × Genomics Chromium (GSE245561)	FVM from PDR patients (n = 4)	Tip ECs, stalk ECs, mature & immature ECs	↑ VEGFC in stalk ECs; loss of tip-stalk hierarchy	Corano Scheri et al. (2023)
	PDR	Bulk RNA-seq (CD31⁺ ECs)	Illumina HiSeq 2000 (GSE94019)	FVM from PDR patients (n = 9) and non-diabetic retina controls (n = 3)	NA	↑ TUBB6	Zhang et al. (2025)
	PDR	scRNA-seq	‡ (GSE165784)	FVM from PDR (n = ‡) patients	STINGhigh ECs, STINGlow ECs	↑ STING	He et al. (2024)
	PDR	scRNA-seq	‡ (GSE165784); 10 × Genomics Chromium (GSE245561; GSE196235); Seq-Well (GSE137846)	FVM from PDR patients (n = 8) and non-diabetic retina controls (n = 11)	NA	↑ MAP4K4	Yang et al. (2025)
Kidney	DKD	snRNA-seq	10 × Genomics Chromium (GSE131882)	Kidney biopsies from DKD patients (n = 3) and non-diabetic controls (n = 3)	GEC0, GEC1	↑ ADAM17, ITGB8, TGFB2, PDGFB, PSEN1 in GEC1	Li et al. (2021)
	DKD	scRNA-seq	10 × Genomics Chromium (‡)	Kidney samples from DKD patients (n = 44) and non-DKD controls (n = 18)	NA	↑ EDN1, DLL4, TGFB1, GAS6, VEGFA	Stefansson et al. (2022)
	DKD	snRNA-seq	10 × Genomics Chromium (GSE131882)	Kidney biopsies from DKD patients (n = 3) and non-diabetic controls (n = 3)	NA	↑ VEGFC, VCAM1, NR4A1, ITGB1, MEF2C in ECs	Wilson et al. (2019)
	DKD	snRNA-seq	10 × Genomics Chromium (KPMP-HuBMAP human kidney atlas)	Kidney biopsies from DKD patients (n = 7) and non-DKD controls (n = 18)	EC-GC, prEC and dEC	Expansion of prEC state	Ferreira et al. (2026)
Skin	DFU	scRNA-seq	10 × Genomics Chromium (GSE165816)	Wound-edge skin from DFU healers (n = 7), DFU non-healers (n = 4), and non-diabetic controls (n = 10)	15 HE-VasEndo clusters, LECs	↓ CCND1, ENO1, HIF1A, SERPINE1 in HE-VasEndo of non-healers	Lu et al. (2023)
	DFU	scRNA-seq	10 × Genomics Chromium (GSE165816)	Wound-edge skin from DFU patients (n = 14) and non-diabetic controls (n = 11)	HDMECs (8 EC subclusters)	↓ RAB17	Du et al. (2023)
	DFU	scRNA-seq	10 × Genomics Chromium (GSE245703)	Skin wounds from DFU patients (n = 5) and non-diabetic wounds (n = 4)	VECs, LECs	↓ SELE, RPS4Y1, NFKBIA, C2CD4B, JUNB in VECs	Wang et al. (2024)
	Diabetic wounds	scRNA-seq	10 × Genomics Chromium (GSE182208)	Skin samples from diabetic (n = 3) and non-diabetic (n = 5) donors	ECs, LECs	↓ PLCγ2 in ECs	Rustagi et al. (2022)
	DFU	scRNA-seq	10 × Genomics Chromium (GSE165816)	Wound-edge skin from healing (n = ‡) and non-healing DFUs (n = ‡)	ECs, LECs	↑ SLFN5 in non-healing DFU ECs	Fan et al. (2024)
	DFU	scRNA-seq	10 × Genomics Chromium (GSE165816)	Skin samples from non-diabetic (n = 11), diabetic non-DFU (n = 8), healed DFUs (n = 9), and non-healing DFUs (n = 5) donors	ECs (11 clusters), LECs	↑ SH3BGRL3 in ECs	Zhao et al. (2025)
	Diabetic wounds	scRNA-seq	10 × Genomics Chromium (GSE176417, GSE268834)	Wound-edge skin from T2D (n = 5) and non-diabetic (n = 5) donors	5 EC subclusters (arterial, venous, capillary)	↓ PLCγ2	Verma et al. (2025)
	DFU	scRNA-seq	10 × Genomics Chromium (GSE165816)	Foot skin (total n = 24) from healthy, healing DFU, and non-healing DFU donors	ENTPD1⁺ ECs, ENTPD1⁻ ECs	↑ ENTPD1 associated with DFU healing	Zhang et al. (2025)

Abbreviations: DFU diabetic foot ulcer, DKD diabetic kidney disease, dEC degenerative endothelial cell, EC endothelial cell, EC-GC glomerular capillary EC, FVM fibrovascular membrane, GEC glomerular EC, HE-VasEndo healing-enriched vascular EC, HuBMAP Human BioMolecular Atlas Program, HDMEC human dermal microvascular EC, KPMP Kidney Precision Medicine Project, LEC lymphatic EC, PDR proliferative diabetic retinopathy, prEC proliferative EC, scRNA-seq single-cell RNA sequencing, snRNA-seq single-nucleus RNA-seq, T2D/T2DM type 2 diabetes mellitus, VEC vascular EC

^†Endothelial subpopulation nomenclature is retained as reported in the original studies

^‡Data not explicitly reported in the manuscript

Fig. 1.

Multi-layer regulatory mechanisms underlying tissue-specific endothelial states across diabetic vascular complications. Diabetes induces endothelial dysfunction through interacting regulatory layers, including epigenetic, transcriptional/post-transcriptional, translational/post-translational, and intercellular signaling mechanisms. The combination of these processes drives the emergence of distinct tissue-specific endothelial states across diabetic vascular complications. Created in BioRender. Alshehhi, R. (2026) https://BioRender.com/gqrdfuk

EC dysfunction in diabetic arteries

Macrovascular complications affect large blood vessels, including the aorta, coronary arteries, and arteries supplying blood to the brain and limbs. These complications contribute to coronary artery disease, peripheral artery disease (PAD), and cerebrovascular complications, which collectively drive morbidity and mortality in T2D (Plutzky 2011). Because ECs regulate vascular homeostasis, their dysfunction provides a unifying mechanism underlying these complications and represents a key therapeutic target for macrovascular disease in diabetes. Emerging single-cell studies of human diabetic mesenteric arteries indicate that arterial endothelial dysfunction is not driven by a single molecular abnormality but arises through the convergence of multiple interconnected regulatory layers, including epigenetic reprogramming, post-translational dysregulation, and pathological immune-endothelial communication. Together, these mechanisms reinforce one another to shift ECs from a homeostatic state toward a persistent pro-inflammatory, pro-fibrotic, and angiogenesis-impaired endothelial state.

LINC00607 long non-coding RNA drives profibrotic endothelial dysfunction

Epigenetic modifications are among the earliest regulatory mechanisms that contribute to the establishment of the dysfunctional arterial endothelial state in diabetes. These include chromatin-associated RNAs (caRNAs), which act as enhancer-embedded regulators that shape transcriptional programs driving EC dysfunction (Calandrelli et al. 2020). A notable example is LINC00607, a super-enhancer (SE)-derived long non-coding RNA identified through scRNA-seq in arteries from T2DM and non-T2DM donors (Sriram et al. 2022). LINC00607 is upregulated in T2DM ECs and exhibits artery-specific expression, supported by histological validation and large-scale transcriptomic datasets such as ENCODE and GTEx. LINC00607-associated DEGs are enriched for pathways related to ECM organization, inflammation, wound healing, and angiogenesis, implicating it in basal EC functions. Consistent with these roles, loss of LINC00607 in human umbilical vein ECs (HUVECs) impairs vascular endothelial growth factor A (VEGFA)-driven angiogenic sprouting and reduces endothelial migration and proliferation, confirming its requirement for maintaining normal angiogenic capacity (Boos et al. 2023). Moreover, inhibition of nuclear LINC00607 in high-glucose and TNFα-treated (HT) HUVECs reduces monocyte adhesion and downregulates the pro-inflammatory, pro-fibrotic gene SERPINE1 (Calandrelli et al. 2020). The frequent co-expression of LINC00607 and SERPINE1 observed in both HT-treated HUVECs and diabetic donor-derived ECs likely reflects RNA-chromatin interactions observed between LINC00607 and SERPINE1SE (Fig. 2a). SERPINE1 encodes PAI-1, a key inhibitor of fibrinolysis and ECM degradation that promotes fibrosis (Schneider and Sobel 2012). Notably, although normalizing glucose partially restores eNOS and ICAM expression, it fails to reverse SERPINE1 or other inflammatory and ECM-remodeling genes (Calandrelli et al. 2020), which remain responsive to LINC00607 knockdown. This finding highlights a role for LINC00607 in metabolic memory, whereby prior hyperglycaemic exposure produces persistent pathogenic gene-expression programs despite subsequent glucose normalization (Sriram et al. 2022). Collectively, these findings identify LINC00607 as an epigenetic regulator that stabilizes a pro-inflammatory and pro-fibrotic arterial endothelial state. By sustaining pathogenic transcriptional programs despite glucose normalization, LINC00607 may contribute to the persistence of endothelial dysfunction and impaired vascular repair in diabetic arteries. Targeting LINC00607 could therefore represent a strategy to reprogram this maladaptive endothelial state in combination with conventional glucose-lowering therapies.

Fig. 2.

a Diabetic stress induces c-Myc, leading to transcriptional upregulation of the super-enhancer-derived lncRNA, LINC00607. LINC00607 participates in stress-induced RNA-chromatin interactions at the SERPINE1 super-enhancer, thereby promoting SERPINE1 transcription and increasing PAI-1 protein levels, which inhibit fibrinolysis and ECM degradation. b Diabetic conditions activate JNK in ECs, resulting in suppression of the E3 ubiquitin ligase, WWP2. Normally, WWP2 promotes proteasomal degradation of DDX3X via K63-linked polyubiquitination. When WWP2 is lost, DDX3X accumulation drives endothelial injury. c Diabetes stress increases TREM2 expression in mononuclear phagocytes while concomitantly elevating endothelial expression of the corresponding ligands CLU, APP, and APOE. This reciprocal signaling strengthens EC-mononuclear phagocyte interactions, impairing EC migration, wound healing, and endothelial angiogenic potential. Black arrows indicate physiological pathways; red arrows indicate diabetic conditions. Created in BioRender. Alshehhi, R. (2026) https://BioRender.com/8v1ir93

WWP2 loss promotes endothelial injury

Beyond epigenetic regulation, disruption of post-translational protein control represents a second regulatory layer that contributes to the establishment of the dysfunctional arterial endothelial state in T2D. E3 ubiquitin ligases maintain protein homeostasis by selectively tagging proteins for degradation, and disruption of this process contributes to endothelial dysfunction. Arterial scRNAseq data show reduced expression of WWP2, an endothelial E3 ubiquitin ligase, in human T2DM arteries (You et al. 2023). Under high glucose plus palmitate (HG/PA) conditions, JNK activation in HUVECs downregulates WWP2, preventing WWP2-mediated K63-linked polyubiquitination and proteasomal degradation of the helicase DDX3X, thereby permitting DDX3X accumulation and endothelial injury (Fig. 2b). Consistently, endothelial-specific Wwp2 knockout in high-fat diet/streptozotocin-induced mice exacerbates endothelial injury and promotes vascular remodeling, confirming the protective role of WWP2 in ECs. These findings are reinforced by inhibition of Wnt5a-JNK signaling, which enhances nitric oxide generation in ECs from diabetic patients (Bretón-Romero et al. 2016), supporting the concept that JNK-driven suppression of WWP2 impairs EC function. Additionally, endothelial/myeloid WWP2 deletion worsens angiotensin II-induced oxidative stress in mice (Zhang et al. 2020). Taken together, these observations demonstrate WWP2 loss as a post-translational mechanism that reinforces the dysfunctional arterial endothelial state by promoting endothelial injury and impairing vascular homeostasis. Consequently, the JNK-WWP2-DDX3X axis represents a potential therapeutic target for restoring endothelial function in diabetic macrovascular disease.

TREM2-dependent immune-endothelial cross-talk impairs angiogenic function

While epigenetic and post-translational defects reveal intrinsic vulnerabilities in diabetic ECs, endothelial dysfunction is further amplified by pathological interactions with immune cells. Ligand-receptor mapping of T2D arterial cells demonstrates increased expression of the myeloid receptor TREM2 in mononuclear phagocytes, alongside increased expression of its ligands (CLU, APP, and APOE) in diabetic ECs, indicating enhanced immune-endothelial signaling (Fig. 2c) (Malhi et al. 2024). This signaling configuration promotes pathological immune-endothelial engagement as reflected by increased adhesion of TREM2^high myeloid cells isolated from the spleens of T2D donors to HUVECs. Moreover, conditioned media from TREM2-expressing macrophages exposed to diabetic stimuli induce pro-inflammatory adhesion molecules in human microvascular ECs and impair their migratory capacity, an effect abolished by TREM2 knockdown. Consistent with these findings, Trem2 neutralization improves post-ischemic perfusion and neovascularization in diabetic mice, highlighting its pathogenic role in diabetic vasculopathy and PAD. However, TREM2 function is context-dependent across tissues. In metabolic tissues, Trem2 overexpression promotes insulin resistance in high-fat diet models (Sharif et al. 2021), whereas Trem2 blockade suppresses adipogenesis and reduces weight gain (Park et al. 2015). In contrast, in diabetic kidney disease (DKD), Trem2 knockout exacerbates glomerular injury and tubular epithelial damage, suggesting a protective role in the kidney (Subramanian et al. 2024). Overall, these observations identify TREM2 as an amplifier of pathological immune-endothelial communication that reinforces the pro-inflammatory and angiogenesis-impaired arterial endothelial state. However, the divergent effects of TREM2 across metabolic tissues and the kidney highlight the need for vascular-bed-specific therapeutic strategies when targeting this pathway.

In summary, single-cell studies indicate that endothelial dysfunction in diabetic macrovascular disease arises through the convergence of epigenetic, post-translational, and immune-endothelial regulatory mechanisms. Together, these processes drive a persistent pro-inflammatory, pro-fibrotic, and angiogenesis-impaired arterial endothelial state. These findings support therapeutic strategies aimed at restoring arterial endothelial homeostasis through state-directed reprogramming rather than targeting individual pathways in isolation. Such approaches may include combining epigenetic interventions targeting LINC00607, restoration of WWP2-dependent protein homeostasis, and modulation of pathological TREM2-mediated immune-endothelial signaling to improve vascular repair and limit macrovascular disease progression.

While diabetic macrovascular disease is characterized by a pro-inflammatory, pro-fibrotic, and angiogenesis-impaired endothelial state, endothelial dysfunction in microvascular complications manifests differently depending on the affected tissue. This is exemplified in diabetic retinopathy (DR), where diabetic stress promotes excessive pathological angiogenesis and a barrier-disruptive endothelial cell state, rather than the impaired reparative angiogenesis observed in diabetic arteries.

EC dysfunction in diabetic retinopathy

DR is a microvascular complication of diabetes mellitus categorized into non-proliferative and proliferative stages (Sivaprasad et al. 2023). Proliferative DR (PDR) is characterized by pathological retinal angiogenesis driven by ischemia and hypoxia, resulting in hemorrhage, retinal detachment, and irreversible vision loss (Zhang et al. 2022). The newly formed vessels often extend into the vitreous and can form epiretinal fibrovascular membranes (FVMs) at the vitreoretinal interface. These pathological lesions are composed of ECs, myofibroblasts, inflammatory cells, glial cells, and extracellular matrix and contribute to retinal traction and vision loss (Kim et al. 2015). Inflammation caused by hyperglycemia activates glial cells and induces inflammatory cytokine production, leading to EC dysfunction, destruction of intercellular junctions, and impairment of the inner blood-retinal barrier (iBRB), thereby causing vascular leakage (Yang and Liu 2022; Tang et al. 2023). Although anti-VEGF therapies targeting the pro-angiogenic molecule VEGF have transformed DR management, many patients do not respond adequately to this treatment (Arima et al. 2020; Antonetti et al. 2021), suggesting that alternative pro-angiogenic pathways contribute to DR. Human single-cell studies suggest that retinal endothelial dysfunction is not driven by a single angiogenic pathway but arises through the convergence of multiple regulatory processes, including disruption of endothelial-state specification, activation of pathological angiogenic programs, and inflammatory pathways that compromise barrier integrity. Together, these mechanisms shift ECs from a homeostatic state toward a pathological angiogenic and barrier-disruptive inflammatory endothelial state. Given their pivotal role in retinal angiogenesis, ECs represent key therapeutic targets in PDR, and scRNA-seq now enables high-resolution dissection of their molecular heterogeneity, offering novel insights into disease mechanisms and potential druggable pathways.

Aberrant tip- and stalk-cell identities promote pathological angiogenesis

Disruption of endothelial-state specification represents an early mechanism contributing to the pathological angiogenic endothelial state in PDR. Under physiological conditions, tip cells (COL4A2, MCAM, TP53I11, ESM1, ANGPT2) lead vessel sprouting, while stalk cells (IGFBP3, AQP1, PLPP1, EFNB2, ADAMTS1) proliferate to elongate the sprout (Del Toro et al. 2010; Zhao et al. 2018; Corano Scheri et al. 2023). The coordinated interaction between tip and stalk cells is essential for normal angiogenesis, whereas disruption of these programs promotes endothelial dysfunction and consequently pathological neovascularization in diabetic vasculopathy. Analyses of PDR FVMs identified disruption of the distinct transcriptional identities of tip and stalk ECs (Corano Scheri et al. 2023). Specifically, stalk cells aberrantly express tip-cell genes (ESM1, ANGPT2), while tip cells acquire stalk-cell markers (IGFBP3, ETS1), and lose expression of canonical tip markers such as PLAUR (Fig. 3a). In addition to this transcriptional mixing, VEGFC expression, typically involved in lymphangiogenesis (Karkkainen et al. 2004), is ectopically expressed in stalk cells, with immunofluorescence confirming its localization in CD31⁺ ECs in PDR FVMs (Corano Scheri et al. 2023). Beyond lymphangiogenesis, VEGFC promotes neovascularisation in the mouse cornea and, in porcine aortic ECs, leads to VEGFR2/VEGFR3-dependent actin reorganization that is essential for angiogenic sprouting (Cao et al. 1998). Ligand-receptor analyses further identify VEGFC-mediated signaling between ECs and myofibroblasts, suggesting a mechanism by which pathological EC states contribute to FVM proliferation (Corano Scheri et al. 2023). Collectively, these discoveries highlight VEGFC signaling as a potential therapeutic target in PDR. Although direct clinical evidence in PDR is lacking, insights from related neovascular conditions may be informative. For example, in a phase 2b clinical trial in neovascular age-related macular degeneration (NCT03345082), inhibition of VEGF-C/D through OPT-302 trap, in combination with ranibizumab (anti-VEGFA), improved superior visual and anatomical outcomes compared with standard anti-VEGFA therapy (Jackson et al. 2023). These findings identify disruption of tip- and stalk-cell identities as a mechanism that promotes the pathological angiogenic endothelial state in PDR. Therapeutically, targeting VEGFC may complement existing therapies by restoring endothelial-state specification and limiting pathological neovascularization.

Fig. 3.

Molecular pathways underlying the pathological angiogenic and barrier-disruptive endothelial state in diabetic retinopathy. a In a healthy retina, EC tip and stalk cells display distinct transcriptional programs that coordinate ordered sprouting angiogenesis. Single-cell RNA-sequencing analyses of fibrovascular membranes from PDR patients reveal the mixing of tip- and stalk-associated signatures. b Under diabetic stress conditions, YBX1 binds to the 3′UTR of TUBB6 mRNA and stabilizes it, leading to increased TUBB6 expression. Elevated TUBB6 enhances WNT3A production and activates endothelial WNT3A-FZD8 signaling, promoting angiogenic endothelial phenotypes. c Diabetic stress activates endothelial pathways that destabilize the inner blood-retinal barrier. Diabetic stress induces CREB activation and mitochondrial dysfunction, leading to mitochondrial DNA leakage. Cytosolic DNA is sensed by cGAS, resulting in the production of 2′3′-cGAMP and activation of STING. Activated STING activates TBK1, leading to phosphorylation of IRF3 and NF-κB and induction of interferons and pro-inflammatory cytokines that disrupt tight and adherens junction proteins. These effects are attenuated by GLP-1 receptor agonists. Additionally, diabetes increases MAP4K4 expression via reduced YTHDF2-mediated m⁶A-dependent mRNA decay, stabilizing MAP4K4 transcripts. Elevated MAP4K4 activates NF-κB, induces MMP2 and MMP9, and drives junctional protein loss and therefore retinal vascular leakage. Black arrows indicate physiological pathways; red arrows indicate diabetic conditions. Created in BioRender. Alshehhi, R. (2026) https://BioRender.com/sz6bbv3

TUBB6-WNT signaling promotes pathological angiogenesis

Beyond disruption of tip- and stalk-cell identities, activation of pro-angiogenic transcriptional programs represents a second layer that reinforces the pathological angiogenic endothelial state in PDR. Microtubule dynamics, which regulate cell shape and migration, have emerged as novel regulators of this process. TUBB6, encoding tubulin beta-6 (Maurin et al. 2021), was found by RNA expression profiling to be significantly upregulated in PDR and nAMD patients’ retinal microvascular ECs (Zhang et al. 2025). Similarly, Tubb6 expression is increased in retinas from oxygen-induced retinopathy (OIR) mice, particularly in tip and stalk ECs, supporting a role in angiogenic EC states. Functional studies further show that Tubb6 knockdown reduces sprouting angiogenesis and impairs tip cell formation in the retinas of OIR mice, confirming its role in pathological angiogenesis. Mechanistically, hypoxia-induced YBX1 stabilizes the TUBB6 mRNA (in vitro), increasing protein expression and promoting the upregulation of both WNT3A and FZD8. The binding of WNT3A ligand to FZD receptors activates canonical Wnt/β-catenin signaling, resulting in β-catenin stabilization and nuclear translocation, where it induces the expression of pro-angiogenic mediators such as VEGF and promotes endothelial proliferation, migration, sprouting, and tip cell specification (Fig. 3b) (Zhang et al. 2025; Chen and Ma 2017). This signaling axis is consistent with single-cell findings in PDR, where WNT-related genes (TSPAN12, LRP5, LRP6) are upregulated in FVM ECs. In line with these observations, inhibition of WNT signaling reduces retinal inflammation, vascular leakage, and neovascularisation in DR models (Chen et al. 2009). Collectively, these findings identify the TUBB6-WNT axis as a transcriptional regulator that reinforces the pathological angiogenic endothelial state in DR by promoting endothelial proliferation, migration, and neovascularization. Given its upstream role in regulating multiple pathogenic angiogenic processes, inhibition of the TUBB6-WNT3A/FZD8 signaling axis may offer broader therapeutic benefits in DR treatment than current anti-VEGF therapy.

STING and MAP4K4 signaling disrupt the inner blood-retinal barrier

While dysregulated angiogenesis is a hallmark of advanced DR, the pathological retinal endothelial state is further reinforced by inflammatory pathways that disrupt inner blood-retinal barrier integrity. Stimulator of interferon genes (STING) is abundantly expressed in ECs from PDR patients as identified through single-cell transcriptomic profiling (He et al. 2024). Diabetic stress in human retinal vascular ECs (HRVEC) activates CREB, which results in mitochondrial DNA leakage and STING activation, thereby increasing TBK-1/NF-κB/IRF3-driven inflammatory signaling and loss of endothelial junctional protein integrity (ZO-1, VE-cadherin, and claudin-5), promoting vascular leakage (Fig. 3c). These findings are consistent with observations in PDR FVM, where tight junction proteins such as claudin-1 and claudin-12 are absent or reduced in ECs, respectively (Corano Scheri et al. 2023). Since claudin-1 and claudin-5 are major claudin proteins in the BRB, their loss contributes to the formation of leaky BRBs (Luo et al. 2011). Importantly, these pathogenic effects are attenuated by treatment with GLP-1 receptor agonists (GLP-1RA), which suppress STING signaling and preserve endothelial junctional integrity in HRVECs (He et al. 2024). Supporting this mechanism, GLP-1RA treatment improves retinal vascular integrity in diabetic mouse models and is associated with a lower incidence of DR in patients with T2DM compared with other glucose-lowering therapies (He et al. 2024). Overall, these findings identify STING signaling as a key inflammatory mechanism contributing to the barrier-disruptive component of the pathological retinal endothelial state. The ability of GLP-1RAs to suppress this pathway further highlights the therapeutic potential of targeting endothelial inflammation to preserve vascular integrity in DR.

Although GLP-1RAs partially preserve endothelial junctions by suppressing STING signaling, persistent vascular leakage in DR indicates the involvement of additional endothelial pathways that independently regulate barrier integrity. Among these, the proinflammatory kinase MAP4K4 has emerged as a key regulator of intercellular junctions and endothelial barrier function, with potential relevance to multiple disorders associated with endothelial dysfunction (Zou et al. 2021). MAP4K4 expression is specifically increased in ECs of PDR FVM compared with non-diabetic patients, a finding validated at the protein level (Yang et al. 2025). Mechanistically, MAP4K4, regulated by the m⁶A reader, YTHDF2, activates the NF-κB signaling pathway, leading to the upregulation of matrix metalloproteinase (MMP) expression and subsequent downregulation of endothelial junctional proteins within the iBRB in primary human retinal microvascular ECs (HRMECs) and in vivo diabetic mouse models. Inhibition of MAP4K4 with DMX-5804 reduces vascular leakage and pathological neovascularisation in diabetic mouse retinas without impairing physiological endothelial function, supporting its therapeutic potential in DR. Although DMX-5804 shows limited efficacy for cardioprotection in large-animal models (Lintel et al. 2021), its ability to alleviate retinal microvascular leakage highlights its potential for repurposing to target microvascular complications of DR (Yang et al. 2025). Together, these findings identify MAP4K4 as a regulator that reinforces the barrier-disruptive inflammatory endothelial state by promoting junctional instability and vascular leakage.

In summary, single-cell studies indicate that endothelial dysfunction in DR arises through the convergence of disrupted endothelial-state specification, pathological angiogenic signaling, and inflammatory pathways that compromise barrier integrity. These processes drive a pathological angiogenic and barrier-disruptive inflammatory retinal endothelial state that promotes neovascularization, vascular leakage, and disease progression. This framework also provides a mechanistic explanation for the heterogeneous responses to anti-VEGF therapy, highlighting the potential value of combination strategies that simultaneously target pathological angiogenesis and endothelial barrier dysfunction.

In contrast to DR, where endothelial dysfunction is characterized by excessive pathological neovascularization and barrier breakdown, endothelial dysfunction in diabetic nephropathy (DN) is driven by fibrosis, maladaptive endothelial-glomerular cross-talk, and maladaptive angiogenesis that progressively compromise renal structure and function.

EC dysfunction in diabetic nephropathy

DKD or DN is a major microvascular complication of diabetes mellitus and the leading cause of end-stage renal disease worldwide (Liu et al. 2025). Early pathological changes include thickening of the glomerular basement membrane, expansion of the mesangial matrix, arteriolar hyalinosis, and, ultimately, the development of Kimmelstiel-Wilson nodules, podocyte loss, and glomerulosclerosis in advanced stages (Anders et al. 2018; Najafian et al. 2011). These structural alterations compromise the glomerular filtration barrier, leading to albuminuria, a hallmark clinical feature that often reflects endothelial and podocyte dysfunction (Thomas et al. 2015). Chronic exposure to elevated glucose, lipids, and inflammatory mediators makes glomerular ECs (GECs) highly vulnerable to diabetic injury, positioning them as key contributors to DKD pathogenesis and potential therapeutic targets (Chen et al. 2020). Current single-cell studies indicate that glomerular endothelial dysfunction is not limited to endothelial injury alone but involves pathological interactions between GECs and neighboring glomerular cells, together with endothelial reprogramming that alters angiogenic behavior. These processes contribute to a pro-fibrotic and maladaptive angiogenic/proliferative endothelial state in DKD.

GECs cross-talk associated with podocyte loss and mesangial expansion

Although GECs are among the earliest glomerular cell types affected by diabetic metabolic stress, the progression of DKD is not driven solely by endothelial injury. Instead, GECs actively shape disease progression through pathological crosstalk with neighboring glomerular cells, including podocytes and mesangial cells. ADAM17 expression is increased in GECs from patients with DN, a finding supported with immunohistochemical validation in human diabetic kidneys (Li et al. 2021; Lattenist et al. 2016). Ligand-receptor analysis further identified upregulated ERBB4 on podocytes as a predicted target of endothelial ADAM17 signaling, supporting an active endothelial-podocyte axis (Fig. 4a) (Li et al. 2021). Consistent with this mechanism, ERBB4 is increased in human and mouse diabetic glomeruli and is induced by TGF-β1 in podocytes (Lee et al. 2017). Pharmacological inhibition of ErbB signaling with the pan-ErbB inhibitor erlotinib attenuates murine podocyte injury in vitro and reduces albuminuria in diabetic mouse models, hallmarks of DKD. Collectively, these studies identify the ADAM17-ERBB4 axis as a mechanism through which pathological signals from GECs are propagated to podocytes in DKD.

Fig. 4.

Mechanisms driving the pro-fibrotic and maladaptive angiogenic/proliferative endothelial state in diabetic nephropathy. a Human single-cell RNA sequencing analysis predicts ligand-receptor interactions mediating communication between GECs and neighboring podocytes, mesangial cells, immune cells, fibroblasts, and lymphocytes. “Send” denotes ligands expressed by GECs that signal to other cell types, whereas “Receive” denotes ligands expressed by other cell types that signal to GECs. In each ligand-receptor pair, the first molecule represents the ligand and the second the cognate receptor. Blue text indicates upregulated ligand-receptor interactions, while red text indicates downregulated interactions in DKD. b In healthy glomeruli, pro-angiogenic and anti-angiogenic signaling is balanced in ECs. The MEF2A-MEF2C-TRPS1 transcription factor network maintains balanced angiogenesis by regulating transitions between EC-GCs, prECs, and dECs. In DKD, hyperglycemia disrupts this balance by maintaining anti-angiogenic signaling and promoting pro-angiogenic pathways (thicker arrow). This imbalance drives pathological neovascularization. SGLT2i can partially restore MEF2C network activity and shifts endothelial signaling toward a more physiological balance. Created in BioRender. Alshehhi, R. (2026) https://biorender.com/msa2hw9

In parallel with podocyte injury, GECs contribute to mesangial expansion by transmitting profibrotic signals that promote mesangial cell activation, proliferation, and excessive ECM deposition, leading to renal fibrosis and eventually glomerulosclerosis (Hu et al. 2024). Consistent with GECs acting as drivers of profibrotic signaling, single-cell ligand-receptor analysis identified an endothelial-to-mesangial signaling axis in DKD, in which GECs secrete ligands such as endothelin-1 (EDN1), DLL4, TGFB1, Gas6, and VEGFA (Stefansson et al. 2022). These ligands engage corresponding mesangial receptors (EDNRA, NOTCH4, TGFBR2, AXL, and NRP1), activating pathways that promote cytoskeletal remodeling, angiogenic signaling, and mesangial matrix expansion. Functional studies in murine system support this axis, showing that Edn1 knockout in GECs or blockade of EDN1/EDNRA signaling in mesangial cells attenuated mesangial proliferation and collagen IV production in vitro (Zou et al. 2019). Clinically, inhibition of EDNRA with atrasentan reduces albuminuria in patients with diabetes and CKD, highlighting the translational relevance of this pathway (NCT00920764) (Andress et al. 2012). Together, these findings suggest that therapeutically targeting endothelial-to-mesangial cross-talk may represent a promising strategy for limiting fibrosis and slowing DKD progression.

Overall, single-cell studies position GECs as central signaling hubs in DKD, coordinating pathological communication with neighboring glomerular cells. Importantly, endothelial-podocyte and endothelial-mesangial signaling converge to drive DKD progression through complementary mechanisms. Endothelial-mediated podocyte injury disrupts the glomerular filtration barrier, leading to albuminuria, while endothelial-derived profibrotic signals promote mesangial expansion and ECM accumulation. Progressive mesangial matrix expansion further distorts glomerular architecture and exacerbates filtration barrier dysfunction, thereby accelerating fibrosis and amplifying proteinuria. These interconnected processes couple albuminuric injury with structural remodeling of the glomerulus, ultimately promoting DKD progression. Targeting endothelial-derived signaling axes, including ADAM17-ERBB4 and EDN1-EDNRA pathways, therefore represents a potential strategy to mitigate proteinuria and limit fibrosis in DKD.

Aberrant neovascularisation associated with intercellular cross-talk and MEF2C signaling

While pathological signaling from GECs promotes podocyte loss and mesangial expansion, the resulting fibrotic microenvironment feeds back onto GECs, reshaping their phenotypic and functional states. Early renal fibrosis, driven by mesangial matrix expansion and podocyte injury, is thought to induce angiogenesis as an initial adaptive response to preserve tissue perfusion and delay further fibrotic progression (Hu et al. 2024). In line with this concept, angiogenic activity in GECs appears to be influenced by mesangial-derived signals in DN (Wilson et al. 2019). Predicted intercellular signaling interactions identify factors such as SLIT3 and CCN1 as mediators of endothelial activation through their respective receptors, ROBO2 and ITGB3. In vitro functional studies corroborate these predictions, demonstrating that SLIT3 enhances rat endothelial progenitor cell migration (Dou et al. 2018), and CCN1 promotes HUVECs sprouting and migration (Park et al. 2019), linking intercellular signaling to angiogenic endothelial activation in DKD. Collectively, these data indicate that fibrotic remodeling drives a feedback loop in which intercellular mesangial-derived signals reprogram GECs toward angiogenic states.

Although intercellular signaling from neighboring glomerular cells activates angiogenic programs in GECs, this response is not uniformly adaptive. In DKD, angiogenesis becomes dysregulated as pro-angiogenic signaling is maintained while anti-angiogenic regulatory pathways are lost (Fig. 4b). This imbalance reflects intrinsic endothelial reprogramming that contributes to neovascularisation and thereby potentially exacerbates fibrotic remodeling. In particular, a transcription factor network involving MEF2C, MEF2A, and TRPS1 has been identified in DKD and is predicted to regulate SEMA6A and PLXNA2 (Ferreira et al. 2026), an anti-angiogenic ligand receptor pair (Segarra et al. 2012; Toledano et al. 2023). In silico studies suggest that this network shifts GECs towards proliferative (prEC) or death (dEC) states, whereas in DKD the balance shifts towards prEC (Ferreira et al. 2026). This was evidenced by glomeruli with high MEF2C activity and prEC abundance, which were linked to histological neovascularization and nodular mesangial sclerosis. Independent scRNA-seq datasets further support expansion of angiogenic GEC populations in DN compared with healthy controls (Li et al. 2021). Importantly, scRNA-seq analysis of a human DKD cohort demonstrated that the dysregulated MEF2C gene regulatory network in EC-GCs can be reversed by SGLT2i treatment (Ferreira et al. 2026). Overall, these findings identify the MEF2C regulatory axis as a key determinant of the maladaptive angiogenic/proliferative endothelial state in DKD and a potential therapeutic target to limit pathological glomerular remodeling.

Taken together, single-cell studies indicate that DKD is sustained by a vicious cycle in which pathological endothelial-glomerular communication drives podocyte injury and mesangial expansion, while the resulting fibrotic microenvironment feeds back onto GECs to promote maladaptive angiogenic reprogramming. Collectively, these processes reinforce a pro-fibrotic and maladaptive angiogenic/proliferative endothelial state that drives progressive glomerular remodeling and renal dysfunction. Therapeutically, strategies that simultaneously target endothelial cross-talk and angiogenic reprogramming may be more effective than single-pathway interventions for slowing DKD progression.

These findings further support the concept that DVCs are driven by distinct tissue-adapted endothelial states. Whereas DKD is characterized by endothelial-mediated fibrosis, pathological endothelial-glomerular cross-talk, and maladaptive angiogenesis, diabetic foot ulcers (DFUs) are dominated by impaired reparative angiogenesis and failure of endothelial state transitions required for effective wound healing.

EC dysfunction in diabetic foot ulcers

DFUs are among the most common and debilitating complications of T2DM, characterized by chronic non-healing due to impaired wound repair in the diabetic milieu (Dawi et al. 2025). Impaired wound healing arises from the convergence of multiple physiological disturbances, including defective angiogenic responses, altered macrophage function, reduced epidermal innervation, and impaired keratinocyte and fibroblast migration (Brem and Tomic-Canic 2007). Among these factors, insufficient angiogenesis represents a barrier to effective tissue repair, with vascular ECs (VECs) being key contributors (Weng et al. 2025). Importantly, scRNA-seq data demonstrate comparable EC numbers in healing and non-healing DFUs, indicating that differences in wound repair arise from functional alterations in ECs rather than changes in EC abundance (Lu et al. 2023; Du et al. 2023). This insight highlights the significance of identifying endothelial state-specific dysfunctions that underlie impaired angiogenesis and chronic non-healing in diabetic wounds.

Failure of endothelial state transitions impairs healing

To define the endothelial mechanisms underlying these divergent healing outcomes, recent scRNA-seq studies have focused on resolving endothelial state heterogeneity in non-healing versus healing DFUs. Within diabetic wound samples, scRNA-seq studies have segregated ECs into VECs and lymphatic ECs (LECs), although LECs remain poorly characterized, VECs have been divided into clusters (Lu et al. 2023; Theocharidis et al. 2020; Wang et al. 2024; Verma et al. 2025; Rustagi et al. 2022; Fan et al. 2024; Zhao et al. 2025). In non-healing DFU, clustering indicates that ENTPD1- VEC (Zhang et al. 2025) and EC-5 VEC (Zhao et al. 2025) predominate, while in healing DFU, healing-enriched vascular EC (HE-VasEndo) (Lu et al. 2023), ENTPD1 + VEC (Zhang et al. 2025), EC-8 VEC (Zhao et al. 2025) are enriched (Fig. 5a). ECs in non-healing DFUs exhibit impaired immunoregulatory capacity, as evidenced by reduced Th1, Th2, and Th17 differentiation, as well as suppression of the antigen processing and presentation pathway (Lu et al. 2023). In contrast, ECs from healing ulcers show transcriptional enrichment of angiogenic, immune, and inflammatory pathways, including AGE-RAGE, focal adhesion, PI3K/Akt/mTOR, Relaxin, IL-17, TNF, NF-κB, and HIF-1 signaling (Zhang et al. 2025; Lu et al. 2023). These transcriptional differences are reflected functionally, as human dermal microvascular ECs (HDMECs) isolated from DFU patients exhibit impaired tube formation compared with healthy controls, confirming reduced angiogenic capacity in DFU (Du et al. 2023). This defect is associated with the downregulation of the angiogenesis-related gene RAB17 in DFU, as identified by scRNA-seq and bulk RNA-seq analyses. Consistent with this angiogenic defect, non-healing DFUs exhibit significant downregulation of key angiogenic and inflammatory genes, including CCND1, ENO1, HIF1α, and SERPINE1, at both transcriptomic and histological levels (Lu et al. 2023). Trajectory analyses further revealed that ECs normally transition from an early inflammatory (e.g., EC-5 VEC) state to a later angiogenic (e.g., EC-8 VEC) state during successful healing (Zhao et al. 2025). In non-healing ulcers, this transition is disrupted, leading to the persistence of inflammatory endothelial states and the loss of angiogenic programs.

Fig. 5.

Mechanisms driving the inflammatory and anti-angiogenic endothelial state in diabetic foot ulcers. a Single-cell RNA sequencing of human DFUs revealed that non-healing DFUs exhibit immune-suppressed, pro-inflammatory endothelial phenotypes with reduced angiogenic signaling, supported by enrichment of non-healing EC subpopulations, impaired endothelial communication and downregulation of genes required for angiogenesis. In contrast, healing DFUs display immune-competent, pro-angiogenic endothelial states, characterised by dominance of EC subsets that promote angiogenesis, strengthened interactions with keratinocytes, and immune cells, and increased expression of pro-angiogenic genes. Arrow thickness represents the relative increase or decrease in cell–cell communication strength. b In non-healing DFUs, SLFN5 is upregulated in ECs and is activated by the AP-1 family member c-Fos upon diabetic stress. Elevated SLFN5 suppresses PI3K/Akt signaling, thereby reducing NRF2 nuclear translocation and weakening antioxidant defense. As a result, ECs exhibit increased oxidative stress, and activation of apoptotic and pyroptotic pathways, resulting in reduced angiogenesis. c Hypermethylation of PLCγ2 promoter results in reduced PLCγ2 expression in diabetic wounds. When PLCγ2 is at physiological levels, it acts as a downstream effector of VEGF signaling, triggering MAPK44/42 activation, stabilizing HIF-1α, and facilitating its nuclear translocation, thereby increasing endogenous VEGF production and establishing a positive feedback loop that amplifies pro-angiogenic signaling. Loss of endothelial PLCγ2 impairs VEGF-dependent metabolic activation and diminishes angiogenic responsiveness. Black arrows indicate physiological pathways; red arrows indicate diabetic conditions.Created in BioRender. Alshehhi, R. (2026) https://BioRender.com/s88f45i

This failure of endothelial state transition is further reinforced by impaired intercellular communication within the wound microenvironment. In healing DFUs, VECs both respond to and generate signals that promote angiogenesis and tissue repair. In contrast, receptor-ligand interactome analyses indicate that VECs in non-healing DFU receive fewer pro-angiogenic signals from immune and epithelial ligands from B lymphocytes (CXCL8, CD44, CCL5), keratinocytes (VEGFA, TNFRSF10A/B, SEMA4A, PLD2, NRP2), and M1 macrophages (CXCL1, HLA-C, NAMPT) (Lu et al. 2023). Similarly, VECs in non-healing ulcers transmit fewer regulatory signals to B lymphocytes (FAM3C, FLT1, LGALS9, MIF, NRP1, SELE) and keratinocytes (NRP1/2, NOTCH1, LGALS9, FLT1), impairing coordinated immune, epithelial, and endothelial responses required for wound repair. Fibroblast-EC cross-talk provides an additional layer of regulation that influences endothelial fate decisions. Ligand-receptor analyses indicate that fibroblast-derived cues can bias ECs toward either non-healing EC-5 states (via SLIT3, APOE, SERPINF1, JAG1) or healing-associated EC-8 states (via CXCL12, COL14A1, COL6A6) (Zhao et al. 2025). This cross-talk is associated with the activation of downstream targets, including KRT14, KRT17, KRT6A, MT2A, and CXCR4, which support endothelial adhesion, migration, and vascular remodeling, all of which are essential for the revascularization of ischemic tissue. Collectively, these findings highlight that endothelial dysfunction in DFUs arises from both intrinsic defects in angiogenic competence and disrupted intercellular communication within the wound niche.

Taken together, non-healing DFUs are characterized by ECs trapped in inflammatory states that cannot transition to angiogenic programs. This defect, compounded by impaired intercellular signaling, results in insufficient angiogenesis and persistent non-healing.

Impaired wound healing driven by intrinsic anti-angiogenic programs

While restoring angiogenic endothelial states appears therapeutically attractive, single-cell analyses caution against nonspecific therapies that promote the expansion of healing-associated endothelial clusters, as they can also harbor pathogenic features. For example, ENTPD1⁺ VECs overexpress diabetic PAD risk genes (ANKIB1 and ANP32E) and underexpress protective genes (SLC30A7 and TMF1) (Zhang et al. 2025), suggesting that indiscriminate expansion of endothelial states may be maladaptive. Therapeutic approaches should therefore selectively target endothelial-intrinsic anti-angiogenic pathways.

Several such regulators have been identified through scRNA-seq studies. SH3BGRL3, which is upregulated in ECs from healing DFUs, promotes endothelial proliferation, migration, and angiogenesis under high-glucose conditions in HUVECs (Zhao et al. 2025). In vivo, endothelial overexpression of SH3BGRL3 accelerates wound closure and enhances neovascularisation in diabetic mice, supporting its role in reparative angiogenesis. In contrast, Schlafen-5 (SLFN5), an interferon-induced protein, is overexpressed in DFU ECs (Fan et al. 2024; Metzner et al. 2022). Functional studies demonstrate that SLFN5 suppresses angiogenic capacity by promoting oxidative and stress-associated injury (Fig. 5b); thereby, SLFN5 knockdown restores the migration, tube formation, and angiogenesis of HUVECs in vitro (Fan et al. 2024). Consistent with these findings, endothelial SLFN5 suppression improves wound healing in diabetic mouse models, supporting its role as a negative regulator of reparative angiogenesis. These studies indicate that selectively targeting endothelial-intrinsic pathways can promote the transition from maladaptive anti-angiogenic states toward reparative angiogenic states, thereby restoring vascularization and improving wound healing in DFUs.

In addition to targeting endothelial-intrinsic anti-angiogenic programs, restoring endothelial responsiveness to VEGF represents a complementary therapeutic strategy to enhance angiogenic endothelial states in DFU. Single-cell studies of human diabetic skin identify reduced expression of phospholipase Cγ2 (PLCγ2), a downstream effector of VEGF receptor signaling, in diabetic ECs (Fig. 5c) (Verma et al. 2025; Rustagi et al. 2022; Wang et al. 2020). This reduction is driven by hypermethylation of the PLCγ2 promoter, confirmed by bisulfite sequencing of human diabetic wound edge ECs (Verma et al. 2025). Functional rescue experiments demonstrate that restoring PLCγ2 expression reactivates VEGF-responsive signaling (Verma et al. 2025; Rustagi et al. 2022). In diabetic ischemic wounds, this intervention enhances angiogenesis, improves perfusion, increases vessel abundance, and accelerates wound closure, with the greatest effects observed when PLCγ2 rescue is combined with VEGF therapy. Together, these observations identify restoration of PLCγ2 signaling as a promising therapeutic strategy to overcome VEGF resistance and promote the transition of ECs toward reparative angiogenic states that support wound healing.

Overall, single-cell studies indicate that non-healing DFUs arise from endothelial dysfunction at multiple levels, including impaired state transitions, defective intercellular communication, activation of intrinsic anti-angiogenic programs, and reduced VEGF responsiveness. Collectively, these defects trap ECs in maladaptive angiogenesis-impaired states that limit revascularisation and contribute to persistent non-healing. Therapeutically, these findings suggest that strategies aimed at restoring endothelial plasticity, selectively suppressing anti-angiogenic pathways, and re-sensitizing ECs to VEGF signaling may be more effective than broadly stimulating angiogenesis alone for promoting wound healing in diabetic patients.

Cross-tissue comparison of endothelial states: translational and therapeutic implications

Across the DVCs, diabetes drives ECs into maladaptive states characterized by inflammation, fibrosis, aberrant angiogenesis, or impaired reparative angiogenesis. Single-cell studies presented in this review demonstrate that endothelial dysfunction across DVCs is not governed by a single pathway but instead arises from a shared principle: the convergence of multiple regulatory layers, including epigenetic memory, post-translational control, transcriptional reprogramming, and intercellular signaling. While perturbation at any single level can impair endothelial function, diabetes induces convergent dysregulation across these layers, thereby locking ECs into persistent pathological states. We propose that DVCs are driven by tissue-specific EC states that emerge from the integration of these layers, rather than by isolated signaling pathways. For example, in diabetic arteries, EC dysfunction is shaped by epigenetic reprogramming (e.g. LINC00607), post-translational dysregulation (e.g. JNK-WWP2-DDX3X axis), and immune-endothelial crosstalk (e.g. TREM2), collectively driving (1) a pro-inflammatory and pro-fibrotic endothelial state with impaired angiogenic capacity (Calandrelli et al. 2020; Sriram et al. 2022; You et al. 2023; Malhi et al. 2024); in DR, regulatory layers converge to produce (2) a pathological angiogenic and EC barrier-disruptive inflammatory state (Corano Scheri et al. 2023; Zhang et al. 2025; He et al. 2024; Yang et al. 2025); in DN, these processes promote (3) a pro-fibrotic and maladaptive angiogenic/proliferative endothelial state (Li et al. 2021; Stefansson et al. 2022; Wilson et al. 2019; Ferreira et al. 2026); whereas in DFUs, ECs are trapped in (4) an inflammatory, anti-angiogenic states. This indicates that endothelial dysfunction across organs is driven by combinations of tissue-adapted (a) pro-inflammatory, (b) pro-fibrotic, and (c) dysregulated angiogenic programs, representing a recurring theme between these DVCs. Under physiological conditions, ECs dynamically transition between quiescent, inflammatory, and angiogenic states in response to environmental cues (Citrin et al. 2025). Across DVCs, however, a common consequence of diabetes is impaired endothelial plasticity, which prevents normal state transitions and promotes the persistence of tissue-specific dysfunctional endothelial states.

Although these regulatory principles are shared, their functional outcomes are highly tissue-specific, necessitating vascular bed-targeted therapeutic strategies. This is particularly evident in the regulation of angiogenesis as revealed by the single-cell studies. While dysregulated angiogenesis represents a recurring theme across DVCs, its manifestation differs markedly between vascular beds. In diabetic arteries, ECs exhibit impaired angiogenic capacity within a pro-inflammatory and pro-fibrotic state (Calandrelli et al. 2020; Sriram et al. 2022; You et al. 2023; Malhi et al. 2024). In DR, ECs adopt a pathological angiogenic state characterized by excessive and disorganized neovascularization (Corano Scheri et al. 2023; Zhang et al. 2025; He et al. 2024; Yang et al. 2025). In DN, angiogenesis becomes maladaptive and contributes to glomerular remodeling (Dou et al. 2018), whereas in DFU, ECs exhibit impaired reparative angiogenesis, leading to chronic non-healing wounds (Zhang et al. 2025; Lu et al. 2023; Du et al. 2023; Wang et al. 2024; Verma et al. 2025; Rustagi et al. 2022; Fan et al. 2024; Zhao et al. 2025). Historically, therapies have attempted to modulate angiogenesis through VEGF targeting, with anti-VEGF strategies in retinopathy, and pro-angiogenic approaches such as recombinant VEGF delivery in wound healing (Arrigo et al. 2022; Bang et al. 2024). However, these approaches have shown incomplete efficacy in some patients (Arima et al. 2020; Antonetti et al. 2021; Bang et al. 2024), where the findings synthesized in this review provide a mechanistic explanation: angiogenesis in diabetic EC dysfunction is not solely VEGF-driven. For instance, in DR, endothelial dysfunction involves VEGF-independent mechanisms, including disruption of tip- and stalk-cell identities, VEGFC signaling, TUBB6-WNT activation, and STING/MAP4K4-mediated barrier disruption (Corano Scheri et al. 2023; Zhang et al. 2025; He et al. 2024; Yang et al. 2025). In contrast, in DFU, impaired angiogenesis is driven by intrinsic anti-angiogenic programs such as SLFN5 activation and reduced VEGF responsiveness due to PLCγ2 suppression (Verma et al. 2025; Rustagi et al. 2022; Fan et al. 2024). These findings indicate that diabetic endothelial dysfunction does not uniformly suppress or promote angiogenesis, but rather disrupts the balance between quiescence, repair, and growth, resulting in excessive angiogenesis in DR, maladaptive angiogenesis in DN, and insufficient reparative angiogenesis in diabetic arteries and DFUs. This also helps explain the limitations of VEGF-targeted therapies, which do not account for endothelial heterogeneity. For example, although effective for some DR patients, intravitreal anti-VEGF therapy has been associated with an increased risk of DFU, reflecting systemic VEGF suppression (Hong et al. 2025). The available evidence suggests that therapeutic strategies must account for vascular bed specificity, as targeting shared pathways without considering endothelial context may lead to unintended systemic effects.

Collectively, this review highlights that endothelial dysfunction in diabetes is sustained by overlapping and interconnected regulatory mechanisms, resulting in heterogeneous endothelial states across tissues. As such, single-pathway interventions are unlikely to provide a durable therapeutic benefit. Instead, these findings support the development of combination strategies that target multiple tissue-specific regulatory layers in ECs. The integrated framework presented in Fig. 1 illustrates endothelial states across tissues, which may guide future combination therapies. For instance, in diabetic arteries, combined targeting of epigenetic (LINC00607), post-translational (WWP2), and immune (TREM2) pathways, which were identified from the same single-cell dataset (GSE156341), may more effectively restore endothelial function than targeting any pathway separately. The translational status of endothelial-state regulators identified through human single-cell studies across DVCs is summarized in Table 2. Importantly, most endothelial-state regulators identified through single-cell studies remain at the pre-clinical stage, highlighting the need for further translational investigation and evaluation of combination therapeutic approaches, with some of the key outstanding questions summarized in Table 3.

Table 2.

Translational status of endothelial-state regulators identified through human single-cell studies of diabetic vascular complications

DVC	Target	Translational status	Intervention	Experimental system	Outcome	Clinical trial	Ref
Diabetic Artery	LINC00607	Pre-clinical	Nuclear LINC00607 inhibition	HUVECs (cultured in diabetic conditions)	↓ Monocyte adhesion, ↓ SERPINE1 expression	None	Sriram et al. (2022)
	WWP2	Pre-clinical	Wwp2 endothelial-specific knockout	Diabetic mice	↑ Endothelial injury & vascular remodeling	None	You et al. (2023)
	TREM2	Pre-clinical	Trem2 neutralization	Diabetic mice	↑ Post-ischemic perfusion & neovascularization	None	Malhi et al. (2024)
PDR	VEGF-C	Indirect clinical evidence	OPT-302 + ranibizumab (VEGF-A/C/D inhibition)	nAMD patients	↓ Neovascular activity	Phase 2b (NCT03345082)	Jackson et al. (2023)
	TUBB6-WNT	Pre-clinical	Tubb6 knockdown	OIR mouse retinas	↓ Sprouting angiogenesis & impaired tip-cell formation	None	Zhang et al. (2025)
	STING	Pre-clinical	GLP-1 receptor agonists	HRVECs (cultured in diabetic conditions) & diabetic mice	↓ STING signaling & preserved endothelial junction integrity	None	He et al. (2024)
	MAP4K4	Pre-clinical	DMX-5804	Diabetic mice	↓ Retinal vascular leakage & pathological neovascularization	None	Yang et al. (2025)
DKD	ADAM17-ERBB4	Pre-clinical	Pan-ErbB inhibition (erlotinib)	Diabetic mice	↓ Podocyte injury & albuminuria	None	Lee et al. (2017)
	EDN1-EDNRA	Clinical validation	Atrasentan (EDNRA inhibition)	Patients with T2DM-associated CKD	↓ Albuminuria & risk of renal disease progression	SONAR, Phase 3 (NCT01858532)	Heerspink et al. (2019)
		Clinical validation	Zibotentan + dapagliflozin (EDNRA inhibition)	CKD patients	↓ Albuminuria	ZENITH-CKD, Phase 2b (NCT04724837)	Wasehuus et al. (2025)
	MEF2C	Indirect clinical evidence	SGLT2 inhibitors	Human DKD patients	Partial restoration of the dysregulated MEF2C endothelial regulatory network	Supported indirectly by SGLT2i studies	Ferreira et al. (2026)
DFU	SH3BGRL3	Pre-clinical	Endothelial SH3BGRL3 overexpression	Diabetic mice	↑ neovascularisation & wound closure	None	Zhao et al. (2025)
	RAB17	Pre-clinical	RAB17 overexpression	HDMECs (cultured in diabetic conditions) & diabetic wounds	↑ HIF1α, angiogenesis & wound healing	None	Du et al. (2023)
	SLFN5	Pre-clinical	SLFN5 knockdown	HUVECs (cultured in diabetic conditions) & diabetic mice	Restored migration, tube formation, angiogenesis, & wound healing	None	Fan et al. (2024)
	PLCγ2	Preclinical	PLCγ2 restoration + VEGF therapy	Diabetic ischemic wound models	↑ Angiogenesis, perfusion, vessel abundance, & wound closure	None	Verma et al. (2025); Rustagi et al. (2022)

Abbreviations: CKD chronic kidney disease, DFU diabetic foot ulcer, DKD diabetic kidney disease, EDNRA endothelin receptor type A, ErbB erythroblastic oncogene B receptor family, GLP-1RA glucagon-like peptide-1 receptor agonist, HDMEC human dermal microvascular endothelial cell, HRVEC human retinal vascular endothelial cell, HUVEC human umbilical vein endothelial cell, nAMD neovascular age-related macular degeneration, OIR oxygen-induced retinopathy, PDR proliferative diabetic retinopathy, SGLT2i sodium-glucose cotransporter-2 inhibitor, SONAR Study Of Diabetic Nephropathy with Atrasentan, T2DM type 2 diabetes mellitus, VEGF vascular endothelial growth factor, VEGF-C vascular endothelial growth factor C, VEGF-D vascular endothelial growth factor D

Table 3.

Outstanding questions

• Do distinct molecular mechanisms driving endothelial dysfunction (e.g., LINC00607 activation, TREM2-mediated signaling, and WWP2 loss) co-occur within the same ECs, or do they define separate endothelial subpopulations across diabetic vascular beds?

• Which single-cell-defined endothelial states act as causal drivers of diabetic vascular complications, and which represent adaptive or bystander responses to tissue injury?

• Do pathogenic endothelial states precede clinical disease onset, and can they be translated into circulating, imaging, or functional vascular biomarkers for early prediction and patient stratification?

• How can AI-based models integrate multi-omic endothelial datasets to reconstruct disease trajectories and predict the vascular bed-specific impact of therapeutic perturbations?

• What are the optimal combinations of targets across distinct regulatory layers to effectively reprogram tissue-specific endothelial states in DVCs?

• How can VEGF-independent pathways be targeted within existing therapies to overcome treatment resistance and improve outcomes across different vascular beds?

Overall, this review supports a paradigm shift in the treatment of DVCs from targeting individual pathways to reprogramming maladaptive endothelial states in a tissue-specific manner. This approach not only explains the limited efficacy of current therapies but also provides a conceptual foundation for targeting tissue-specific dominant dysfunctional programs to develop next-generation vascular therapeutics.

Limitations and future directions

While single-cell transcriptomic approaches have provided unprecedented resolution into endothelial heterogeneity and dysfunction in DVCs, several limitations should be considered when interpreting these findings. First, relatively few human scRNA-seq studies have examined ECs in diabetes, with most available single-cell vascular datasets generated in mouse models, where major human diabetic organs, such as the liver, pancreas, and adipose tissue, remain uncharacterized at endothelial resolution, thereby limiting direct clinical extrapolation and the scope of studies reviewed here. Even within the vasculature, all arterial studies discussed here relied exclusively on mesenteric arteries (Calandrelli et al. 2020; Sriram et al. 2022; You et al. 2023; Malhi et al. 2024), which do not reflect the distinct endothelial phenotypes of coronary, peripheral, cerebral, or aortic beds, thereby limiting generalisability. Moreover, important macrovascular and microvascular diabetic complications remain underrepresented in currently available human endothelial single-cell datasets, despite the growing evidence implicating endothelial dysfunction in their pathogenesis. These include diabetic peripheral vascular disease, diabetic cardiomyopathy, and diabetic neuropathy. This limitation restricts our ability to compare endothelial states across the full spectrum of diabetic vascular disease and highlights the need for future endothelial-focused single-cell studies in these complications. Furthermore, because ECs constitute only a small fraction of most arteries (Hu et al. 2021), and are highly fragile and stress-responsive to enzymatic dissociation (Oh et al. 2022), whole-tissue scRNA-seq approaches often fail to resolve rare or disease-specific endothelial states. EC enrichment strategies, such as CD31⁺ magnetic bead enrichment (MACS) (Zhao et al. 2021), will therefore be essential for accurately defining endothelial heterogeneity when isolating ECs, while broader adoption of snRNA-seq may help mitigate dissociation-induced artefacts (Oh et al. 2022).

Moreover, human samples are costly to profile, invasive, and require clinically indicated procedures. Thus, many available datasets are derived from small (Table 1), heterogeneous patient cohorts that lack balanced representation of sex and ethnicity, which reduces statistical power. Additionally, some studies lack true non-diabetic equivalents. For example, FVMs from PDR lack non-diabetic equivalents because healthy individuals do not undergo vitrectomy, restricting the interpretability of diabetic specific EC signatures (Corano Scheri et al. 2023). Furthermore, currently available datasets do not capture the full clinical spectrum of DVCs. For example, single-cell studies in DR are largely limited to PDR, while endothelial profiling of earlier non-proliferative stages remains unavailable. Similarly, DFU datasets often lack detailed stratification by ulcer subtype (e.g., neuropathic versus ischemic), which may influence endothelial phenotypes and integration and cross-study comparisons. Consequently, no longitudinal single-cell studies on human diabetic ECs currently exist, preventing investigation of the progression or regression of endothelial dysfunction. To address this gap, future work may adapt artificial intelligence (AI) tools, such as UNAGI (Zheng et al. 2025), to analyse time-series single-cell data and predict changes in endothelial state or treatment responses when longitudinal human samples are unavailable.

Additionally, despite several studies analysing the same publicly available scRNA-seq datasets, a notable inconsistency remains in EC cluster identification and annotation. Distinct groups report different numbers of EC subclusters and rely on different markers for EC definition, such as PECAM1, CDH5, CLDN5 (Zhao et al. 2025), or CDH5^hi, VWF^hi, LYVE1^low (Verma et al. 2025), and sometimes use a single marker, such as ACKR1 (Lu et al. 2023), in diabetic wounds, which risks misclassifying non-EC populations as ECs. This challenge is compounded by the fact that no single marker is universally expressed across all endothelial subtypes, while several commonly used markers may also be expressed in non-endothelial populations or vary according to tissue context and disease state. Consequently, differences in marker selection can substantially influence both the number and identity of EC clusters identified within the same dataset. These discrepancies are further exacerbated by substantial variation in computational analysis pipelines and by differences in tissue dissociation and library preparation protocols across tissues, collectively limiting cross-study comparability and integration. This variability highlights the urgent need for a standardized endothelial nomenclature and harmonized analytical frameworks. Future studies may benefit from a standardized annotation framework that incorporates expression of a core set of endothelial markers and systematically excludes non-endothelial lineage markers by adopting a shared reference-based annotation platform (e.g., CellMarker 2.0 (Hu et al. 2023)). Continued refinement of such resources through incorporation of newly identified endothelial markers could further improve annotation accuracy, reproducibility, and cross-study comparability over time.

Furthermore, validation studies frequently rely on HUVECs, which poorly recapitulate the tissue-specific transcriptional and functional states of human diabetic ECs. For example, direct comparison between in vitro HT treated HUVECs and donor-derived arterial ECs revealed SERPINE1 was consistently ranked first by odds ratio in both models (Calandrelli et al. 2020). However, the ranking of the other co-expressed genes, such as TRIO and RUNX1, differed, suggesting that the in vitro model may not fully mimic the human donor-derived cells. Therefore, future work should incorporate patient-specific endothelial models generated by reprogramming peripheral blood-derived endothelial progenitor cells (EPCs) into iPSCs and differentiating them into ECs (Chang et al. 2013; Harding et al. 2017), to better recapitulate inter-individual endothelial dysfunction. These models can be further integrated into vessel-on-chip platforms, which more closely mimic the native human endothelial microenvironment through co-culture with perivascular cells under physiological flow (Cochrane et al. 2019). Moreover, many in vivo validations are performed using global or non-EC-specific knockout models, making it difficult to attribute phenotypes directly to endothelial dysfunction rather than systemic effects. This gap can be addressed by utilizing endothelial-specific Cre systems (e.g., Cdh5-Cre) to generate EC-restricted knockout models (You et al. 2023).

Addressing these limitations will be critical to translating insights into effective therapies for endothelial dysfunction in different tissues. Therefore, future work should prioritize the development of a single-cell, multi-modal, multi-omic, and cross-organ diabetic EC atlas on post-mortem organs, ideally from the same individuals. However, because endothelial identity is strongly shaped by tissue context, direct integration of ECs from different organs may obscure biologically meaningful vascular bed-specific differences by treating them as batch effects, resulting in misleading biological interpretations. Future studies may therefore benefit from defining endothelial states within individual tissues before performing cross-tissue comparisons of shared and tissue-specific programs.

Single-cell multi-omics enables simultaneous, paired profiling of multiple molecular modalities within the same individual cell, providing more accurate mechanistic insights into disease causes in humans (Guan and Quek 2025). Single-cell epigenomics (e.g., scATAC-seq), genomics (e.g., scDNA-seq), transcriptomics (e.g., scRNA-seq), and proteomics (e.g., CyTOF) can be combined in different combinations, such as genome-transcriptome, epigenome-transcriptome, and transcriptome-proteome. For example, pairing single-cell epigenomic profiling with transcriptomics enables a direct assessment of DNA regulatory features, such as methylation, chromatin accessibility, histone marks, and other epigenetic mechanisms that are linked to the cell’s own transcripts (Baysoy et al. 2023). Such approaches will be important for defining the contribution of epigenetic mechanisms to tissue-specific endothelial dysfunction in diabetes, which cannot be fully resolved by scRNA-seq alone. Importantly, the studies synthesized in this review indicate that endothelial dysfunction arises through the integration of multiple regulatory layers, including epigenetic, transcriptional, post-translational, and intercellular signaling mechanisms, which collectively shape tissue-specific endothelial states. Single-cell multi-omic approaches will therefore be essential to define how these regulatory layers interact within the same EC and how their relative contributions differ across organs. Such insights may facilitate the identification of dominant disease-driving mechanisms within specific vascular beds and support the development of more precise, state-directed therapeutic strategies. Recently, various AI tools, such as scGPT, have been demonstrated to perform multi-omic integration, batch correction, perturbation prediction, cell type annotation, and gene network inference (Cui et al. 2024). Incorporating tools such as scGPT into future multi-organ diabetic endothelial atlases may help integrate diverse data types and predict perturbation effects that may contribute to organ-specific endothelial dysfunction.

In parallel, spatial transcriptomics will be needed to situate endothelial states within their microenvironment and predict the bidirectional interactions between ECs and neighboring cell populations (Vandereyken et al. 2023). As highlighted throughout this review, endothelial dysfunction is not solely cell-autonomous but is shaped by reciprocal signaling between ECs and surrounding cells. By preserving tissue architecture, spatial approaches can identify the cellular neighborhoods associated with specific endothelial states and predict how ECs both influence and are influenced by local signaling networks. Such analyses may uncover therapeutic opportunities that not only target dysfunctional ECs but also the surrounding cellular microenvironment that sustains pathological endothelial states.

Conclusion

Single-cell transcriptomic studies across human DVCs demonstrate that endothelial dysfunction is not driven by a single pathway but arises from the convergence of distinct regulatory layers into coordinated, tissue-specific EC states. By integrating human, endothelial-specific single-cell data, this framework captures the heterogeneity and context-dependent mechanisms of EC dysfunction across vascular beds with greater translational relevance than bulk or animal-based approaches. This model also provides a mechanistic explanation for the limited efficacy of current pathway-directed therapies, such as anti-VEGF strategies, which show incomplete efficacy because they fail to address the broader endothelial state driving the disease. Therefore, this review identified tissue-specific EC states and highlighted potential therapeutic strategies, including alternative VEGF-targeting approaches and combinatorial therapies tailored to the dominant endothelial state within each complication. Collectively, these insights support a paradigm shift from a single pathway-centric intervention toward state-specific endothelial reprogramming, providing a foundation for more precise and clinically relevant treatments for DVCs.

Abbreviations

ADAM17

A disintegrin and metalloproteinase domain-containing protein 17

ADAMTS1

A disintegrin and metalloproteinase with thrombospondin motif 1

AGE

Advanced glycation end products

AI

Artificial intelligence

ANGPT1

Angiopoietin-1

ANGPT2

Angiopoietin-2

ANKIB1

Ankyrin repeat and IBR domain-containing protein-1

AP-1

Activator protein-1

APOE

Apolipoprotein E

APP

Amyloid precursor protein

ARE

Antioxidant response element

ASC

Apoptosis-associated speck-like protein containing a CARD

AQP1

Aquaporin-1

AXL

AXL receptor tyrosine kinase

BRB

Blood-retinal barrier

cAMP

Cyclic adenosine monophosphate

c-Fos

FBJ murine osteosarcoma viral oncogene homolog

c-Myc

Cellular myelocytomatosis oncogene

cGAMP

Cyclic guanosine monophosphate-adenosine monophosphate

cGAS

Cyclic GMP-AMP synthase

CCL5

C–C motif chemokine ligand 5

CCN1

Cellular communication network factor-1

CCND1

Cyclin D1

CD31 (PECAM1)

Platelet endothelial cell adhesion molecule-1

CD44

CD44 antigen

CDH5

Cadherin-5

CKD

Chronic kidney disease

CLU

Clusterin

COL4A2

Collagen type IV alpha-2 chain

COL6A5/COL6A6

Collagen type VI alpha-5 / alpha-6 chains

COL14A1

Collagen type XIV alpha-1 chain

CREB

CAMP-response element-binding protein

CXCL1/CXCL8/CXCL12

C-X-C motif chemokine ligand 1 / 8 / 12

CyTOF

Cytometry by time-of-flight

dEC

Degenerative endothelial cell

DDX3X

DEAD-box helicase 3 X-linked

DFU

Diabetic foot ulcer

DKD

Diabetic kidney disease

DLL4

Delta-like ligand 4

DM

Diabetes mellitus

DM-PAD

Diabetes mellitus-associated peripheral artery disease

DN

Diabetic nephropathy

DR

Diabetic retinopathy

DVC

Diabetic vascular complication

EC

Endothelial cell

EC-GC

Glomerular capillary endothelial cell

ECM

Extracellular matrix

EDN1

Endothelin-1

EDNRA

Endothelin receptor type A

EFNB2

Ephrin-B2

ENO1

Enolase-1

ENTPD1

Ectonucleoside triphosphate diphosphohydrolase-1

EPHA4

Ephrin type-A receptor 4

EPHB1

Ephrin type-B receptor 1

EPC

Endothelial progenitor cell

ERBB4

Erb-B2 receptor tyrosine kinase 4

ERK1/2

Extracellular signal-regulated kinases 1 and 2

ESM1

Endothelial cell-specific molecule-1

ETS (family)

ETS proto-oncogene family of transcription factors

ETS1

ETS proto-oncogene 1

FAM3C

Family with sequence similarity 3 member C

FLT1

Fms-related receptor tyrosine kinase 1 (VEGFR1)

FVM

Fibrovascular membrane

FZD

Frizzled class receptor

GAS6

Growth arrest-specific protein 6

GC

Glomerular capillary

GEC

Glomerular endothelial cell

GLP-1

Glucagon-like peptide-1

GLP-1RA

Glucagon-like peptide-1 receptor agonist

GLP-1R

Glucagon-like peptide-1 receptor

GSK3β

Glycogen synthase kinase-3 beta

HDMEC

Human dermal microvascular endothelial cell

HE-VasEndo

Healing-enriched vascular endothelial cell

HG

High glucose

HG/PA

High glucose plus palmitate

HLA-C

Major histocompatibility complex class I C

HIF-1

Hypoxia-inducible factor 1

HIF-1α

Hypoxia-inducible factor-1 alpha

HRMEC

Human retinal microvascular endothelial cell

HRVEC

Human retinal vascular endothelial cell

HT

High glucose plus TNF-α treatment

HuBMAP

Human BioMolecular Atlas Program

HUVEC

Human umbilical vein endothelial cell

iBRB

Inner blood-retinal barrier

ICAM

Intercellular adhesion molecule

ICAM1

Intercellular adhesion molecule-1

IFN

Interferon

IGFBP3

Insulin-like growth factor-binding protein-3

IL-1β

Interleukin-1 beta

IRF3

Interferon regulatory factor 3

iPSC

Induced pluripotent stem cell

ITGB1/ITGB3

Integrin beta-1 / beta-3

ITGAV

Integrin alpha-V

JAG1/JAG2

Jagged ligand 1 / 2

JNK

C-Jun N-terminal kinase

KPMP

Kidney Precision Medicine Project

KRT14/KRT17/KRT6A

Keratin 14 / 17 / 6A

LEC

Lymphatic endothelial cell

LGALS9

Galectin-9

LINC00607

Long intergenic non-protein coding RNA 607

lncRNA

Long non-coding RNA

LRP2/LRP5/LRP6

Low-density lipoprotein receptor-related protein 2 / 5 / 6

MACS

Magnetic-activated cell sorting

MAP4K4

Mitogen-activated protein kinase kinase kinase kinase 4

MAPK

Mitogen-activated protein kinase

MAPK44/42 (ERK1/2)

Extracellular signal-regulated kinases 1 and 2

MCAM

Melanoma cell adhesion molecule

MEF2A/MEF2C

Myocyte enhancer factor-2A / -2C

MIF

Macrophage migration inhibitory factor

MMP/MMP2/MMP9

Matrix metalloproteinase / -2 / -9

m⁶A

N6-methyladenosine

mtDNA

Mitochondrial DNA

mTOR

Mechanistic target of rapamycin

MT2A

Metallothionein 2A

NADPH

Nicotinamide adenine dinucleotide phosphate

nAMD

Neovascular age-related macular degeneration

NAMPT

Nicotinamide phosphoribosyltransferase

NF-κB

Nuclear factor kappa-B

NR4A1

Nuclear receptor subfamily 4 group A member 1

NRF2

Nuclear factor erythroid 2-related factor 2

NRP1/NRP2

Neuropilin-1 / -2

NOTCH

Notch receptor

NLRP3

NOD-like receptor family pyrin domain-containing 3

OIR

Oxygen-induced retinopathy

PA

Palmitate

PAI-1

Plasminogen activator inhibitor-1

PDGFB

Platelet-derived growth factor subunit B

PDGFR

Platelet-derived growth factor receptor

PAD

Peripheral artery disease

PDR

Proliferative diabetic retinopathy

PECAM1

Platelet endothelial cell adhesion molecule-1

PI3K/Akt

Phosphoinositide 3-kinase / protein kinase B

PLAUR

Plasminogen activator urokinase receptor

PLCγ2

Phospholipase C gamma-2

PLD2

Phospholipase D2

PLPP1

Phospholipid phosphatase-1

PLXNA2

Plexin-A2

prEC

Proliferative endothelial cell

PSEN1

Presenilin-1

RAB17

Ras-related protein Rab-17

RAGE

Receptor for advanced glycation end products

RUNX1

Runt-related transcription factor 1

ROBO2

Roundabout guidance receptor 2

scRNA-seq

Single-cell RNA sequencing

SE

Super-enhancer

SELE

E-selectin

SEMA4A/SEMA6A

Semaphorin-4A / -6A

SERPINE1

Serpin family E member 1

SERPINF1

Serpin family F member 1

S1PR1

Sphingosine-1-phosphate receptor 1

SGLT2i

Sodium-glucose cotransporter-2 inhibitor

SH3BGRL3

SH3 domain-binding glutamate-rich protein-like 3

SLFN5

Schlafen family member 5

SLIT3

Slit guidance ligand 3

snRNA-seq

Single-nucleus RNA sequencing

STING

Stimulator of interferon genes

TBK-1

TANK-binding kinase 1

TEK (TIE2)

Endothelial tyrosine kinase receptor

TGFB1/TGFB2/TGFB3

Transforming growth factor beta-1 / -2 / -3

TGFBR2

Transforming growth factor beta receptor-2

TNF/TNF-α

Tumour necrosis factor/alpha

TNFRSF10A

Tumor necrosis factor receptor superfamily member 10A

TNFRSF10B

Tumor necrosis factor receptor superfamily member 10B

TP53I11

Tumor protein p53 inducible protein 11

TREM2

Triggering receptor expressed on myeloid cells 2

TRIO

Triple functional domain protein

TRPS1

Trichorhinophalangeal syndrome type-1

TSPAN12

Tetraspanin-12

TUBB6

Tubulin beta-6 chain

T2D/T2DM

Type 2 diabetes mellitus

UTR

Untranslated region

VEC

Vascular endothelial cell

VE-cadherin

Vascular endothelial cadherin

VEGF/VEGFA/VEGFC

Vascular endothelial growth factor / A / C

VEGFR

Vascular endothelial growth factor receptor

VEGFR2/VEGFR3

Vascular endothelial growth factor receptor 2 / 3

VWF

Von Willebrand factor

WNT/WNT2B/WNT3A

Wingless/Integrated signaling pathway / family member 2B / 3A

WWP2

WW domain-containing E3 ubiquitin protein ligase 2

YBX1

Y-box binding protein 1

YTHDF2

YTH N6-methyladenosine RNA-binding protein 2

ZO-1

Zonula occludens-1

Authors’ contributions

R.A. conceived the review, designed its structure, performed the literature search and synthesis, prepared all figures, and wrote the original draft of the manuscript. M.M. contributed to conceptual development, supervised the study, and critically revised the manuscript. L.L. and T.A.K. critically revised the manuscript. S.S.A. and A.H.V.C. contributed to supervision and critically revised the manuscript. P.C. contributed to conceptualization and critically revised the manuscript. H.A. contributed to conceptualization, secured funding, and critically revised the manuscript. All authors read and approved the final manuscript.

Funding

This research was supported by Center for Biotechnology, Khalifa University of Science and Technology (KU-BTC). This research was funded by Khalifa University of Science and Technology through the Biomedical Science and Discovery Program under Project ID: KU-EXT-BISDI-2022–8434000456.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

This article is a review of previously published studies and publicly available data. Ethics approval and consent to participate are therefore not applicable.

Consent for publication

This manuscript does not contain any individual person’s data. Consent for publication is therefore not applicable.

Competing interests

The authors declare no competing interests.