This prospective study followed the principles of the Declaration of Helsinki and was approved by the Ethics Committee of Beijing Tongren Hospital, Capital Medical University. All subjects signed an informed consent form before participation.

A total of 131 patients with type 2 DM, including 77 males and 54 females, aged 27–76 years old were recruited from the outpatient clinic of Beijing Tongren Hospital from April 2016 to September 2020. The participants were assigned to the DM group if they had no DR [32 patients, aged 37–75 years, median (IQR): 56 (48–65) years], non-proliferative diabetic retinopathy (NPDR) [56 patients, aged 29–76 years, 56 (51–61)] years, and proliferative diabetic retinopathy (PDR) [43 patients, aged 27–74 years, 55 (49–60) years].

Participants with type 2 DM were included in the study. Type 2 DM was defined according to the 2020 American Diabetes Association (ADA) guidelines of DM (5) and 2017 A Position Statement of DR (6). Those with the ability to provide informed consent were included in the study. Exclusion criteria included people with type 2 DM with macular edema secondary to other retinal vascular diseases; co-existent other retinal diseases such as age-related macular degeneration, uveitis, and inherited retinal diseases; recent history of posterior segment or cataract surgery; ocular media opacity; and unable to tolerate examinations due to severe system diseases. Patients with history of lipid disorders or on lipid-lowering therapy were also excluded.

The definition was followed to the national cholesterol education program (NCEP) adult treatment panel-III (ATP-III) report in 2001 and the guidelines for the prevention and treatment of dyslipidemia in Chinese adults 2016. Dyslipidemia was defined as high cholesterol (TC) group (TC > 5.17 mmol/L), high TG group (TG > 1.70 mmol/L), or high low-density lipoprotein cholesterol (low-density lipoprotein cholesterol, LDL-C) group (LDL-C > 3.37 mmol/L).

Arteriosclerosis indices including API (LDL-C/HDL-C), AI [(TC-HDL-C)/HDL-C), and AIP (log(TG/HDL-C)] were calculated. Atherogenic plasma index, AI, and AIP with high sensitivity and specificity were selected as the cutoff values on receiver operating characteristic (ROC) curve. Patients with API > 2.24 (AUC: 0.746; sensitivity = 0.708, specificity = 0.517), AI > 2.91 (AUC, 0.723; sensitivity = 0.629; specificity = 0.724) or AIP > 0.01 (AUC 0.564; sensitivity = 0.607, specificity = 0.552) were assigned to high API, high AI, and high AIP groups, respectively.

The Luminex technology (Luminex 200™ liquid chip detector, Millipore, Boston, Massachusetts, USA) was applied to detect the plasma level of cytokines, VEGF-A, VEGF-C, VEGF-D, and PlGF according to the manufacturer's instructions.

Sample size was determined by using the Cochran's sample size formula and also reference the previous studies (12). Sample size was calculated at 95% confidence level with a margin of error of ±5%. To detect the difference (0.83) between means of expression level of angio-cytokines with a significance level of (alpha) 0.05. The minimum sample size was calculated to be 25 in each group. However, since there is variability in the circulating level of angio-cytokines/serum lipid profiles and highly variable parameters, the sample size was increased to above 30 in the present study.

Statistical analysis was performed using SPSS software (SPSS, Inc., 23.0, Chicago, IL, USA). Normality was assessed by Kolmogorov-Smirnov test and Shapiro-Wilk test. Variance homogeneity was tested by the Levene's test. Age of participant, duration of diabetes, fasting blood glucose, glycosylated hemoglobin, hypertension, TG, TC, and LDL-C were described by means ± standard deviation (mean ± SD) or median (interquartile range). The comparisons among groups were analyzed by one-way analysis of variance (ANOVA) or the Kruskal–Wallis test. The circulating levels of cytokines VEGF-A, VEGF-C, VEGF-D, and PlGF were reported as mean ± SD or median (interquartile range), comparisons between groups [DM, NPDR, and PDR groups, or DM and DR groups (combined NPDR with PDR as DR group)] was analyzed by independent sample t-test or Mann–Whitney U-test according to the data distribution (If the data were not normally distributed or variance homogeneity was not met). Bonferroni corrections was applied for comparison between the three groups. The effects of age, gender, the duration of diabetes, AI, API, AIP as well as their relationships with VEGF-A, VEGF-C, VEGF-D, and PlGF on DR patients were analyzed by multiple logistic regression analysis. P < 0.05 indicated statistical significance.

No statistically significant differences in age or gender were detected between the DM (with no DR), NPDR and PDR groups (P_age = 0.692; P_gender = 0.756). The average duration of DM was shorter in the DM group compared to that in the NPDR and the PDR group (12.59 ± 5.79 vs. 9.14 ± 6.76 years, P_NPDRvs.DM = 0.013; 12.86 ± 6.76 vs. 9.14 ± 6.76 years, P_PDRvs.DM = 0.021). There was no statistical significance of fasting blood glucose in the DM, NPDR, and PDR groups (P = 0.151). The difference between the NPDR and the DM groups [6.80 (6.18–7.83) vs. 7.75 (7.03–8.98) years, P = 0.018)], PDR and the DM groups [6.80 (6.18–7.83) vs. 7.80 (6.68–9.10) years, P = 0.017] of the glycosylated hemoglobin levels were statistically significant. No significant difference was detected in the duration of hypertension among the three groups (χ² = 4.27, P = 0.118) (Table 1).

The level of TC in the three groups differed significantly [4.30 (3.91–5.03) vs. 4.59 (3.74–5.68) vs. 4.99 (4.32–5.89) mmol/L, P = 0.033], with the level of TC in the PDR group being higher than that in the DM group [4.99 (4.32–5.89) vs. 4.30 (3.91–5.03) mmol/L, P = 0.006]. In addition, LDL-C also showed statistical differences [2.50 (2.05–3.09) vs. 2.80 (2.03–3.60) vs. 3.06 (2.58–3.92) mmol/L, P = 0.011], with the level in the PDR group being higher than that in the DM group [3.06 (2.58–3.92) vs. 2.80 (2.03–3.60) mmol/L, P = 0.002]. No statistically significant difference was detected in the three groups with respect to TG and HDL (P_TG = 0.312; P_HDL−C = 0.562) (Table 1).

The API in the three DR groups showed a statistically significant difference (2.19 ± 0.71 vs. 2.42 ± 0.91 vs. 2.82 ± 0.95, P = 0.008), with the value of the PDR group higher than that of the DM group (P = 0.003). The difference of AI in the three groups also differed significantly (2.75 ± 0.97 vs. 2.92 ± 1.17 vs. 3.44 ± 1.20, P = 0.019), with the value in the PDR group significantly higher than that in the NPDR and DM group (P_PDRvs.NPDR = 0.024, P_PDRvs.DM = 0.010). There was no difference of the AIP-value between the three groups (P = 0.247) (Table 1).

The plasma level of VEGF-A differed significantly [15.47 (11.40–25.92) vs. 27.64 (20.62–35.32) vs. 24.85 (18.59–33.54) pg/ml, P < 0.001], with the level of VEGF-A in NPDR and PDR groups significantly higher than that in the DM group (P_NPDRvs.DM = 0.002, P_PDRvs.DM = 0.001). There was also a significant difference in VEGF-D between the three groups (P < 0.001), with higher level in NPDR and PDR groups than that in DM group (P_NPDRvs.DM < 0.001, P_PDRvs.DM < 0.001). The level of PlGF in the three groups differed significantly [1.41 (0.58–2.44) vs. 2.69 (1.89–3.31) vs. 2.41 (1.65–3.87) pg/ml, P < 0.001], with the level of PlGF in the NPDR and PDR groups significantly higher than that in the DM group [2.69 (1.89–3.31) vs. 1.41 (0.58–2.44) pg/ml, P_NPDRvs.DM = 0.012, 2.41 (1.65–3.87) vs. 1.41 (0.58–2.44) pg/ml, P_PDRvs.DM < 0.001]. No statistical significance was detected in the three groups with the circulating level of VEGF-C (P = 0.088) (Table 1).

There was a significant difference in the plasma level of VEGF-D between the normal and abnormal API group [154.10 (71.78–274.36) vs. 255.58 (147.91–400.44) pg/ml, P = 0.009]. Similarity, differed expression level of VEGF-D was found between the normal AI and abnormal group [179.77 (72.09–280.18) vs. 255.59 (147.91–400.45) pg/ml, P = 0.016]. There is no significant difference in the expression level of VEGF-A, VEGF-C, and PlGF in between the normal API, AI, and AIP groups. No significant difference was found in the expression level of VEGF-D between the normal AIP and abnormal AIP groups (Table 2).

The level of VEGF-A, VEGF-C, VEGF-D, PlGF is significantly higher in the abnormal TC group in comparison with the normal TC group (P_VEGF−A = 0.008; P_{VEGF−C =} 0.001, P_VEGF−D = 0.035; P_PlGF = 0.004). Similarity, in the abnormal LDL-C group, the level of VEGF-A, VEGF-C, VEGF-D, and PlGF is significantly higher than that in the normal LDL-C group (P_VEGF−A = 0.032; P_VEGF−C = 0.006; P_VEGF−D = 0.018, P_PlGF = 0.006). No statistical difference was found between the circulating level of VEGF-A, -C, -D, and PlGF in the abnormal TG, HDL-C groups, compared with that in the normal TG and HDL-C groups, respectively, indicating the circulating level of the angiogenic cytokines is correlated with the abnormal serum level of TC and LDL-C (Table 3).

Table 4 shows the correlations of the plasma level of VEGF-A, VEGF-C, VEGF-D, and PlGF with atherogenic indices on the occurrence and severity of DR using multiple ordinal logistic regression models. When DM, NPDR, and PDR were considered as independent variables, after controlling the duration of diabetes and hypertension, the serum level of hemoglobins and fasting glucose, AI and API are strong risk factors for the occurrence and severity of DR. Atherogenic index is associatd with VEGF-D [AI: P = 0.038, OR = 1.38 (1.02–1.86); VEGF-D: P = 0.002, OR = 1.00 (1.00–1.01)] and PlGF [AI: P = 0.021, OR = 1.43 (1.06–1.92); PlGF: P = 0.004, OR = 1.50 (1.14–1.98)] contributed to DR (Table 4A). Atherogenic plasma index is associated with VEGF-D (Table 4B): [API: P = 0.027, OR = 1.56 (1.05–2.30); VEGF-D:P = 0.002, OR = 1.00 (1.00–1.01)] and PlGF [API: P = 0.011, OR = 1.66 (1.12–2.47); PlGF: P = 0.005, OR = 1.49 (1.13–1.96)] (Table 4B). The correlations etween AIP and VEGF-C, VEGF-D, VEGF-A, PlGF were summarized in Table 4C.

The above analysis has proved that API and AI strongly associated with DR severity. A greater area under the ROC (AUC) means a more useful test of the statistical model. The AUC of AI (AUC:0.72; 95% CI: 0.62–0.82) and API (AUC: 0.75;95%CI: 0.65–0.84) is much higher than TC (AUC:0.68; 95%CI: 0.58–0.78), TG (AUC: 0.54; 95% CI 0.41–0.67) LDL-C (AUC:0.72; 95% CI: 0.62–0.82), HDL-C (AUC:0.57, 95% CI: 0.45–0.69), indicating that both API and AI have higher capacity to differential DR.

The Wisconsin Epidemiologic Study of DR first identified that a long-term increase in the serum level of TC might be a risk factor for hard exudates in DR in 1991 (21), FIELD and ACCORD study further confirmed that lipid-lowing therapy with fenofibrate significantly reduce the progression of DR and the need for laser treatment (3, 22), indicating that intensive lipid control is correlated with improved clinical prognosis. Statins are the most effective drug to reduce TC and LDL-C. However, the role of statins is more profound in cardiovascular disease than DR (3, 22–24).

Our study found that the levels of TC and LDL-C in the NPDR and PDR groups were significantly higher than those in the DM group, which are consistent with the results of previous studies on type 1 DM (25). In addition, the serum levels of TC and LDL-C in PDR patients were significantly increased compared to those patients with NPDR and DM, suggesting that the increased levels of TC and LDL-C in serum accelerate the progression of DR.

In this study, the lipid metabolism indices were used to explore the relationship between dysregulated lipid metabolism and the expression level of circulating cytokines and DR severity. In routine clinical practice, LDL-C is calculated (by the Friedewald formula) than measured directly. Several studies have found that the estimation of LDL-C has significant limitations, especially in individuals with hypertriglyceridemia with DM (it may contain abnormal composition of the TG-rich lipoproteins which is not considered in the Friedewald formula) (14, 26–28).

Non-HDL-C represents the cholesterol components carried by atherogenic lipoproteins such as LDL, very low density lipoprotein, and intermediate density lipoprotein and has been recommended as the primary lipid-lowing target by the 2019 American College of Cardiology (ACC) and American Heart Association (AHA) guidelines on the management of blood cholesterol (29).

It has been confirmed by studies in cardiovascular diseases that AI and API reflect two or three traditional parameters, which are simple and economic and can be used as biomarkers. Atherogenic index and API have several key advantages than the traditional lipid parameters: they have explicit pathophysiologic link to the development of atherosclerosis, robust from a laboratory measurement standpoint etc. (14) API, AI, and AIP have been applied for assessing plasma lipid homeostasis and are also implicated in the pathogenesis of macrovascular complications, especially in the early stage of the disease when the unaltered concentration of various lipids is increased (30). Atherogenic plasma index is a sensitive predictor of asymptomatic type 2 DM patients with coronary atherosclerosis and the incidence of microvascular complications and peripheral neuropathy in DM patients increased significantly with the increased AIP (15, 31). Since this is the first study to apply the indices in eye study, we used the ROC curve, the objective method to obtain the cut off value, and also referred to the value of the lipid metabolism in cardiovascular and atherosclerosis studies [API: 0.16–2.24 (32), AI: 3.21–3.37 (33), AIP: 0.6–0.21 (34, 35)].

In this study, we found that the AUC of AI (TC-(HDL-C)/HDL-C) (a) and API (LDL-C/HDL-C) is significantly higher than TC, TG, LDL-C, and HDL-C, indicating these simpler combined parameters have higher diagnostic capacity than the traditional parameters for DR. Furthermore, logistic regression analysis showed that AI and API are independent risk factors for DR, and provide evidence for a link between lipid metabolism and DR.

VEGF-A is not only the most important vascular leakage inducer but also a key angiogenesis factor and a target of anti-VEGF therapy in ocular conditions (36, 37). However, macular oedema does not resolve completely in a number of patients with DR suggesting that there are other influencers in the pathogenesis of DR. Our study shows that other family members of VEGF are also raised in DR, indicating the need for novel molecular targets. Both VEGF-C and VEGF-D are secreted glycoproteins that exhibit structural homology but have differential receptor binding (Flk-1 and Flt-3, respectively) and regulatory mechanism is a vital mediator to angiogenesis in ischemic heart disease in type 2 DM (38). These cytokines have been demonstrated as independent predictors for suspected coronary artery disease and are associated with all-cause mortality (39). In a previous study, VEGF-A-FLK-1 signaling has been shown as a primary mediator of endothelial cell mitogenesis, survival, and microvascular permeability (40). Together with VEGF-A, VEGF-C, and VEGF-D promote blood vessel development (angiogenesis) by binding and activating VEGF-R-2 and VEGF-R-3. VEGF-C is also a potent inducer of vascular permeability and angiogenesis in AMD. The ongoing phase III trial is investigating a soluble form of VEGF-R-3, which comprises the extracellular domains 1–3 of human VEGF-R3. In addition, the Fc fragment of human IgG1 suppresses the binding of VEGF-C and -D to VEGF-R 2 and 3, respectively (ClinicalTrials.gov Identifier: NCT02543229). The positive outcome from the phase II trial has shown that the combined intravitreal administration of OPT-302 and ranibizumab is better than ranibizumab alone for the patient's vision prognosis. Furthermore, PlGF is deemed to play a crucial role in both experimental animal models and retinal vascularization (41, 42). A high expression level of PlGF has also been detected in vitreous and PDR membrane in patients with PDR (43–46).

In the present study, based on a significant body of evidence in preclinical and clinical studies that support the pathological function of VEGF-A, -C, -D, and PlGF, a protein chip-Luminex technology was applied to quantitatively detect the circulating expression levels in patients with abnormal lipid metabolism group and normal group, respectively. We found that the circulating level of VEGF-D significantly contributed to the occurrence and servierty of DR associated with AI and API, confirmed the previous finding that VEGF-D is also an important regulator of genes related to lipid metabolism and inflammation (47–50). Our results also provide evidence for the first time that besides diabetes duration, AI, API, VEGF-D, and PlGF are strong risk factors for the occurrence and progression of DR. Atherogenic index and API have synergic effects with VEGF-D and PlGF. Placental growth factor may release VEGF-A by binding to VEGF-R-1 to activate VEGF-R-2, thereby indirectly stimulating angiogenesis, or combined with VEGF-R-1 to promote the proliferation of endothelial cells and angiogenesis (44). Animal experiments also have shown that PlGF promotes the infiltration of macrophages in early atherosclerotic lesions in apolipoprotein E deficiency (Apo E^−/−) mice and rabbits with high TC. In this study, by logistic regression analysis, after controlling the age, sex, duration of diabetes and VEGF-A, -C, -D, the serum level of TC and LDL was found to concomitant with higher level expression of PlGF in either DM or DR group, HDL was found to be negatively with PlGF expression. Thus, it could be speculated that high non-HDL-C promotes the development of DR by upregulating the expression of PlGF.

Presently, the specific regulatory mechanism of abnormal lipid metabolism involved in the progression of DR is yet to be clarified. Nevertheless, the present study has some limitations. Due to the small sample size, confounding factors could not be eliminated in the analysis process, and various measurement methods and correction factors might also affect the results of this study. Well-matched and large-scale prospective clinical trials and cohort studies are warranted to validate the conclusions in the study.

In summary, the ratios of LDL-C/HDL-C and TC-LDL-C/HDL-C are useful and simple indicators of DR. The present study further found that dysregulation of lipid metabolism promotes the development of DR and are associated with upregulation of circulating VEGF-A, VEGF-C, VEGF-D, and PlGF. The current results may partly explain why the response to anti-VEGF varies between individuals. Strengthening the management of blood lipids in patients with DM may help to reduce the levels of VEGF and PlGF and resultant oedema. Therefore, patients with DM should focus on the management of blood lipids in addition to their glycemic status. Further understanding the correlation between the arteriosclerosis-associated plasma indices and circulating levels of angiogenic cytokines can be translated to novel therapeutic targets for DR.