introduction

Proliferative diabetic retinopathy (PDR) represents the most advanced stage of diabetic retinopathy. PDR affects approximately 17 million people globally and is a leading cause of blindness among adults [1], making it a significant global health and economic problem. Characterized by neovascularization at the interface of perfused and non-perfused tissue, fragile new vessel formation and progressive fibrosis places patients with PDR at risk of severe visual compromise from vitreous haemorrhage, tractional retinal detachment, and neovascular glaucoma. While the pathophysiology of PDR is complex and not yet completely understood, it is known that microvascular ischemia, chronic inflammation, and retinal neurodegeneration are all important in the disease process [2].

As a result of prolonged hyperglycaemia, retinal blood vessels undergo pericyte and endothelial cell apoptosis [3, 4]. The resulting capillary occlusion and ischemia leads to upregulation of hypoxia-inducible factor 1 and angiogenic factors, the most studied of which is vascular endothelial growth factor (VEGF). VEGF contributes to the progression of PDR through increased vascular permeability and the promotion of endothelial cell proliferation [2]. The management of PDR with anti-VEGF therapy, therefore, has become increasingly common, although pan-retinal photocoagulation remains the mainstay of treatment. However, some patients may progress despite available therapies, including anti-VEGF agents and there remains a need for additional therapeutic targets and improved treatment algorithms.

The search for other intraocular cytokines that are involved in PDR pathogenesis has yielded a large body of literature. In addition to VEGF, angiogenic factors including insulin-like growth factor-I, basic fibroblast growth factor, platelet derived growth factor, placental growth factor, and angiopoietin have all been implicated in retinal neovascularization [5, 6]. However, there remains inconsistency between studies regarding which cytokines are associated with PDR and the magnitude of this association. Clarity on the role of cytokines in PDR will aid in predicting disease severity, progression, treatment response, and identifying novel therapeutic targets. We have therefore conducted a systematic review and meta-analysis on the association of intraocular cytokines and PDR to address these inconsistencies and to quantitatively summarize the literature.

Methods

A detailed account of the methods used in this meta-analysis is available in our companion article on nonproliferative diabetic retinopathy (NPDR) in this issue, and is similar to those used in our previous work on a diabetic macular oedema [7]. In brief, a systematic literature search was done using Ovid MEDLINE, Embase, and Web of Science databases without year limitation until January 18, 2021. The search statements are available in Supplementary Fig. S1.

Studies were assessed by two independent reviewers using the following exclusion criteria: (1) the study did not examine a correlation, clinical outcome or response to treatment of an aqueous or vitreous cytokine; (2) was on subjects other than human adults; (3) included subjects with diabetic macular oedema; and (4) was a review article, editorial, or opinion piece. Studies that assessed the correlation of cytokines to clinical outcomes (prognostic biomarkers) or disease correlation to cytokine concentrations (diagnostic biomarkers) were evaluated for their risk of bias using the Quality in Prognosis Studies (QUIPS) tool or the Quality Assessment of Diagnostic Accuracy Studies (QUADAS) tool, respectively [8, 9]. Modifications to QUIPS and QUADAS are provided in Supplementary Fig. S2.

Primary outcome measures were mean and standard deviation intraocular cytokine concentration and p-values for comparisons between those with PDR and non-diabetic participants, if a control group was available. Review Manager (Version 5.3.5, Nordic Cochrane Centre, Copenhagen, Denmark) was used to compute standardized mean difference (SMD) and 95% confidence interval for cytokine concentrations between patients with PDR and controls using the inverse variance method when there was sufficient data of at least three data points. The formulation of the SMD used was Hedges’ adjusted g, selected as it includes an adjustment for small sample bias. A random-effects model was chosen because within-study and between-study variances were hypothesized to influence the true effect size. The magnitude of the SMD was defined as being very small if <0.20, small if 0.20-0.49, medium if 0.50–0.79, and large if ≥0.80. Between-study heterogeneity was assessed using the Cochrane Q test and I2 statistic as described previously. Statistical difference for the Cochrane Q test was set as p < 0.05; I2 statistics of 0.25, 0.50 and 0.75 denoted low, medium and high levels of heterogeneity, respectively, as per previous literature [10].

A sensitivity analysis was undertaken by removing one study at a time to assess outcome stability; if the effect size significantly changed with the removal of a single study, the data for that cytokine was deemed to have failed the sensitivity analysis. To determine if the results of this study were influenced by participants having previously received treatment for their PDR, two subgroup analyses were performed: one included only studies where patients had received no treatments for PDR within 3 months of sample collection; the other included only studies where all patients were treatment naïve.

Results

The initial search identified 2947 records, of which 1681 remained after automated removal of duplicates. Following review at the full-text level to ensure that no cytokine concentrations were missed, 480 studies met the inclusion/exclusion criteria and 341 of those were specific for PDR (Fig. 1). These 341 studies encompassed for 10379 eyes with PDR and 6269 eyes from healthy controls [11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124,125,126,127,128,129,130,131,132,133,134,135,136,137,138,139,140,141,142,143,144,145,146,147,148,149,150,151,152,153,154,155,156,157,158,159,160,161,162,163,164,165,166,167,168,169,170,171,172,173,174,175,176,177,178,179,180,181,182,183,184,185,186,187,188,189,190,191,192,193,194,195,196,197,198,199,200,201,202,203,204,205,206,207,208,209,210,211,212,213,214,215,216,217,218,219,220,221,222,223,224,225,226,227,228,229,230,231,232,233,234,235,236,237,238,239,240,241,242,243,244,245,246,247,248,249,250,251,252,253,254,255,256,257,258,259,260,261,262,263,264,265,266,267,268,269,270,271,272,273,274,275,276,277,278,279,280,281,282,283,284,285,286,287,288,289,290,291,292,293,294,295,296,297,298,299,300,301,302,303,304,305,306,307,308,309,310,311,312,313,314,315,316,317,318,319,320,321,322,323,324,325,326,327,328,329,330,331,332,333,334,335,336,337,338,339,340,341,342,343,344,345,346,347,348,349,350,351].

Fig. 1
figure 1

Flow of information through the different phases of the systematic review; the initial search, duplicate removal, full-text elegibility, and manuscripts on proliferative diabetic retinopathy meeting inclusion criteria.

Full size image

Study characteristics

Population and study characteristics are summarized in Table 1. Vitreous cytokines were assessed in 82% (279/341) of studies, with only 12% (41/341) assessing aqueous cytokines and the remaining 6% (21/341) assessing both vitreous and aqueous cytokines. Most studies (57%, 193/341) did not specify the type of diabetes in PDR patients, while 24% (81/341) of studies used participants with type 2 diabetes, 1% (3/341) of studies used those with type 1, and 14% (46/341) used patients with either type 1 or type 2 diabetes. The remaining studies classified patients by treatment status; 0.6% (2/341) were insulin dependent, 0.3% (1/341) were non-insulin dependent, and 4% (14/341) used both insulin dependent and independent patients. The analytical method used to quantify cytokine concentrations was clearly stated in 99% (340/341) studies, with the enzyme-linked immunosorbent assay and multiplex assay being most frequently used in 70% (239/341). Of the 293 studies with a control arm, the most commonly used control was eyes with epiretinal membrane or macular hole (38%, 111/293), with cataract surgery, retinal detachment, macular pucker, vitreous haemorrhage, vitreous floaters, vitreomacular traction, lens dislocation, and lens subluxation also being used as controls.

Table 1 Population characteristics and intraocular markers investigated by the relevant studies identified in the systematic review.
Full size table

Quality assessment

Diagnostic biomarker studies

Two hundred and seventy-nine studies (82%) evaluated the diagnostic potential of intraocular cytokines and were assessed for quality using QUADAS (Fig. 2a), with further details available for individual studies in Supplemental Table S1. A case-control design was used in 88% (246/279) of studies and the selection criteria were clearly stated in 71% (198/279). There was a low risk of selection bias in 70% (195/279). Selected patients were an appropriate match for the study question in every case and all enrolled patients were included in the final analysis in 94% (263/279) of studies. The index text demonstrated a low risk of study bias in all studies, but was never interpreted without knowledge of the reference standard. All studies had each participant receiving the same reference standard, classified PDR appropriately and matched the review question, and in no studies was it interpreted independently from the index test (100%, 279/279). There was an overall low risk of bias in diagnostic studies.

Fig. 2
figure 2

Summary of quality assurance data.

Full size image

Prognostic biomarker studies

Sixty-two studies (18%) evaluated the prognostic potential of cytokines. The overall QUIPS assessment is shown in Fig. 2b, and study details are available in Supplemental Table S2. There was a low risk of bias in study participation in 71% (44/62), and study attrition had a low risk of bias in 85% (53/62) of studies. Data collection and outcome measurement, respectively, had a low risk of bias in 81% (50/62) and 100% (62/62) of studies. An unclear risk of bias in confounding was found in most studies (85%; 53/62) as most studies did not account for non-diabetic ocular or systemic conditions that may influence intraocular cytokine concentrations. Finally, 81% (50/62) of studies had an appropriate statistical analysis. Prognostic studies were found to have an overall low risk of bias.

Main association of DME with cytokine levels

Fifty-four cytokines had sufficient data for inclusion in the meta-analysis, with results detailed in Table 2. Aqueous concentrations (standard mean difference, 95% confidence interval, and p value) of IL-1β (5.76, 2.45–9.07, p = 0.0006), IL-6 (2.69, 1.34–4.03, p < 0.0001), IL-8 (2.77, 1.11–4.43, p = 0.001), MCP-1 (1.80, 0.72–2.88, p = 0.001), TNF-α (9.74, 5.01 to 14.47, p < 0.0001), and VEGF (2.31, 1.61–3.00, p < 0.00001) were significantly higher in patients with PDR when compared to healthy nondiabetic controls. For vitreous cytokines, concentrations of IL-2 (0.61, 0.18 to 1.04, p = 0.006), IL-4 (1.07, 0.30–1.84, p = 0.007), IL-6 (1.72, 1.30–2.13, p < 0.00001), IL-8 (1.78, 1.16–2.39, p < 0.00001), angiopoietin-2 (1.46, 0.93–1.98, p < 0.00001), eotaxin (0.99, 0.36–1.63, p = 0.002), erythropoietin (1.37, 0.81–1.93, p < 0.00001), GM-CSF (0.72, 0.34–1.10, p = 0.0002), GRO (1.40, 0.49–2.32, p = 0.003), HMGB-1 (0.86, 0.62–1.10, p < 0.00001), IFN-γ (0.75, 0.33–1.17, p = 0.0004), IGF (0.48, 0.17–0.80, p = 0.003), IP-10 (1.53, 0.87–2.20, P < 0.00001), MCP-1 (2.53, 1.76–3.30, p < 0.00001), MIP-1 (1.09, 0.28–1.91, p = 0.009), MMP-9 (0.92, 0.32–1.52, p = 0.003), PDGF-AA (1.28, 0.77–1.78, p < 0.00001), PlGF (1.33, 0.80–1.87, p < 0.00001), sCD40L (1.15, 0.30–2.01, p = 0.008), SDF-1 (1.60, 0.79–2.41, p = 0.0001), sICAM-1 (1.40, 0.73–2.07, p < 0.0001), sVEGFR (2.40, 0.92–3.89, p = 0.002), TIMP (0.73, 0.34–1.13, p = 0.0003), TNF-α (1.05, 0.34–1.77, p = 0.004), and VEGF (1.96, 1.69–2.23, p < 0.00001) were significantly elevated in subjects with PDR as compared to healthy controls. For all other cytokines intraocular concentrations were not significantly different between PDR and controls, did not pass the sensitivity analysis, or had insufficient data for inclusion in the meta-analysis.

Table 2 Summary of the outcomes and description of the role in vivo of each cytokine in the meta-analysis.
Full size table

Supplementary Table S3 shows the known associations of intraocular cytokines with PDR from the previous studies identified in the systematic review and contrasts those with the results of this meta-analysis. Of the 31 aqueous and vitreous cytokines for which we found a significant elevation in cytokine concentration in PDR versus controls, previous studies had found them to be either not significantly different or even reduced in PDR versus controls in 17% (120/700) of the cases. Forest plots for each of the analysed cytokines are available in Supplementary Fig. S3, and funnel plots for each cytokine that had at least five data points are available in Supplementary Fig. S4.

Some studies included patients that had received previous treatment for PDR, such as a laser photocoagulation or intravitreal anti-VEGF. To determine if the results of the meta-analysis was influenced by patient treatment status, two subgroup analyses were performed (Supplementary Fig. S5). When including only studies with patients that had no treatments for at least 3 months prior to study enrolment, the effect sizes for aqueous cytokines was 5.76 for IL-1β (2.45–9.07, p = 0.0006), 3.56 for IL-6 (2.03 to 5.09, p < 0.00001), 5.58 for IL-8 (2.97 to 8.19, p < 0.0001), 2.40 for MCP-1 (0.68–4.12, p = 0.006), 9.74 for TNF-α (5.01 to 14.47, p < 0.0001), and 2.25 for VEGF (1.54–2.97, p < 0.00001). Similarly, for vitreous cytokines the effect size was 1.48 for IL-6 (0.75–2.21, p < 0.0001), 1.37 for IL-8 (0.25–2.48, p = 0.02), 1.16 for IL-10 (0.44–1.89, p = 0.002), 1.20 for IFN-γ (0.47–1.94, p < 0.0001), 3.29 for MCP-1 (0.87–5.72, p = 0.008), –2.29 for PEDF (–3.87 to –0.70, p = 0.005), 0.91 for TNF-α (0.36–1.47, p = 0.001), and 3.11 for VEGF (2.29 to 3.93).

When including only studies with patients that had no previous treatments the effect size for aqueous VEGF was 3.08 (2.25–3.90, p < 0.00001), vitreous PEDF was –2.89 (–4.41 to –1.37, p = 0.0002), and vitreous VEGF was 3.71 (2.90–4.51, p < 0.00001). All other cytokines failed the sensitivity analysis.

Discussion

Panretinal photocoagulation has been a longstanding treatment for PDR. However, this treatment can restrict the visual field, reduce visual acuity at night, and has the potential to worsen macular oedema [352, 353]. The efficacy of intravitreal ranibizumab, an anti-VEGF agent, was compared to panretinal photocoagulation in DRCR Network Protocol S, a randomized trial of 394 eyes. This study found that ranibizumab was non-inferior to panretinal photocoagulation in terms of mean change in visual acuity and the proportion of eyes without neovascularization [354]. In both groups, however, at two years more than 40% of participants had active neovascularization on fundus photography and 2–3% had developed neovascular glaucoma. This suggests that proinflammatory or proangiogenic cytokines in addition to VEGF are likely involved in the disease process and may therefore be appropriate treatment targets.

Since 1992 there have been 341 studies on the association of intraocular fluid cytokines with PDR, with the vast majority being published in the last decade. Interpretation of this wealth of data has been complicated by inconsistencies in which cytokines may be involved in disease pathogenesis, with some studies even finding no difference in VEGF concentrations between patient with PDR and controls (for example Nishiguchi et al. 2013 [218]; Semeraro et al. 2014 [257]). Previous meta-analyses were highly selective in the cytokines investigated or did not address PDR specifically and looked at a relatively small number of studies [355, 356]. We therefore attempted to undertake a more comprehensive systematic review and meta-analysis on intraocular cytokines in PDR. This work summarizes 10379 eyes with PDR and 6269 eyes from healthy controls.

Previous studies have identified a correlation between aqueous and vitreous cytokine concentrations in diabetic retinopathy [324], and one might expect similar cytokine profiles in the aqueous and vitreous samples. In the end stage of severe retinal ischemic disease proinflammatory and proangiogenic cytokines can migrate from the posterior to anterior segment of the eye and promote new vessel formation on the iris surface or anterior chamber angle, leading to neovascular glaucoma [161, 275]. Over eighty percent (279/341) of the included studies utilized vitreous humour samples alone, which is appropriate for investigating PDR pathogenesis due to the proximity between the vitreous and the retina. However, this resulted in many aqueous cytokines having insufficient data for inclusion in the meta-analysis. Given the relative ease of aqueous paracentesis, it would be beneficial for future studies to collect both aqueous and vitreous humour samples to further elucidate the relationship between posterior and anterior disease processes.

This study found significant elevation of aqueous IL-1β, IL-6, IL-8, IP-10, MCP-1, TNF-α, and VEGF, and vitreous IL-4, IL-6, IL-8, IL-12, angiopoietin-2, eotaxin, erythropoietin, GRO, HMGB-1, IFN-γ, IGF, IP-10, MCP-1, MIP-1, MMP-9, PDGF-AA, PlGF, sCD40L, SDF-1, sICAM-1, sVEGFR, TIMP, TNF-α, and VEGF. Most of the cytokines significantly associated with PDR had a medium (0.50–0.79) to large (≥0.80) effect size and none were sensitive to the result of any one study. Many of these molecules are known to have an important proinflammatory and proangiogenic role, and for some there may be a synergistic or antagonistic affects amongst cytokines. Angiopoietin-2 influences blood-retinal barrier stabilization and vessel remodelling through mediation of Tie-2 phosphorylation and is elevated in the retina of diabetic patients with chronic hyperglycaemia [357, 358]. IFN-γ is a key participant in inflammation in diabetic retinopathy through the breakdown of the blood-retina barrier and upregulation of other proangiogenic cytokines [359]. This also induces the release of IP-10, another CXC chemokine that in turn prevents neovascularization and inhibits IL-8-induced chemotaxis [86, 360]. PDGF is a ubiquitous growth factor produced in the retina by the retinal pigment epithelium, astrocytes, and ganglion cells, and directly contributes to neovascularization and fibrovascular proliferation in PDR [361, 362]. PlGF, a member of the VEGF family, is not required for physiologic angiogenesis but does play a role under pathological conditions where levels correlate with PDR activity [133]. This cytokine is a target of aflibercept and concentrations may also decrease following conbercept injection [350]. TNF-α and sCD40-L, which is a member of the TNF superfamily, promote angiogenesis and may induce the expression of VEGF in vivo [169, 363].

The number of elevated cytokines in PDR is much greater than that seen in previous work on retinal vein occlusion [364], neovascular age-related macular degeneration [365], and diabetic macular oedema [7], and in our companion article on cytokines in NPDR. This is reflective of the complexity of PDR pathogenesis and the advanced disease state, where dysregulation of several biochemical and molecular signalling pathways is driven by oxidative stresses [366]. In the NPDR meta-analysis it was found that aqueous IL-6, IL-17, and VEGF as well as vitreous VEGF were elevated when compared to controls, all with large effect sizes. Apart from IL-17, which failed the sensitivity analysis, similar results were obtained for PDR. In addition, aqueous IL-1β, IL-8, and TNF-α were elevated in NPDR versus controls but failed sensitivity analysis, making their significance inconclusive. Given that these cytokines were elevated in PDR, it suggests that the importance of their role in the nonproliferative form of DR may be better elucidated with additional studies.

When including only studies where all participants were without treatment for their PDR in the preceding 3 months, aqueous IL-1β, IL-6, IL-8, MCP-1, TNF-α, and VEGF and vitreous IL-6, IL-8, IL-10, IFN-γ, MCP-1, TNF-α, and VEGF were found to be significantly elevated and, in most cases, had larger effect size than found in the primary analysis. Furthermore, when using only data from treatment-naive patients only aqueous and vitreous VEGF was significantly elevated in PDR, with other cytokines having an insufficient number of studies or a failed sensitivity analysis. Due to notable heterogeneity in treatment type, duration, and the use of combination therapy, we were not able to stratify patients by the type of prior treatment. Our analysis was further limited as some studies did not provide sufficient detail on the nature of prior treatments. It would be interesting to investigate the influence of specific prior treatments on the cytokine milieu since these may have different mechanisms and durations of action, allowing for a better understanding of the post-treatment cytokine profile.

In both the treatment-free for 3 months and treatment naïve subgroups PEDF was significantly lower in PDR than in controls, with an effect size of -2.29 (-3.87 to -0.70) in those without treatment for 3 months and -2.89 (-4.41 to -1.37) for those without any prior treatments. PEDF can counteract VEFG-induced vascular permeability and inhibit retinal neovascularization [362, 367, 368], acting as an inhibitor of VEGF function through its action on the VEGF receptor [369]. It is disruptions to the balance of proangiogenic and antiangiogenic cytokines because of chronic hyperglycaemia in diabetes that leads to proliferative retinopathy and its sequelae [53, 370]. It therefore fits mechanistically that patients with more active neovascularization would have lower levels of PEDF. While angiogenic stimuli such as VEGF are required for PDR to occur, pathology can only develop when the ‘break’ of negative regulators such as PEDF fail [371].

Of the 31 aqueous and vitreous cytokines for which we found a significant elevation, 17% (120/700) of previous investigations had found them to either be not significantly different or reduced in PDR when compared to controls (Table S3). There are several possible reasons for this inconsistency in the literature. There is a variable nature to an individual’s cytokine profile. While there is little data available on longitudinal intraocular cytokine concentrations, work on plasma cytokines indicates that levels are sensitive to the duration of diabetes [372], acute episodes of hyperglycaemia [373], and systemic factors such as hypertension [374] and dyslipidaemia [375]. Furthermore, the ‘healthy’ control groups often contained patients of different ethnicity and with either systemic or ocular conditions such as macular hole or epiretinal membrane that may influence cytokine concentrations [376, 377]. As studies generally had small sample sizes, with a median of 27 cases and 20 controls, they may be particularly susceptible to these influences. Future work using large ocular fluid biobanks may provide a normative database that identify genetic, temporal, and comorbid variations in the cytokine profile, and upon which changes specifically due to PDR can be further elucidated. Over 70% of studies utilized enzyme linked immunosorbent assays for cytokine quantification, with more recent studies being predominantly based on multiplex analysis. There remains both intra-assay and inter-assay variability due to differences in sample collection, handling, and storage as well as the types of buffers and antibodies used. It is unclear to what extent these impact reported cytokine concentrations, and if a correction factor should be applied when comparing data from different assay types.

Conclusions

Previous studies have shown conflicting associations for most cytokines assessed in PDR. This meta-analysis demonstrated elevated aqueous concentrations of IL-1β, IL-6, IL-8, MCP-1, TNF-α, and VEGF, and vitreous concentrations of IL-2, IL-4, IL-6, IL-8, angiopoietin-2, eotaxin, erythropoietin, GM-CSF, GRO, HMGB-1, IFN-γ, IGF, IP-10, MCP-1, MIP-1, MMP-9, PDGF-AA, PlGF, sCD40L, SDF-1, sICAM-1, sVEGFR, TIMP, TNF-α, and VEGF in patients with PDR.

When assessing patients without recent treatment, levels of the anti-angiogenic cytokine PEDF were low. This work identifies a number of candidate cytokines other than VEGF that are implicated in PDR and adds clarity to the large body of literature. These findings suggest potential biomarkers of PDR development and severity and point to potential therapeutic targets.

Supplemental information is available at Eye’s website.