The Innate Immune System in Diabetic Retinopathy

Abstract

The prevalence of diabetes has been rising steadily in the past half-century, along with the burden of its associated complications, including diabetic retinopathy (DR). DR is currently the most common cause of vision loss in working-age adults in the United States. Historically, DR has been diagnosed and classified clinically based on what is visible by fundoscopy; that is vasculature alterations. However, recent technological advances have confirmed pathology of the neuroretina prior to any detectable vascular changes. These, coupled with molecular studies, and the positive impact of anti-inflammatory therapeutics in DR patients have highlighted the central involvement of the innate immune system. Reminiscent of the systemic impact of diabetes, immune dysregulation has become increasingly identified as a key element of the pathophysiology of DR by interfering with normal homeostatic systems. This review uses the growing body of literature across various model systems to demonstrate the clear involvement of all three pillars of the immune system: immune-competent cells, mediators, and the complement system. It also demonstrates how the relative contribution of each of these requires more extensive analysis, including in human tissues over the continuum of disease progression. Finally, although this review demonstrates how the complex interactions of the immune system pose many more questions than answers, the intimately connected nature of the three pillars of the immune system may also point to possible new targets to reverse or even halt reverse retinopathy.

1. Introduction

Diabetes, characterized by the dysregulation of carbohydrate and lipid metabolism, results from impaired insulin secretion and/or insulin resistance. Its prevalence has more than quadrupled over the past four decades, from 108 million in 1980 to more than 425 million worldwide (Internation Diabetes Federation, 2019; World Health Organization, n.d.). This, coupled with one-third of all diabetic individuals experiencing related vision complications (Ko et al., 2012), has resulted in diabetic retinopathy (DR) becoming the leading cause of preventable vision loss in working-age individuals (Yau et al., 2012). DR has classically been viewed as a microvascular complication of diabetes and categorized based on those vascular abnormalities (Bursell et al., 2001; Treatment and Retinopathy, 1991a, 1991b, 1991c; Wilkinson et al., 2003). However, more recent discoveries of neurodegeneration and immune dysregulation early in DR patients have called into question this vascular-centric view (Kong et al., 2016). Consistent with these findings, inflammatory genes are most closely tied to DR, as identified by genome-wide association studies (Abhary et al., 2009).

The mechanisms behind these immune system–based manifestations in human DR patients have been investigated in rodent models of DR. However, no rodent model fully recapitulates the human presentation of DR (Olivares et al., 2017). This major limitation has driven investigators to use a combination of models to study different human manifestations of DR, including drug-induced (e.g., streptozocin), genetic (i.e. Akita [Ins2^Akita], and leptin receptor deficient [db/db]), and non-diabetic injury–based (e.g., oxygen-induced retinopathy [OIR]; and ischemia/reperfusion) (Lai and Lo, 2013; Olivares et al., 2017) models. Interestingly, a common feature that has emerged from these disparate models is the central role of the innate immune system. This manuscript will examine our evolving understanding of DR from the perspective of these experimental in vivo models contextualized to human studies and in vitro data. In particular, we will focus on the role of the innate immune system in DR-related neurodegeneration and vasculopathy.

2. The intricacy of the ocular innate immune system

The retina is exquisitely sensitive to metabolic perturbations due to the combination of high metabolic demand and limited vascular supply (Dai et al., 2014; Joyal et al., 2018, 2016; Kooragayala et al., 2015). Importantly, the innate immune system, through its role as the early responder to environmental perturbations, maintains the homeostasis and visual function of this environment (Murakami et al., 2020). Much of this homeostasis is maintained and controlled by the innate immune system in a set of critically important anatomic barriers made up of and regulated by highly specialized cell types.

a. Blood-retinal barrier (BRB) as the first line of defense

A series of anatomic barriers is the first component of protection provided by the innate immune system. From the perspective of the retina, the first most-distal barrier is the inner BRB and consists of endothelial cells woven together by tight junctions and regulated by pericytes (Figure 1). This barrier limits the ability of circulating cells, proteins, and other molecules from crossing between the systemic circulation and the highly sensitive neuroretinal tissue (Black et al., 1985). The molecules that do make it past this first inner BRB layer next come in contact with the glial limitans, a second, more-proximal barrier made up of Müller glial cells and astrocytic processes. Finally, the inner and most-proximal barrier is the outer BRB, which is formed by the retinal pigment epithelium (RPE) via tight junctions (Figure 1). These barriers not only limit the retina’s exposure to foreign elements but also modulate the retina’s access to nutrients from the vasculature (Black et al., 1985; Muoio et al., 2014). To coordinate these protective and metabolic functions, the different cells that make up the BRB communicate closely with neurons and microglia to make up the neurovascular unit (NVU; Figure 2) (Simó et al., 2018). Separate from their role in forming immune barriers, many of these cells, including microglia, Müller cells, and astrocytes, play important cellular roles in innate immunity, which will be discussed in later sections. Over the past decade, our group and others have brought to light the central and unique roles played by these different retinal cells in the pathophysiology of DR.

Figure 1. Anatomical blood-retinal barrier.

The layers of the retina, retinal pigment epithelium (RPE) and choroid are labeled on the left with details of the outer blood retinal barrier (oBRB) expanded in the diagram on the right depicting the tight and adherent junctions between the RPE cells in red. Additionally, a cartoon of the inner BRB (iBRB) is shown in the lower right with tight junction between endothelial cells, which are surrounded by pericytes, Müller glia, and astrocytes. Additionally, neurons of the ONL (outer nuclear layer) and INL (inner nuclear layer) are nearby. GCL: ganglion cell layer.

Figure 2. The complex multi-cellular dysfunction in diabetic retinopathy.

Diabetes virtually affects all retinal cells in a sequence that remains hard to elucidate as they function as a neurovascular unit and influence each other. (1) Vessel dysfunction reflects pericytes and endothelial cells dysfunction and loss, a phenomenon influenced by factors secreted by dying neurons and activated micro- and macroglia. (2) Microglia (brown) and macroglia (green) become activated in response to the alteration of the retinal environment associated with diabetes, a phenomenon enhanced by neuronal cell death and vascular preturbations. (3) Alteration of the retinal homeostasis by diabetes leads to neuronal dysfunction and ultimately cell death, a phenomenon enhanced by the increasingly pro-inflammatory environment and vascular perturbations.

b. Specific innate immune cells of the ocular system

i. Microglia

The NVU is composed of various cell types, each playing key roles in the integration and regulation of ocular homeostasis, which includes maintenance of the immune privilege status. Central nervous system microglia, including those in the retina, are derived from yolk sac primitive macrophages during development and are distributed evenly throughout three layers of the retina: the ganglion cell layer, the inner plexiform layer, and the outer plexiform layer (Herbomel et al., 2001; Nimmerjahn et al., 2005; Wake et al., 2009; Wang et al., 2016). As any other microglia of the central nervous system, retinal microglia continuously survey their environment by extending and retracting their processes in all directions (Damani et al., 2011; Karlstetter et al., 2015; Lee et al., 2008; Okunuki et al., 2018). This comprehensive access makes them well suited for phagocytosing retinal debris, supporting neighboring cells, and modulating synapses (Dando et al., 2016; Okunuki et al., 2018; Paolicelli et al., 2011; Rathnasamy et al., 2019). During development, microglia selectively target neurons and synapses for phagocytosis and complement-mediated destruction, without which neuropathy, synaptic degeneration, and vision loss result (Ferrer-Martín et al., 2014; Marín-Teva et al., 2004; Paolicelli et al., 2011; Roumier et al., 2004; Schafer et al., 2012). These necessary functions further require an intimate relationship with macroglial cells due to their prime location at the interface between the systemic circulation and the neuroretina. This relationship is all the more important because microglia depend on adenosine triphosphate supplied by nearby Müller cells to power these energy-intensive processes (Fontainhas et al., 2011; Wang and Wong, 2014). This co-dependence and communication between microglia and macroglia are becoming increasingly important for our understanding of the contribution of the innate immune response in the onset and progression of DR.

These surveying microglia are morphologically homogenous with small soma and ramified structures (McMenamin et al., 2019; Nimmerjahn et al., 2005); however, single-cell RNA-sequencing and gene expression profiles have determined that microglia are transcriptionally heterogeneous based on location and are dramatically different from peripherally circulating monocytes (Goldmann et al., 2016; Grabert et al., 2016). Together this is consistent with the important role that microglia, as the resident immune cells, play in the local maintenance of retinal function.

ii. Non-microglial cells

Several other innate immune cells, including perivascular macrophages, persistent hyalocytes, and dendritic cells, have been suggested to participate in the regulation of the unique retinal immune situation. Perivascular macrophages have been suggested to reside between the inner BRB and the glial limitans and putatively monitor this space for foreign proteins that pass through the barriers (Lewis et al., 2005; Mato et al., 1996; Mendes-Jorge et al., 2009). Pharmacologic disruption of the inner BRB has been reported to result in perivascular macrophage recruitment and activation (Mendes-Jorge et al., 2009). These observations suggest that this immune cell population could function as an active pseudo-barrier in clearing proteins and debris; however, there continues to be ongoing discussions on these cells’ exact nature and origin, including during diabetes, because it is difficult to differentiate them from other immune cells, local or circulating. Hyalocytes are bone marrow–derived macrophages located at the vitreoretinal interface during development, and their purpose is to phagocytose the transient vasculature of the lens (i.e., tunica vasculosa lentis) (McMenamin et al., 2002; Qiao et al., 2005). Persistent hyalocytes remain between the vitreous membrane and the inner limiting membrane after development. Although their purpose is not completely understood, they have been hypothesized to be antigen-presenting cells that transport antigens to the spleen and help to modulate the immunosuppressive environment of the eye via regulatory T cells (Murakami et al., 2020). Although their role in DR remains unknown, this population could certainly play an important role locally and systemically and participate in both the maladapted immune response that develops over time during diabetes and the onset and progression of DR. Such a role has been suggested based on the finding of a significant increase in hyalocyte number as well as a change in morphology in a rodent model of diabetes (Vagaja et al., 2012). These cells could particularly be important in the pathophysiologic mechanisms responsible for tractional detachment because they have been reported, along with glial cells, in epiretinal membranes from proliferative vitreoretinopathy and proliferative DR (PDR) (Oberstein et al., 2011).

Similar to perivascular macrophages, it remains controversial whether dendritic cells can be found in the retina. A small population of cells was originally identified using various cell markers and was thought to be resident dendritic cells (Lehmann et al., 2010; Xu et al., 2007). However, more recent studies have asserted that these cells are indistinguishable from neighboring microglial and macrophage subpopulations (Dando et al., 2016; McMenamin et al., 2019). Thus, although it remains unclear if dendritic cells play a role in DR, including at later stages and during BRB breakdown, applications of new technologies such as multiplex staining and light-sheet microscopy are currently being used to resolve this controversy (Li et al., 2017; Power and Huisken, 2017).

In addition, even if retinal dendritic cells do exist in the retina, it is not clear the extent to which they would fulfill their classic role as antigen-presenting cells (Dando et al., 2016), particularly given the presence of hyalocytes and perivascular macrophages (Gregerson and Yang, 2003; Lehmann et al., 2010). Indeed, recent single-cell RNA-sequencing work has identified the possibility for macroglia to function as antigen presenters as well (Van Hove et al., 2020). Clearly, there remain many questions surrounding the various putative innate immune cell types in the retina, and more research into all these populations is needed to elucidate their roles in DR pathogenesis. For now, however, much of the research on the cellular immune response is limited to microglia and other immune-competent cells, such as astrocytes and Müller cells.

c. The complement system as a key local player

The final component of the innate immune system is the complement system, which consists of more than 30 proteins that act as sentinels (Walport, 2001; Ricklin et al., 2010). Many stressors activate complement components, while endogenous complement regulatory molecules inhibit the complement activation cascade to maintain homeostasis. In this manner, complement components are activated when they encounter abnormal entities (e.g., pathogens), at which point the amplification of the complement response proceeds through three major pathways (the classical, lectin, and alternative pathways); this results in the subsequent destruction of the entities responsible for the perturbation (Zipfel and Skerka, 2009). The role of the complement system in the pathophysiology of DR has been supported by observations of activation markers in the retinal tissue from late-stage PDR donors. In addition, the identification of complement involvement at earlier stages of the disease (Shahulhameed et al., 2020; Zhang et al., 2002) suggests a role similar to the one it plays during neurodevelopment.

Canonically, complement proteins are thought to be produced by hepatocytes. However, complement components and regulators were recently not only found in the retinas of humans and rodents but also shown to be produced locally, including by microglia, macroglia, and RPE (Anderson et al., 2010; Luo et al., 2011). These findings are consistent with additional functions for the complement system such as the one it plays in synaptic pruning by microglia during normal development (Schafer et al., 2012; Stevens et al., 2007). It is indeed consistent with the key role the complement system plays in protecting the retina from foreign disruptions and environmental alterations. Together the complement system, the resident immune-competent cells, and the physical barriers afford the retina relative immune privilege, or exemption from the systemic immune response (Streilein et al., 2002); this plays a central role in other retinal diseases and was recently reviewed (Murakami et al., 2020).

3. The complex role of the immune response in the early stages of the disease: Preclinical diabetic retinopathy

Preclinical DR begins when diabetes is initially diagnosed and lasts until the first vascular abnormalities are detected, which usually takes many years (Abcouwer and Gardner, 2014). In contrast to its historical categorization as a clinically silent phase, preclinical DR is a period of significant disease progression, primarily through immune dysregulation. Diabetes, like many other stress conditions, initiates the microglial immune response as part of its defense mechanism and “normal” response, a response that is usually counterbalanced by immunosuppressive mechanisms. However, during the lasting environmental and metabolic stress associated with diabetes, these regulating systems are overwhelmed; instead of being progressively resolved, the microglial activation and associated immune response are amplified through the cross-activation of immune-competent macroglia (Liu and Steinle, 2017; Wang et al., 2012, 2011; Yoshida et al., 2004).

This ever-increasing and self-perpetuating inflammatory condition involves a complicated interplay of inflammatory mediators with dire consequences on retinal neurons, function, and barrier integrity, all of which start during preclinical DR. Importantly, recent findings in patients with preclinical DR suggest that they may experience deficits in peripheral vision, night vision, color-hue discrimination, and contrast discrimination (Jackson and Barber, 2010; Trento et al., 2017; Wolff et al., 2015), which may be explained by the neuronal thinning observed in patients before development of vascular abnormalities (Adams and Bearse, 2012; van Dijk et al., 2011). Therefore, the active immune response during this period may explain the symptoms these patients experience and their detectable signs of vision impairment.

a. Early induction of the innate immune response

Diabetes results in the generation of a number of products thought to initiate the immune response, including advanced glycated end products and advanced lipoxidation end products (Steinle, 2020; Stitt, 2003). Alternatively, the dyslipidemia and altered lipid metabolism observed in diabetes have also been proposed as initiators of the innate immune response (Atawia et al., 2020; Eid et al., 2019). A recent study specifically identified monounsaturated oleic acid as a possible source for dyslipidemia-induced DR (Chang et al., 2020).

In the retina, these various diabetic signals bind to the receptors for damage-associated molecular patterns (DAMPs) and pathogen-associated molecular patterns (PAMPs) on microglia and macroglia; this represents one of the major mechanisms for how diabetes activates these cells (Atawia et al., 2020; Eid et al., 2019; Steinle, 2020; Stitt, 2003). Upon activation, microglia proliferate and experience a morphologic change by retracting their processes and becoming more amoeboid (Lim et al., 2019; Wang et al., 2007; Wong et al., 2001). Examination of human diabetic eyes has confirmed this microglial activation early in DR (Zeng et al., 2008), which mirrors the microgliosis observed in diabetic animal models only weeks after diabetes onset (Barber et al., 2005; Krady et al., 2005).

Pharmacologic inhibition of glycated end products dampens microglial activation in diabetic rodents, supporting the notion that glucose plays a role in the genesis and enhancement of the immune response during diabetes (Ibrahim et al., 2011). However, numerous studies have shown that glucose is clearly not the only culprit; for example, immune responses are only partially reduced by pharmacologic reduction of retinal glucose levels (You et al., 2018). It is becoming increasingly clear that glucose is only one of many systemic and local factors, including lipids, metabolites, cytokines, and trophic factors, important for microglial activation. In addition to the nature of the factors themselves, recent studies have suggested that the extent of their fluctuations is particularly important in triggering this activation (Hsieh et al., 2019). Together these studies also suggest that the early metabolic dysregulation in diabetes activates microglia soon after disease onset and well ahead of clinically detectable manifestations.

Activated microglia amplify the innate immune response by releasing inflammatory mediators such as interleukin-1β (IL-1β) and tumor necrosis factor α (TNFα) and recruiting additional immune cells as a mechanism of elimination of the inciting stressor (Ibrahim et al., 2011; Krady et al., 2005; Wang et al., 2007). Although this microglial response is extremely proficient in correcting short-term perturbations and typically works well in maintaining the delicate homeostasis of the retina (Chovatiya and Medzhitov, 2014; Miller et al., 2011), it is clearly less effective under chronic perturbation such as diabetes; during diabetes, activated microglia eventually become active contributors of the overt inflammatory environment. Indeed microglia participate in this unresolved and self-sustained inflammatory response which contributes to DR progression (Ding et al., 2018; Medzhitov, 2008).

Although the impact of the diabetic environment on retinal metabolism has received increasing attention (Eid et al., 2019; Fort et al., 2014), the specific impact on the intracellular metabolism of macrophages and other controllers of innate and adaptive immunity is still unclear. This is particularly important in light of the recent finding that nutrients and other environmental clues can induce profound metabolic reprogramming in macrophages and dendritic cells (O’Neill and Pearce, 2016). Although additional work in the context of diabetes and the retina is sorely needed, a number of reviews have detailed the general interdependence of metabolism with immune response activation (McGettrick and O’Neill, 2020; Ryan and O’Neill, 2020). The need for further investigation of the specific impact of the diabetic environment on retinal immune cells activation is particularly highlighted by the following: the metabolic dysregulated and hypoxic state of diabetes (Atawia et al., 2020; Eid et al., 2019; Klein et al., 2007; Kur et al., 2012; Wong et al., 2002); the specificities of retinal metabolism, such as retinal photoreceptors’ unusual use of aerobic glycolysis (Du et al., 2013; Petit et al., 2018; Winkler et al., 1986); the conversion of glucose to lactate despite the presence of oxygen (Vander Heiden et al., 2009; Warburg, 1956); and the neuron/macroglia metabolic interdependence.

b. Counterbalance by immunosuppressive mechanisms

The activated microglia-initiated inflammatory response is countered by a number of immune-suppressive mechanisms in place to re-establish equilibrium in the retina. These mechanisms are mediated by RPE cells, neurons, macroglia, and microglia themselves and, under normal conditions, maintain the retina in an overall anti-inflammatory state (Cai et al., 2018; Karlstetter et al., 2014; Murinello et al., 2019).

RPE cells, which also make up the outer BRB, express the ligand for CD95 (Fas ligand), which, when bound to CD95 (Fas) on microglia, initiates apoptosis in these cells (M. Chen et al., 2019; Griffith et al., 1995; Jørgensen et al., 1998; Kazama et al., 2008). As with other ligands involved in apoptosis induction, CD95 has a complicated role in retinal physiology and immune suppression and requires tight regulation. Indeed, in addition to targeting microglia, Fas ligand might be involved in killing cells foreign to the eye and preventing neovascularization through regulation of endothelial cells proliferation, which might be important in preventing PDR progression (Murinello et al., 2019); however, Fas ligand has also been demonstrated as a critical factor in retinal detachment-associated photoreceptor cell death, exemplifying the complexity of this regulatory process (Zacks et al., 2007). Apart from Fas ligand, RPE cells also secrete a number of anti-inflammatory mediators, including transforming growth factor β2, to prevent overt inflammation in the retina (Cousins SW et al., 1991; D’Orazio and Niederkorn, 1998; Paglinawan et al., 2003).

Neurons are also incredibly active in immunosuppression by expressing α-melanocortin-stimulating hormone, transmembrane glycoprotein CD200, and CX3CL1 (fractalkine). α-Melanocortin-stimulating hormone binds to microglia and macroglia to inhibit their expression of proinflammatory cytokines, including IL-1β and TNFα (Taylor and Lee, 2010), and has experimentally been shown to preserve BRB integrity and slow DR progression (Cai et al., 2018). CD200, secreted by neurons and endothelial cells, binds to its receptor (CD200R) on microglia to inhibit inflammation (Broderick et al., 2002; Hoek et al., 2000). This is also supported by the genetic and pharmacologic disruption of CD200-CD200R, which increases microgliosis and inflammation (Banerjee and Dick, 2004; Hoek et al., 2000). Fractalkine has similarly been shown to inhibit microgliosis (Cardona et al., 2015). To test the role of fractalkine in microglia-neuron interactions and the neuroinflammation associated with diabetes, fractalkine receptor CX3CR1-null rodents were crossed to the diabetic Ins2^Akita model. These CX3CR1-null diabetic rodents experience enhanced microgliosis compared with their diabetic wild-type littermates (Beli et al., 2016; Cardona et al., 2015). Moreover, CX3CR1-null rodents exhibit both increased neurodegeneration (Cardona et al., 2015) and accelerated vascular abnormalities compared with diabetic controls (Beli et al., 2016). These findings, coupled with the elevated vitreous levels of fractalkine in human DR patients, suggest that neurons attempt, through fractalkine, to limit and possibly resolve the microgliosis that accompanies DR (You et al., 2007).

Microglia also possess mechanisms of “self-control” to limit their own activation and to limit damage associated with unresolved inflammation. When activated, microglia express the mitochondrial translocator protein TSPO (Karlstetter et al., 2014). The use of TSPO agonist XBD173 experimentally reverses microgliosis and decreases levels of pro-inflammatory genes (Karlstetter et al., 2014; Scholz et al., 2015). Furthermore, microglia endogenously produce diazepam-binding inhibitor, whose derivatives bind to TSPO and experimentally dampen the inflammatory response (Wang et al., 2014). These findings suggest a self-regulating mechanism whereby microglia are capable of not only mounting an immune response but also suppressing that inflammation and regaining homeostasis. Together these findings demonstrate how non-immune and immune cells exert immunosuppressive pressures in the attempt to maintain retinal homeostasis.

c. Glia as immune-competent cells

Contrary to how non-immune RPE and neurons apply immunosuppressive pressures, the immune-competent macroglia contribute to the pro-inflammatory retinal environment during diabetes. Given the central role of the innate immune response during diabetes, a more nuanced understanding of astrocytes and Müller cells is necessary in the understanding of DR.

Astrocytes develop along the vasculature in the central nervous system and are part of the glial limitans and NVU (Muoio et al., 2014; Watanabe and Raff, 1988). They are necessary for normal vasculature development in a number of mammalian species (Schnitzer, 1988) and for the continuous maintenance of the vascular network through the production of angiogenic mediators such as vascular endothelial growth factor (VEGF) (Kur et al., 2012). Studies examining the blood-brain barrier found that astrocytes specifically maintain the integrity of the blood-brain barrier by secreting various molecules, including Sonic hedgehog (Alvarez et al., 2011). Indeed, the disruption of this Hedgehog pathway results in increased permeability and decreased tight junction protein expression (Alvarez et al., 2011). Due to all these roles, astrocytes are considered the stewards of the vasculature.

Müller cells are specific to and span all layers of the retina. In this way, Müller cells are able to interact with other retinal cells to maintain homeostasis and metabolism, including glycogen storage and nutritional support (Lindsay et al., 2014; Xue et al., 2015). Müller cells recycle glutamate and neurotransmitters released by neurons, prevent potassium ion accumulation, and control water flux (Freitas et al., 2016; Vogler et al., 2016). Like astrocytes, Müller cells are part of the NVU and are involved in vasculature maintenance. Ablation of Müller cells causes BRB breakdown and has been implicated in DR progression (Abukawa et al., 2009; Coughlin et al., 2017; Fu et al., 2015; Shen et al., 2012).

Like microglia, macroglia become activated in response to perturbations in the environment, including glucose level fluctuations (Picconi et al., 2019). Although microglia are not true glia and this misnomer only artificially categorizes them with astrocytes and Müller cells, it is useful in this context due to their shared response to environmental stress and the immune-competent status of glial cells. Indeed, diabetes activates astrocytes and Müller cells and causes morphologic and functional changes, including the secretion of inflammatory mediators (Gerhardinger et al., 2005; Lieth et al., 1998; Mizutani et al., 1998; Puro, 2002; Rungger-Brändle et al., 2000; Shin et al., 2014). Moreover, a recent single-cell RNA-sequencing examination of macroglia in diabetic rodents has reaffirmed this through the identification of markers of reactive gliosis (Van Hove et al., 2020).

Exposure to the inflammatory milieu that they perpetuate and other metabolic perturbations associated with diabetes results in Müller cell dysfunction and impairs their normal homeostatic functions (Hassan et al., 2017). Recent electron microscopic analysis of human diabetic retinas confirmed Müller cell alterations during the development of DR (Fehér et al., 2018); these findings were suggested by previous studies in diabetic rodents models (Fernandez-Bueno et al., 2017) and highlight the potential key role of glial cell activation in the onset and progression of retinal dysfunction that characterizes DR.

d. The complex interactions of microglia and macroglia

Both microglia and macroglia are activated by environmental stressors such as diabetes; however, this activation, especially when not resolved, may not always be beneficial to the retina. Microgliosis exemplifies the dual effects a continuously activated immune response could have in both limiting and exacerbating DR pathology. Indeed, as previously noted, microglia are inherently heterogenous in nature, under physiologically healthy states (Goldmann et al., 2016; Grabert et al., 2016) and even more so under pathologic conditions. Thus, additional work focusing on untangling their heterogenous responses is needed to better understand their involvement in DR. There is ongoing discussion regarding the most appropriate way to view and conceptualize the varied effects activated microglia can have on the environment (Ransohoff, 2016), but there does appear to be a significant spectrum ranging from the anti-inflammatory slowing of disease to the pro-inflammatory acceleration of disease (Figure 3). Experimentally, activated microglia in diabetic rodents undergo shifts from one response to another (Arroba et al., 2016). The role of activated microglia during DR is therefore in continuous flux and represents a promising area of continued investigation.

Figure 3. Microglial activation during progression of diabetic retinopathy.

The quiescent microglia observed in normal physiology is ramified in form and initially responds by becoming activated and transforming into an amoeboid state. During this appropriate response to diabetes, activated microglia secrete increasing amounts of neurotrophic and anti-inflammatory mediators and help preserve the state of no diabetic retinopathy. Over time and in absence of resolution of the unbalanced environment, the population of M2 appropriately active microglia shifts to a more M1 disease amplifying state, which participates in the progression of DR.

Similar to microglia, macroglial activation results in the secretion of inflammatory mediators such as TNFα, IL-1β, and monocyte chemoattractant protein-1 (MCP-1); this contrasts drastically with their physiologic role of secreting anti-inflammatory mediators (Behzadian et al., 1995; Capozzi et al., 2016; Shin et al., 2014; Yong et al., 2010). Furthermore, in co-culture systems, activated microglia activate Müller cells (Wang et al., 2011), which in turn secrete a number of mediators to reciprocally activate microglia (Liu and Steinle, 2017; Wang et al., 2012; Yoshida et al., 2004). This self-perpetuating cycle involving communication between immune-competent cells can perpetuate unforgivingly until the immune response, initially appropriate for the perturbation, becomes unwieldy and excessive. It has been hypothesized that the transformation from quiescent to activated macroglia may transition with an intermediary state that produces anti-inflammatory and neurotrophic mediators (Figure 4) (Tsai et al., 2018). Although this hypothesis has yet to be confirmed, it is supported by our recent finding that Müller cells express high levels of αA-crystallin, a highly protective chaperone, in response to diabetes during preclinical DR (Figure 5). Although this high level of expression continues to be found in donors with DR, post-translational modifications are associated with a loss of function in these donors and are consistent with the increased neurodegeneration and neuroinflammation observed (Ruebsam et al., 2018). Our most recent work also suggests that αA-crystallin overexpression by Müller glial cells dampens their activation and self-perpetuation of the inflammatory cycle (data not shown). However, this regulation is controlled by specific post-translational modifications, including that shown to be lost in DR, consistent with the increased activation observed in DR patients. Recent single-cell RNA-sequencing work has identified a number of macroglial subpopulations that may represent these states; however, this study examined diabetic rodents at only a single time-point (Van Hove et al., 2020). More work using techniques such as single-cell RNA sequencing is underway to confirm the development of these macroglial subpopulations over the different stages of DR progression and to better characterize the impact of diabetes on macroglia.

Figure 4. Macroglial activation during progression of diabetic retinopathy.

Quiescent macroglia become activated in response to diabetes and while first involved in the adaptive protective response, will ultimately contribute to diabetic retinopathy pathology. The exact nature and duration of this intermediary state between quiescent macroglia and activated macroglia remains to be fully elucidated but may be a key element in our quest to new avenues for DR treatment.

Figure 5. AlphaA-crystallin chaperone protein while increasingly expressed is decreasingly phosphorylated on a key regulatory site in DR donors.

AlphaA-crystallin total protein level was assessed by western-blot and showed to be increased in diabetic donors with or without DR (left). Concomitantly, phosphorylation on T148, a phosphosite controlling its protective function was assessed by multiple reaction monitoring and shown to be dramatically reduced in DR donors. (Adapted from Ruebsam et al., 2018)

e. Mechanisms: The inflammatory mediators

As we have previously discussed, the various immune-competent and non-immune cells of the retina secrete pro- and anti-inflammatory mediators in response to different stimuli (Capozzi et al., 2016; Ibrahim et al., 2011; Paglinawan et al., 2003; Shin et al., 2014). With the retina normally maintained in an overall anti-inflammatory state (Cai et al., 2018; Murinello et al., 2019), the relative increase of pro-inflammatory mediators in diabetic human retinas highlights an important shift in the overall inflammatory state (Loukovaara et al., 2015; Murugeswari et al., 2008; Patel et al., 2008; Sakata et al., 2004; Sundstrom et al., 2018). This shift in the inflammatory milieu with DR progression is thought to involve the cross-activation of macroglia and microglia via inflammatory mediators. However, with human studies generally limited to samples gathered from vitrectomies (Sakata et al., 2004), the inflammatory environment of the human retina during preclinical DR remains opaque. Because diabetic rodents experience increasing levels of pro-inflammatory mediators early in disease (McVicar et al., 2011; Zhou et al., 2012), additional work on the retinal transcriptome and ocular proteome in human samples would deepen our understanding of the extent of involvement these mediators have during preclinical DR.

i. Cytokines

Cytokines are an important group of inflammatory mediators that are not only a consequence of activated microglia and macroglia but are also initiators of escalating DR inflammation. In the vitreous of diabetic patients, TNFα is one of the earliest cytokines detected and increases with disease progression (Demircan et al., 2006; Doganay et al., 2008; Wu et al., 2020). In vitro experiments have shown that TNFα stimulates microglia, macroglia, and endothelial cells to secrete VEGF (Stone et al., 1995; Wang et al., 2010; Yoshida et al., 2004). Importantly, VEGF is a potent cytokine whose inhibition via inhibitors such as bevacizumab has slowed DR progression and improved visual outcomes in human patients (Do et al., 2012; Elman et al., 2010; Ip et al., 2012; Nguyen et al., 2010; Zechmeister-Koss and Huić, 2012). Interestingly, VEGF inhibition suppresses the immune system and lowers levels of other cytokines, such as TNFα, IL-1β, interferon-γ, and chemokine macrophage inflammatory protein-1α (Gnanasekaran et al., 2020; He et al., 2015; Sivaprasad et al., 2017). Thus separate from their anti-angiogenic effects, VEGF inhibitors may derive therapeutic efficacy by limiting the innate immune response and the activation of microglia and macroglia (Stone et al., 1995; Wang et al., 2010; Yoshida et al., 2004). Furthermore, this raises the possibility of other cytokines as targets for DR treatment, especially in those patients who do not respond to anti-VEGF therapies (Bressler et al., 2014; Elman et al., 2010; Nguyen et al., 2010). Indeed, a recent publication identified inflammation as a mechanism behind those with diabetic macular edema (DME) resistant to anti-VEGF therapies (Arima et al., 2020). While the potential of TNFα as a therapeutic target continues to be evaluated, its role in refractory DME and PDR has been the topic of several prospective interventional studies and clinical trials (NCT00695682, NCT00505947) involving TNFα-specific antibodies. One study reported potentially encouraging results relative to visual acuity and retinal function (Sfikakis et al., 2010), whereas another reported potential immunogenicity and toxicity (Giganti et al., 2010). Due to these mix results and the studies’ small sample size and targeted population, further examination is required.

Similar to how TNFα results in the secretion of VEGF, which in turn stimulates the release of TNFα by microglia, macroglia, and endothelial cells (Stone et al., 1995; Wang et al., 2010; Yoshida et al., 2004), an interesting area of continued research is the cross-activation of cytokines during DR. For example, the presence of IL-1β in diabetic humans and rodents (Yang Liu et al., 2012; Yego et al., 2009; Zhou et al., 2012) triggers the release of the pro-inflammatory cytokine IL-8, which is elevated in advanced DR patients (Doganay et al., 2008; Hernández et al., 2005; Liu et al., 2014; Raczyńska et al., 2018; Yoshida et al., 2004).

In addition to these well-known cytokines, we have also found indoleamine 2,3-dioxygenase to be elevated in diabetic eyes (Figure 6) (Nahomi et al., 2018). Although the origin of indoleamine 2,3-dioxygenase remains unclear, studies examining human diabetic donor tissues have implicated microglia, Müller cells, and endothelial cells (Hu et al., 2017; Nahomi et al., 2018). We further identified this elevation of indoleamine 2,3-dioxygenase to be associated with increased levels of the pleiotropic cytokine interferon-γ, which is a known regulator of vascular endothelial cell survival (Figure 7) and is also elevated in streptozocin-induced diabetic rodents (Özay et al., 2020). Together these human, rodent, and in vitro lines of evidence suggest that the preclinical DR retinal environment is suffused with inflammatory cytokines that further the activation of and are further promoted by microglia and macroglia.

Figure 6. IDO expression is elevated in the retina of diabetic donors with DR.

Immunohistochemistry of retinal sections shows IDO (red) partially colocalizing with CD-34 (green) in the capillary endothelium (arrowheads) of the retina of diabetic donors with DR (right). IDO-positive staining was also observed in the capillary endothelium of the retina of diabetic donors without retinopathy (middle) and nondiabetic donors (left), but the signal appears diffuse and less intense (arrowheads). Additionally, some of the IDO signal in the inner retina of diabetic donors with retinopathy appears not to colocalize with CD-34 (arrows), suggesting expression in glial cells. Nuclei are stained with DAPI (blue). (Adapted from Nahomi et al., 2018)

Figure 7. High levels of IFN-γ in the retina of diabetic donors with DR.

IFN-γ levels were measured by ELISA in retinal homogenates from age-matched nondiabetic (n = 9), diabetic (without retinopathy, n = 8), and diabetic (with retinopathy, n = 7) donors. Scale bar: 20 μm. *P < 0.05; NS, not significant. (Adapted from Nahomi et al., 2018)

ii. Chemokines

Chemokines are another class of inflammatory mediators that attract immune cells to further promulgate the immune response. Microglia undergo chemotaxis via their exposure to a number of chemokines, which, like MCP-1, are elevated in diabetic human and rodent eyes (Hernández et al., 2005; Wu et al., 2020). In response to hyperglycemia, Müller cells produce MCP-1 (Dong et al., 2012; Rangasamy et al., 2014; Wang et al., 2012), which activates and recruits microglia to the hypoxic interfaces of OIR rodent retinas (Davies et al., 2008, 2006). Because OIR models of hypoxia are generated from postnatal day 7 exposure to hyperoxic conditions and are used to model the neovascularization observed in PDR (Olivares et al., 2017), these findings of microglial accumulation may help to explain the microgliosis observed in human PDR (Zeng et al., 2008). Unfortunately, this activation and accumulation of microglia may contribute to disease progression (Ritter et al., 2006), because inhibition of MCP-1 and macrophage inflammatory protein-1α significantly decreases neovascularization in OIR rodents (N. Wang et al., 2019; Yoshida et al., 2003).

It remains unclear if the elevated levels of other chemokines such as chemokine ligand 4 and C-X-C motif ligand 9 and 10 observed in DR patients are deleterious (Nawaz et al., 2013). Although these chemokines have been shown to stimulate the innate immune response and contribute to the inflammatory retinal environment (Nawaz et al., 2013), not all chemokines activate the immune response. As previously mentioned, chemokine CX3CL1 (fractalkine) inhibits microgliosis and supports an immunosuppressed environment (Beli et al., 2016; Cardona et al., 2015; Combadière et al., 2007; You et al., 2007). Therefore a more nuanced and specific understanding of how different chemokines affect the diabetic human eye would further our understanding of the pathogenesis during preclinical DR.

4. Mechanisms of preclinical diabetic retinopathy neurodegeneration

Before the presence of any detectable vascular abnormality, preclinical DR patients experience a number of visual dysfunctions measurable through psychophysical testing, including deficits in peripheral vision, night vision, color-hue discrimination, and contrast discrimination (Jackson and Barber, 2010; Trento et al., 2017; Wolff et al., 2015). These functional deficits have further been confirmed and localized by multifocal electroretinopathy to distinct retinal areas in preclinical DR patients (Di Leo et al., 1994; Juen and Kieselbach, 2011; Reis et al., 2014; Tyrberg et al., 2011). Further structural analysis with high-resolution spectral-domain optical coherence tomography has demonstrated progressive thinning of the nerve fiber layer and ganglion cell layer of the retina in multifocal electroretinography–abnormal regions, thereby linking vision impairment with functional and structural abnormalities (Chhablani et al., 2015; Juen and Kieselbach, 2011; Sohn et al., 2016; Tyrberg et al., 2011). Moreover, not only did these neuronal structural abnormalities overlap spatially with the functional deficits found by psychophysical testing and multifocal electroretinography, but these defects found during preclinical DR seem to track with the locations where future vascular lesions occur (Dodo et al., 2015; Harrison et al., 2011; Ng et al., 2008). These findings suggest that neurodegeneration contributes to visual dysfunction in preclinical DR and that there may exist a relationship between neurodegeneration and vascular abnormalities in DR.

a. Direct effect of activated microglia and macroglia on retinal neurodegeneration

Certainly, these findings in human patients have called into question the classic understanding of vascular abnormalities causing neuropathy (Archer, 1999; Bresnick and Palta, 1987; Kohner et al., 1999). Consistent with this manifestation, retinas of diabetic rodents also experience neuronal thinning, detected by spectral-domain optical coherence tomography (Alves et al., 2018), in locations with increased aggregation of activated microglia (van Dijk 2009, van Dijk 2010, van Dijk 2012). Because microglia possess the machinery to kill neurons and prune synapses and because their activation precedes neuronal thinning (Barber et al., 2005), the possibility that activated microglia may contribute to neurodegeneration continues to be an important avenue to explore. Indeed, a number of investigators have found that microglia are able to phagocytose impaired but functional neurons (Brown and Neher, 2014; Kawabori et al., 2015; Kim et al., 2017; Zhao et al., 2015), perhaps explaining the accelerated loss of neurons observed early in diabetes (Abu El-Asrar et al., 2004; Barber et al., 1998; Hammes et al., 1995; Martin et al., 2004; McVicar et al., 2011).

The other immune-competent cell implicated in DR neurodegeneration is the Müller cell. Because Müller cells are normally responsible for maintaining appropriate glutamate and ion levels, when Müller cells are unable to properly maintain homeostasis during diabetes (Chen et al., 2014; Puro, 2002), diabetic human and rodent retinas can experience elevated levels of glutamate (Ambati et al., 1997; Lieth et al., 1998; Pannicke et al., 2006). Consistent with this notion, hyperglycemia or IL-1β activates Müller cells in cell culture and results in increased glutamate and ion levels (Chen et al., 2014; Puro, 2002), whereas chemical inhibition of Müller cell activation decreases glutamate levels and improves ganglion cell health (Ganesh and Chintala, 2011). These experiments suggest that activation of Müller cells early in diabetes can result in alterations of glutamate recycling, potentially leading to regional excitotoxicity and neurodegeneration in preclinical DR. Mechanistically, a number of recent studies have begun to establish the protective role of neuropeptides, such as neuropeptide Y and substance P, for neurons exposed to glutamate excess (Ou et al., 2020, 2019). However, neuropeptides, just like growth factors, play a complex role in DR pathophysiology, because neuropeptide Y has been shown to potentially play an important role in gliosis and its involvement in fibrocellular membranes associated with PDR (Milenkovic et al., 2004). As previously discussed, microglia and macroglia can cross-activate and regulate each other in many ways, including to potentially exacerbate excitotoxicity and neuronal phagocytosis by microglia (Figure 8; (Wang et al., 2011). Consistent with this, the genetic deletion of Toll-like receptor 4 in Müller cells not only decreases pro-inflammatory cytokines in the retina, but it also results in decreased neurodegeneration (Liu and Steinle, 2017).

Figure 8. Activation and self-perpetuation of the response of local immune competent cells in diabetic retinopathy.

Diabetes activates immune competent cells including micro- and macroglia (astrocytes and Müller glial cells) via multiple mechanisms, potentially inducing a self-perpeutating and amplification loop. Diabetes can directly induce activation of quiescent microglia (orange) leading to increase in number, change in morphology, and increased production of pro-inflammatory mediators. Similarly, diabetes also induces macroglia (green) to undergo activation and result in increased production of pro-inflammatory mediators. Finally, endothelial cells and neurons (pink) can also respond to diabetes by producing pro-inflammatory mediators. Any and all of these responses participate in the shift from immune-privileged to a highly pro-inflammatory ocular environment, and amplifies the activation of micro-and macroglia in an autocrine, and paracrine fashion. Once set in motion, this creates a loop in which de-activation of these immune competent cells is rendered highly difficult even when the trigger (i.e. suboptimal controlled diabetes) has been “removed”.

b. Neurodegeneration and tissue homeostasis: Microglia, macroglia, and growth factors

Another area of increasing interest is the secretion of growth factors by microglia and macroglia and their role in promoting cell survival. Not only are the growth factors brain-derived neurotrophic factor (BDNF), nerve growth factor, and pigment epithelium–derived factor (PEDF) depressed in diabetic human and rodent retinas (Boehm et al., 2003; Boss et al., 2017; Yanling Liu et al., 2012; Mysona et al., 2014; Ola et al., 2013), but there is an accumulation of their corresponding pro-forms (Ali et al., 2011). This observation is all the more important because these abundant pro-forms, in addition to not being neuroprotective, have been shown to be pro-inflammatory (Barcelona et al., 2016; Elshaer et al., 2013; Lebrun-Julien et al., 2010) and even further neurodegeneration and vasculopathies (Matragoon et al., 2012; Mohamed et al., 2018). This pathogenic effect has been confirmed by studies showing that the inhibition of these pro-forms is protective against neurodegeneration and vasculopathy in diabetic and OIR rodent models (Barcelona et al., 2016). Consistent with these findings, experimental increase of BDNF and growth hormone–releasing hormone is neuroprotective and slows DR progression (Suzumura et al., 2020; Thounaojam et al., 2017). However, a recent review of the extensive BDNF literature appropriately highlighted the complexity of manipulating such systems. Afarid et al. indeed nicely summarized the divergent reports relative to the success of BDNF treatment on DR progression and the innate immune response; these may depend on the concentration of BDNF but also the relative engagement and loss of sensitivity of the receptor, especially as a function of the different stages of DR (Afarid et al., 2020).

Interestingly, insulin as a growth factor is intimately involved in diabetes and is itself depressed during disease. Furthermore, local administration of insulin in diabetic eyes results in decreased immune response (Table 2) (Fort et al., 2011). Similarly, PEDF injections into diabetic or OIR rodent eyes decrease levels of pro-inflammatory mediators TNFα, VEGF, and MCP-1 (Zhang et al., 2006). In contrast, knock-down of PEDF results in increased secretion of VEGF and TNFα by cultured Müller cells, suggesting that at least part of PEDF-mediated anti-inflammatory effects are via Müller cells (Zhang et al., 2006). Microglia also appear to be involved, because PEDF challenge decreases microgliosis in streptozocin-induced diabetic rodents (Yoshida et al., 2009). Thus growth factors, although still challenging to use, have become increasingly important in the understanding of retinopathy and are exciting avenues of potential therapies, including as a means to temper the increasingly inflammatory environment of the retina.

Table 2. Top ten pathways affected by diabetes and reversed by ocular insulin administration.

List of the top ten pathways obtained considering only the list of genes from the microarray analysis that were affected by diabetes and reversed by local insulin administration using MetaCore^™ (Genego Inc.) software. Pathways are ranked based upon p-value (Adapted from Fort et al., 2011).

ranking	Maps	min(pValue)
1	Immune response_Classical complement pathway	4.112E-12
2	Cell adhesion_Role of tetraspanins in the integrin-mediated cell adhesion	5.144E-09
3	Immune response_Lectin induced complement pathway	9.611E-08
4	Regulation of lipid metabolism_Regulation of lipid metabolism via LXR, NF-Y and SREBP	1.089E-07
5	Development_Role of IL-8 in angiogenesis	5.124E-07
6	Cytoskeleton remodeling_Regulation of actin cytoskeleton by Rho GTPases	8.567E-06
7	Immune response_IL-22 signaling pathway	7.801E-05
8	Cytoskeleton remodeling_Neurofilaments	1.972E-04
9	Immune response_IL-5 signalling	4.053E-04
10	Development_Slit-Robo signaling	4.831E-04

c. Neurodegeneration and tissue homeostasis: Microglia, macroglia, and complement activation

In addition to immune-competent cells and their inflammatory and growth mediators, the complement system is also intimately involved and active during preclinical DR. Although complement components are absent in non-diabetic eyes, their levels increase in the eye (Muramatsu et al., 2013; Shahulhameed et al., 2020; Zhang et al., 2002) and the periphery (Huang et al., 2018) during preclinical DR. Indeed, our own unpublished data reveal a successive increase in complement component RNA in eyes with no diabetes, preclinical DR, and non-proliferative DR (NPDR) (Figure 9). One of the complement components that is most upregulated in the human eye during DR is C3, which is the central component of the three major complement-activating cascades (unpublished data). This correlates with the increased serum levels of C3a, which is generated from the cleavage and activation of C3, and additionally points to the potential involvement of this anaphylatoxin in DR (Figure 10) (Lingjun Zhang et al., 2016). We previously demonstrated that activation of C3-related anaphylatoxin could play a key role in other models of neurodegeneration such as allodynia (Figure 11) (Xu et al., 2018), supporting a similar role of complement system activation in DR-related neurodegeneration. This aligns with the role C3 signaling plays in synaptic pruning (Schafer et al., 2012); however, in this case it is pathologic. Interestingly, microglia are the only central nervous system cells to express the receptor (CR3) and so are thought to be solely responsible for the phagocytosis of cells (including neurons) coated with C3b. With peripheral C3 levels elevated in diabetes, it remains unclear whether complement infiltration into the retina plays a role in DR (Rasmussen et al., 2018).

Figure 9. Complement system components are found in increasing levels in the neuroretina of donors with non-proliferative diabetic retinopathy compared to noon-diabetic donors (age, gender and race matched).

Immunohistochemical staining was performed on retinal sections of human donors without diabetes (Left) and with diabetes and non-proliferative diabetic retinopathy (NPDR, middle). The ganglion cell layer (GCL), inner nuclear layer (INL), and outer nuclear layer (ONL) are displayed in the first two panels. Agrin (red) a marker of blood vessels, and the complement system regulator CD59 (green) are shown with nuclei counterstain (blue). The right panel is a higher magnification image of the central panel demonstrating the presence CD59 away from agrin, in the neuroretina tissue.

Figure 10. Measurement of complement activation products C3a, C4a and C5a in human serum samples.

Levels of C3a (A) and C5a (B), but not C4a (C) were significantly higher in the DR group compared to age-matched healthy controls. Data were mean ± SD. Each dot represents one patient (Adapted from Lingjun Zhang et al., 2016).

Figure 11. C3 activation is important in paclitaxel-induced mechanical allodynia in CIPN.

A. Complement is activated after paclitaxel administration. WT or C3 KO rats were i.p. injected with DMSO (Vehicle) or 1 mg/kg paclitaxel (in DMSO according to the instructions of the manufacturer, Tocris, Bristol, UK) for 4 days (day 1 to 4), and then sera were collected on day 5 when the behavioral tests confirmed the development of mechanical allodynia and probed with an anti-C3 IgG to assess complement (C3) activation. There was a significant reduction of the α band (~130KDa) of C3 and an increased α2 chain of iC3b (~40 KDa) in the paclitaxel-treated WT rats. No C3 protein was detectable in sera from the C3 KO rats. B. C3 KO rats showed increased paw withdrawal threshold (PWT). WT and C3 KO rats (n=10/group) were injected with paclitaxel for 4 days (day 1 to 4), PWT was assessed on days 0, 7 and 11.; * p<0.05. (Adapted from Xu et al., 2018)

Increased levels of another anaphylatoxin, the C5-derived activation product C5a, have been reported in the vitreous of PDR patients (Muramatsu et al., 2013). C3a and C5a, as chemoattractants, are notorious for their ability to induce overwhelming inflammatory responses and activate phagocytes (i.e., microglia); both are elevated in the vitreous of diabetic human eyes (Sacks, 2010; Lingjun Zhang et al., 2016). Like with the C3 receptor, the C5a receptor is also present in microglia, and was shown to be required for their activation (Song et al., 2017). Furthermore, both astrocytes and macroglia express C5a receptors and upregulate this receptor in response to elevated glucose levels (Cheng et al., 2013). So, upon C5a binding, astrocytes and Müller cells become activated and increase their production of pro-inflammatory cytokines such as IL-6 and VEGF (Cheng et al., 2013; Gasque et al., 1995). Together this suggests that complement plays a critical role in activating both microglia and macroglia during preclinical DR and contributes to the increasingly pro-inflammatory milieu. Although these findings suggest that complement may be upstream of immune-competent cells, other investigations have found inflammatory mediators to cause the production of complement components via microglial activation (Amadi-obi et al., 2012; Luo et al., 2013). Regardless of whether complement fragments initiate gliosis or vice versa, these two components of the innate immune system are clearly important during DR and offer the possibility of a combinatorial targeting approaching for DR therapies (Luo et al., 2013).

5. Diabetes, blood-retinal barrier, and the neurovascular unit

BRB breakdown is a hallmark event during preclinical DR (Miyamoto et al., 1999; Schröder et al., 1991) and was recently reviewed (Subauste, 2019). Briefly, during diabetes, macroglia secrete increasing amounts of VEGF, which compromises barrier integrity (Stone et al., 1995; Wang et al., 2010), in the context of amplified microglial activation and levels of pro-inflammatory mediators (Jo et al., 2019). In response to this environment, endothelial cells express cell adhesion molecules (CAMs), including ICAM-1, VCAM-1, E-selectin, and P-selectin (Arita et al., 2013; Capitão and Soares, 2016; Clauss et al., 1990; Lu et al., 1999; Miyamoto et al., 2000; Noda et al., 2014), some of which have been reported to be elevated in human NPDR eyes (McLeod et al., 1995). CAMs are the anchors by which circulating leukocytes, via their expression of integrins CD11a, CD11b, and CD18, bind to vascular endothelial cells (Barouch et al., 2000; Miyamoto et al., 1999). So, as circulating leukocytes bind to endothelial cells on the luminal side, they too release pro-inflammatory and cytotoxic mediators, further contributing to the local pro-inflammatory environment (Joussen et al., 2001; Park et al., 2016; Schröder et al., 1991). This sustained inflammation impairs tight junction function and pericyte and endothelial health, eroding barrier integrity (Joussen et al., 2001; Ogura et al., 2017; Platania et al., 2019; Schröder et al., 1991; Subauste, 2019). These findings are replicated only weeks into diabetes onset in rodents (Barouch et al., 2000; Miyamoto et al., 1999; Noda et al., 2012) and suggest that BRB breakdown may occur early in human retinas. Consistent with this, inhibition and genetic deletion of CAMs and integrins result in significantly reduced BRB breakdown (Barouch et al., 2000; Joussen et al., 2004; Miyamoto et al., 1999). Similarly, suppression of cytokine and chemokine activity protects BRB integrity (Jo et al., 2019; Omori et al., 2018; Platania et al., 2019; Valle et al., 2019; H. Wang et al., 2019). Studies combating inflammatory mediators and the complement system have successfully demonstrated the importance of the immune response in BRB breakdown as well (He et al., 2015; Joussen et al., 2003; Wang et al., 2010; Yun et al., 2017; L. Zhang et al., 2016). In addition, growth factors such as PEDF are important in maintaining BRB function (Barnstable and Tombran-Tink, 2004; Yoshida et al., 2009). Remarkably, the topical application of PEDF to diabetic rodents is neuroprotective, increases tight junction expression, decreases microgliosis, decreases pro-inflammatory mediator production, and decreases vascular permeability (Yanling Liu et al., 2012). Together this highlights the need for continued research efforts in this area to clinically treat BRB breakdown in humans with preclinical DR.

Because the BRB is part of the NVU, breakdown of the former includes impairment of the latter. The NVU importantly autoregulates the perfusion supply to match metabolic demand (Metea and Newman, 2007; Newman, 2013). Pre-diabetic and preclinical DR patients experience impaired autoregulation with a 50% reduction in their vasomotor response to flickering light or hyperoxia (Garhöfer et al., 2004; Lott et al., 2015, 2012). Interestingly, flicker-evoked vasodilation was similarly decreased by 55% in healthy non-diabetic individuals experimentally exposed to short periods of hyperglycemia (Dorner et al., 2003). This suggests that hyperglycemia may be responsible for impairment in the NVU vasomotor response and that the localized hypoxia ultimately contributes to the development of vascular abnormalities such as venous beading (Klein et al., 2007; Kur et al., 2012; Wong et al., 2002). In addition, areas of the retina with impaired flicker-evoked vasodilation correspond to neuroretinal abnormalities detected on multifocal electroretinography, suggesting that NVU impairment may also contribute to preclinical DR neurodegeneration (Lecleire-Collet et al., 2011; Newman, 2015). Therefore, far from being devoid of activity, preclinical DR is a period when an increasingly rampant immune response may contribute to neurodegeneration and functional deficits. Furthermore, these active processes continue to accelerate into the next phases of DR and are important in the vascular abnormalities that have classically defined the disease.

6. Emergence of the role of the innate immune response in diabetic retinopathy progression

DR has long been classified as a microvascular disease with pathophysiology thought to coincide with vascular abnormalities. Certainly, vasculopathy does cause neurodegeneration, vision impairment, and immune dysregulation. However, as we have previously detailed, the innate immune response is an independent driver of DR pathology. Thus, as DR progresses to NPDR and PDR, it becomes more difficult to definitively define causation. Regardless, this section will focus on the role of the innate immune response during these later stages of DR and characterize how it may interact with the various manifestations of disease.

a. Non-proliferative diabetic retinopathy

NPDR begins with the first clinically detectable sign of vascular abnormality (i.e., microaneurysm) and is further categorized as mild, moderate, and severe based on the degree of vasculopathy (Wilkinson et al., 2003). Microaneurysms are due to the pathologic loss of pericytes and endothelial cells, which compromises the structural integrity of capillaries. Experimental models have demonstrated that the immune response causes apoptosis of pericytes and endothelial cells (Barouch et al., 2000; Joussen et al., 2004; Miyamoto et al., 1999; Schröder et al., 1991). A related avenue of investigation has detailed loss of cell-cell interactions between components of the NVU, including those between pericytes and endothelial cells, as contributors to NPDR progression (Klein et al., 2007; Kur et al., 2012; Roy et al., 2015; Stitt et al., 2016; Wong et al., 2002). With breakdown of the BRB and infiltration by peripheral immune cells and proteins, the immune response becomes even more amplified during NPDR (Kur et al., 2012; Wong et al., 2002). Consistent with this, microgliosis increases commensurate to vascular pathology in diabetic human and rodent retinas (Barber et al., 2005; Davies et al., 2006; Karlstetter et al., 2015; Zeng et al., 2008, 2000).

Complement activity increases too during NPDR in diabetic retinas (Figure 9) (Frederich et al., 1995; Lingjun Zhang et al., 2016). These complement proteins can be assembled into the membrane attack complex, which is postulated to cause pericyte and neuron death (Li et al., 2012; Lueck et al., 2011). In addition, sublytic membrane attack complexes can also form and become generators of pro-inflammatory and neurotoxic mediators (Lueck et al., 2011). This complement-driven pathology may be in part due to the glycation and inactivation of complement inhibitors such as CD59 (Qin et al., 2004; Zhang et al., 2002). Interestingly, an experimental increase of CD59 inhibits membrane attack complex formation (Bora et al., 2010; Harhausen et al., 2010) and reduces vasculopathy and neurodegeneration in diabetic rodents (Adhi et al., 2013). However, these protective effects occur with a concurrent increase in retinal microgliosis (Adhi et al., 2013), which has been implicated in pericyte loss (Ding et al., 2017; Mazzeo et al., 2017). Thus the complement system and its interaction with immune cells are promising areas of continued investigation.

The vascular abnormalities defined as part of NPDR amplify the neurodegeneration observed in preclinical DR (Barber et al., 2005, 1998; Chhablani et al., 2015; Juen and Kieselbach, 2011; Tyrberg et al., 2011). Indeed, microglia and macroglia are further activated by vascular perturbations, which result in the additional release of a number of neurotoxic mediators that can directly lead to neuronal death (Krady et al., 2005). Inhibition of microgliosis by minocycline or adenosine A_2a receptor not only decreases the neurotoxic environment but is also neuroprotective (Aires et al., 2019; Krady et al., 2005). Despite these findings in rodent models, a clinical trial examining microglial inhibition by doxycycline did not improve visual or anatomic outcomes in patients (Scott et al., 2014). Truly, the relationship between neurodegeneration and vasculopathy still remains opaque. As stated previously, neurodegeneration and loss of visual function can be detected in preclinical DR (Table 1) (Barber et al., 2005, 1998; Bogdanov et al., 2014). However, other patients experience no visual symptoms even though they do have detectable vascular abnormalities (Wilkinson et al., 2003). Therefore it remains uncertain if neurodegeneration always precedes vascular abnormalities or if they occur in parallel (Hernández et al., 2017; Santos et al., 2017).

Table 1. Categorizations of the vascular, inflammatory and neuronal aspects of Diabetic Retinopathy.

The first row across lists the different clinical diagnoses of diabetic retinopathy progression. Vascular signs are assigned in the second row according to clinical categorization. Neuronal and immune findings are attributed to the different grades in rows 3 and 4, respectively.

Clinical Diagnosis	No Diabetes	Diabetes without retinopathy	Mild NPDR	Moderate NPDR	Severe NPDR	PDR
Vascular	No signs	No signs	Microaneurysm	Microaneurysm Dot blot hemorrhages Venous beading Cotton wool spots Hard exudates	Microaneurysms (4 quadrants) Venous beading (2 quadrants) Microvascular abnormalities	Neovascularization Vitreous hemorrhages Fibrovascular proliferation Retinal detachments
Neuronal	Neuropreservation	Neurodegeneration	Neurodegeneration	Neurodegeneration	Neurodegeneration	Neurodegeneration
Immune	Quiescent microglia Anti-inflammatory mediators Neurotrophins ↑ No complement fragments	Microgliosis Cytokines ↑ Proforms ↑ Complement fragment ↑	Microgliosis Cytokines ↑ Proforms ↑ Complement fragment ↑	Microgliosis Cytokines ↑ Proforms ↑ Complement fragment ↑	Microgliosis Cytokines ↑ Proforms ↑ Complement fragment ↑	Gliosis Systemic cytokines ↑ Systemic leukocytes ↑ Systemic complement ↑

b. Diabetic macular edema

During NPDR, many patients experience DME, which is clinically observed as retinal thickening with hard exudates (Wilkinson et al., 2003). These findings are thought to be a direct consequence of lipoprotein leakage, which causes separation of the inner retinal layers from the RPE (Lang, 2012). DME is responsible for most of the vision loss in individuals with DR (Duh et al., 2017; Rübsam et al., 2018). This was recently reviewed (Murakami et al., 2020), but it is worth highlighting the involvement of the immune system in its development. Aside from impairment of the anatomic barriers, Müller cells are hypothesized to undergo excessive swelling from fluid leakage/accumulation and contribute to DME (Xi et al., 2005). In addition, phagocytosis of debris by microglia contributes to thickening of the retina by creating additional pockets of fluid and cellular debris (Levin, 2009; Murakami et al., 2020). Interestingly, although earlier studies have found the anti-VEGF therapy bevacizumab to be overall superior to steroids (Gillies et al., 2014), a more recent study suggested that intravitreal steroid injections result in improved outcomes compared with anti-VEGF injections for DME patients (Ceravolo et al., 2020). Another recent study (Arima et al., 2020) may explain this efficacy of steroids by demonstrating the involvement of the immune response in BRB breakdown. Regardless, these studies certainly demonstrate the close involvement of the immune system and underscore the importance of the immune response during various stages of DR, including DME.

One of the most interesting immunologic markers of DME progression is the cytokine IL-6, whose levels correlate with severity (Funatsu et al., 2002; Liu et al., 2015; Wu et al., 2020). Although research is currently underway to determine how IL-6 contributes to DME progression, its receptor (IL-6R) has been identified on Müller cells and endothelial cells, cell-types implicated in DME pathogenesis (Izumi-Nagai et al., 2007; Zhao et al., 2014). IL-6 is produced in the eye by microglia and macroglia and is produced in the periphery by lymphocytes, monocytes, and endothelial cells (Izumi-Nagai et al., 2007; Kishimoto, 1989; Krady et al., 2005; Zhao et al., 2014). However, unlike vitreous levels, serum IL-6 levels in PDR patients are comparable with those of non-diabetic patients (Kishimoto, 1989; Mocan et al., 2006). Therefore, although systemically produced IL-6 is known to be capable of crossing the blood-brain barrier as an acute phase reactant, its elevation in the retina is most likely from secretion by retinal immune-competent cells. In vitro evidence suggests that IL-6 protects Müller cells from hyperglycemia via a VEGF-dependent mechanism (Coughlin et al., 2019); this may help to explain why steroids are neuroprotective but detrimental to Müller cells (Pereiro et al., 2018). This aligns with the finding that IL-6 reprograms Müller cells for retinal regeneration in zebrafish; however, IL-6 has also been implicated in contributing to choroidal neovascularization (Izumi-Nagai et al., 2007; Zhao et al., 2014). Therefore IL-6 is an intriguing cytokine in DR pathogenesis, and the current evaluation of IL-6 and IL-6R antibodies in clinical trials NCT02842541 and NCT02511067 in DME patients is incredibly exciting.

c. Proliferative diabetic retinopathy

Neovascularization is the hallmark of PDR, and the use of anti-VEGF therapies has slowed disease progression in 50% to 87.5% of patients (Osaadon et al., 2014; Sivaprasad et al., 2017; Writing Committee for the Diabetic Retinopathy Clinical Research Network et al., 2015). Other cytokines such as TNFα have also been shown to contribute to neovascularization, and the genetic and pharmacologic inhibition of TNFα activity similarly retards DR progression (Aveleira et al., 2010; Gardiner et al., 2005; Santos et al., 2012; Yoshida et al., 1999). Not surprisingly, the human and rodent retina during PDR contains the highest levels of pro-inflammatory mediators and activated microglia (Barber et al., 2005; Davies et al., 2006; Karlstetter et al., 2015; Stone et al., 1995; Wang et al., 2010; Zeng et al., 2008, 2000). Interestingly, activated microglia spatially overlap with neovascularization in OIR rodents (Davies et al., 2006), and myeloid progenitor transplantation improves that neovascularization (Ritter et al., 2006). This emphasizes the phenotypic difference between activated microglia and transplanted progenitors and suggests that microglia have the capacity of playing either an antagonistic or protective role (Figure 3).

Like microglia, the gliotic response of macroglia seems to be at its peak during PDR. This is most clearly represented by their response to the inflammatory environment by adopting a “fibroblast-like” role and creating dense fibrovascular networks presumably as a mechanism of scarification of the damaged retina in order to protect what remains (Figure 12) (Friedlander, 2007; Van Hove et al., 2020). Unfortunately, these gliotic scars, although primarily “protective,” negatively impact the architecture of the retina, prevent potentially regenerative mechanisms, limit vision, and ultimately become points of tension that cause retinal detachment and blindness (Yang et al., 2008). Interestingly, the microglial expression of connective tissue growth factor, which promotes extracellular matrix formation and fibrosis, is also altered during DR (Kuiper et al., 2004). Cross-activation between microglia and macroglia, and perpetuation of the immune response demonstrated using co-culture systems (Liu and Steinle, 2017; Wang et al., 2012, 2011; Yoshida et al., 2004) supports another collaborative function of Müller glial cells, astrocytes, and microglia in the regulation of the hallmark of PDR—the fibrotic process (Friedlander, 2007; Guidry et al., 2003). This process clearly involves both macroglia and microglia, but further work is needed to fully uncover their respective roles in what clearly is a fibroblastic response reminiscent of wound healing, and thus likely initially adaptive but progressively becoming maladaptive.

Figure 12. Summary of the clinical manifestations of diabetic retinopathy and the relative involvement of the vascular, immune, and neuronal systems.

Early on, activated micro- and macroglia (astrocytes and Muller glia) contribute to aneurysm formation through secretion of mediators that influence neighboring endothelial cells and weaken BRB integrity. Additionally, activated microglia secrete neurotoxins that impair the function of neurons in the neuroretina. As disease progress, activated microglia and macroglia contribute to the creation and formation of fibrovascular scars that push the disease stage to proliferative diabetic retinopathy. Activated microglia and astrocytes secrete pro-angiogenic and pro-inflammatory factors that result in the growth of abnormal vessels.

Clearly, the combinatorial effect of diabetes on Müller cells, astrocytes, and microglia outlines once more the communication and mutual activation of microglia and macroglia (Figure 8) (Friedlander, 2007; Guidry et al., 2003). An underlying theme throughout the various DR stages is the active immune response that contributes to the different observed pathologies, from neurodegeneration to vascular abnormalities. Therefore it is incredibly important to further investigate and decipher how the immune response is intricately tied to these processes.

7. Future directions and conclusions

Vision loss is a particularly devastating complication of diabetes. Treatment options remain dramatically limited, with the widely celebrated anti-VEGF therapies effective only in one-third to one-half of patients with vision-threatening stages of the disease (Bressler et al., 2014; Elman et al., 2010; Nguyen et al., 2010). Hence there is more than ever a clear need to develop new therapies that are more effective to preserve vision. This hinges on our understanding of the pathophysiology behind the disease. The innate immune system is receiving increasing attention as a major contributor to DR progression and may be at the heart of homeostatic dysregulation observed in all phases of disease; however, the clinical treatment of DR continues to revolve solely around its vascular consequences (Amoaku et al., 2020). In addition to revealing new potential targets for therapies such as the complement system (Fort et al., 2011; Huang et al., 2018; Muramatsu et al., 2013; Rasmussen et al., 2018), investigation into the immune response has revealed its incredible activity during preclinical DR—classically thought of as the “silent” period of disease. This provides an incredible opportunity to develop treatment for early stages of the disease and points out the need for a novel and more comprehensive classification of diabetic retinal diseases that includes vascular, neural, and inflammatory aspects of the disease (Abramoff et al., 2018).

In addition to the vascular phenomena that define disease stages, the field needs to focus much more of its attention on the other pathologies and dysfunctions observed throughout disease, with an emphasis on earlier stages of DR (Table 1). Among all these factors, the interactions of the various components of the innate immune system appear to be promising targets based on data from human patients and animal models of DR. Indeed, the successful regression of disease with the use of minocycline, topical PEDF derivatives, and adenoviral delivery of CD59 is testament to the importance and interlinked nature of these different components (Adhi et al., 2013; Krady et al., 2005; Yanling Liu et al., 2012; Zhang et al., 2006). In addition, the applications of pharmacotherapies such as resveratrol and laser treatments to decrease microgliosis and inflammatory mediators are promising avenues for intervention (Y. Chen et al., 2019; Midena et al., 2019). However, more work is needed in human patients to anchor and contextualize the findings made in animal models. Indeed, although rodent studies are incredibly important, none of the many DR rodent models fully recapitulates human disease (Olivares et al., 2017). This is particularly important and relevant to the understanding of disease progression and explains the lack of translation of findings in these models to new treatments for human patients. Using human tissues and integrating data from multiple analytical levels have proven to be successful for other diseases or diabetic complications (Afshinnia et al., 2019). Only by combining discoveries in multiple model systems and vertically integrating findings acquired in cells and animals can we begin to understand and untangle the complexities of pathogenesis. Maybe to our advantage, the inextricably linked nature of the different components of the innate immune system will provide multiple therapeutic strategies targeting overlapping or pathway-specific elements.

Abbreviations

BDNF

brain-derived neurotrophic factor

BRB

blood-retinal barrier

CAM

cell adhesion molecule

CX3CL1

fractalkine

CX3CR1

fractalkine receptor

db/db

leptin receptor deficient

DC

dendritic cell

DME

diabetic macular edema

DR

diabetic retinopathy

IL

interleukin

MCP-1

monocyte chemoattractant protein-1

NPDR

non-proliferative diabetic retinopathy

NVU

neurovascular unit

OIR

oxygen-induced retinopathy

PDR

proliferative diabetic retinopathy

PEDF

pigment epithelium–derived factor

RPE

retinal pigment epithelium

TNFα

tumor necrosis factor α

VEGF

vascular endothelial growth factor