Retinal endothelial cell phenotypic modifications during experimental autoimmune uveitis: a transcriptomic approach

In physiological conditions, the eye is isolated from systemic circulation by an anatomical barrier that restricts the entry of potentially toxic molecules, pathogens and immune cells into the eye. The blood-retinal barrier (BRB) is composed of 2 distinct parts. On the one side, the outer BRB is formed by tight junctions between retinal pigment epithelial (RPE) cells, separating fenestrated choroidal vessels from the outer retina. On the other side, the inner BRB is formed by tightly sealed vascular non-fenestrated endothelial cells (ECs), surrounded by pericytes and supported by Müller cell and astrocyte foot processes. Similarly to the blood-brain barrier (BBB) in the central nervous system, many of the properties of the BRB are manifested in ECs [1].

Alteration of BRB properties is a hallmark of ocular inflammatory diseases. Although BRB breakdown may be beneficial, allowing immune cells to clear debris and repair damages, it can also be harmful, leading to edema by accumulation of water and plasma proteins, tissue damage by inflammatory cells and ultimately vision loss. During non-infectious posterior uveitis, activated T cells produce cytokines and chemokines that activate retinal vessels [2], resulting in a drastic change of retinal EC phenotype, comprising among others upregulation of adhesion molecules such as P selectin, ICAM-1 [3] and VCAM-1 [4] and downregulation of tight junction proteins [5]. This leads to the recruitment of a wide range of circulating leukocytes such as monocytes/macrophages, granulocytes, NK cells, NKT cells and γδ T cells that are directly responsible for tissue damage [2]. The BRB then becomes increasingly permeable, allowing more immune cells into the eye, creating an amplification loop of the pathogenic process.

Although some individual alterations contributing to BRB dysfunction have been described, the underlying molecular mechanisms are still largely unknown. A transcriptomic study of total retinal cells and retinal ECs during experimental autoimmune uveitis (EAU) would thus represent an interesting approach to bring to light new target genes and signaling pathways involved in BRB breakdown. In the past decade, a few genome wide expression data sets have been generated for retinal ECs [6‐18] (Table 1). Gene profiling studies have already confirmed that ECs from the immune-privileged central nervous system have a particular gene signature, compared to ECs in other tissues [1]. A few transcriptomic studies have been performed in human retinal ECs [12, 15, 17, 19]. Compared to uveal ECs, retinal ECs were found to express higher levels of transcripts involved in the immune response, including cell adhesion molecules, cytokines, chemokines, receptors and enzymes participating in the synthesis of inflammatory proteins. This is consistent with their suspected role in the regulation of leukocyte trafficking and inflammatory reaction during uveitis [6, 12]. Furthermore, the expression of candidate genes such as adhesion molecules (e.g. ICAM-1, VCAM-1, E and P selectins) and chemokines (CXCL10, CCL20 and CX3CR1) was shown to be induced in cultured human ECs by inflammatory stimuli [19, 20]. However, as illustrated in Table 1, most studies were performed on cultured ECs, which rapidly lose their barrier properties [21]. Furthermore, although the effects of inflammatory stimuli by exposure to toxoplasma gondii, LPS or TNF-α [7, 15, 17, 18] on gene expression were explored in EC cultures, to the best of our knowledge no study has investigated yet how EC gene expression is affected in vivo during non-infectious uveitis.

In this work, we have decided to analyze the transcriptome of freshly isolated total retinal cells and retinal ECs from EAU and naïve retinas used as control. Using RNA-Seq, we identified a series of genes that are modulated during EAU development. The expression of some of those genes was then analyzed by immunofluorescence (IF) and flow cytometry.

In an attempt to investigate the local mechanisms of non-infectious posterior uveitis development at a global level, we aimed to study the gene regulation of total retinal cells and more specifically retinal ECs during EAU.

We first set up a method to purify and gene profile BRB-forming retinal ECs. We then compared the transcriptional profile of total retinal cells during AT EAU to that of naïve mice and finally performed the same analysis for retinal ECs.

The BRB allows regulated transport of nutrients to the retina while preserving visual function by restricting access to toxins, pathogens and immune cells. During non-infectious uveitis and its animal model EAU, leukocytes infiltrate the retina through a disrupted BRB and induce retinal damage. As the main component of the inner BRB, retinal ECs are thus a central player in the development of retinal inflammation. However, the molecular basis of BRB breakdown is only partially understood.

In this study, we first investigated the activation of total retinal cells during EAU, through comparative analysis of their transcriptomic profile in EAU versus naïve retinas. The analysis of transcriptome data reveals significant regulation (mostly up-regulation) of numerous genes implicated in Ag presentation in association to MHC Class II and T cell activation. This mainly reflects local engagement of the adaptive immune response, whose function in EAU development is already well-known [29]. Some of those genes have already been studied as potential therapeutic targets during EAU, e.g. Icos [30], IL17R [30] or CTLA4 [31] but numerous genes have not yet been explored in the field of uveitis. For example, the tyrosine kinase Jak3 is activated by interleukin (IL) receptors IL2R and IL17R (that comprise a common chain) and plays a capital role for immune response initiation.

However, one pitfall of this approach is that cells isolated from EAU retinas comprise both resident retinal cells and infiltrating inflammatory cells. Furthermore, very few endothelial-specific genes are brought to light by this global analysis. We thus decided to investigate specifically the activation of retinal ECs during EAU, through comparative analysis of their transcriptomic profile in EAU versus naïve retinas.

Endothelial cell isolation is challenging. Many techniques such as magnetic purification (Magnetic-activated cell sorting, MACS), freeze-fracture method or scraping of the endothelium do not yield pure endothelial populations [28]. Furthermore, those last 2 methods are not applicable to retinal microvessels due to their small size. Contamination by other cell types may be eliminated by further selection, for example by culturing ECs under adapted cell culture conditions. However, although in vitro models represent a useful tool for functional analysis of candidate proteins and high-throughput testing, cultured ECs rapidly lose their barrier properties, such as the high number of tight junction proteins and low number of transcytosis vesicles [21]. These properties can be partially restored by adding astrocytes and pericytes to EC cultures [32], but no in vitro model ideally reconstitutes all properties observed in vivo, since hemodynamic conditions and effects of circulating cells and proteins are lacking.

Although Tie2-GFP transgenic mice have proved to be very useful for effective EC isolation [33], we experienced highly restrictive low transmission rates when trying to derive the Tie2-GFP transgene into a C57BL/6 J background, that led us to explore an alternative EC isolation strategy based only on the expression of cell-surface EC markers. Isolation of ECs often relies on CD31 expression [34]. However, our IF images provide evidence for the expression of CD31 by immune cells infiltrating the vitreous (Fig. 2a). In agreement, CD31 expression was reported in diverse immune cells [35]. Although endoglin expression has also been described in myeloid precursors and macrophages [36], in our staining experiments, endoglin expression in EAU retinal cryosections remained strictly confined to retinal vessels. We thus decided to purify ECs by flow cytometry sorting of CD31 + endoglin+ double positive CD45- cells.

In our transcriptome analysis, we detected the presence of pericytes in naïve EC samples, whether sorted from Tie2-GFP or WT mice. The intimate relationship between pericytes and ECs was initially described in the 1970s, pericytes being enclosed in the endothelial basement membrane and involved in adhesive junctions with ECs, rendering individualization of each cell type particularly tricky [37]. Pericytes are exceptionally abundant at the level of barrier-type ECs, with a pericyte/EC ratio of 1:1 in the retina and 1:3 in the brain, compared to 1:10 in other microvascular beds [21]. Although the majority of the pericyte-endothelial interface is separated by a basement membrane, at some places the two cell types form focal contacts through N-cadherin and connexins, allowing them to exchange metabolites and even ribonucleic acids [38]. Such pericyte contamination of EC samples was also observed by Daneman et al. in brain ECs sorted from Tie2-GFP transgenic mice [1]. In that study, a double FACS procedure with exclusion of cells positive for the pericyte marker PDGFRß allowed isolation of pure ECs. Unfortunately, the number of ECs that can be isolated from the retina is much lower than that obtainable from the brain, and in our experimental conditions a double FACS procedure would cause excessive cell loss. In an attempt to exclude pericytes, we tested an anti-PDGFRß antibody but observed only a very weak signal, hardly distinguishable from the FMO control (Additional file 9: Fig. S9). However, since it was shown that pericytes actively take part in establishing the barrier properties of the inner BRB [21], including pericytes in the analysis of gene regulation involved in BRB breakdown bears a lot of interest. Interestingly, our RNA-Seq data indicate lower pericyte contamination in EAU ECs in comparison with naïve ECs. This is consistent with the fact that BRB breakdown has been associated with pericyte dysfunction and loss. Pericyte loss in the brain was reported to be associated with upregulation of EC transcytosis and induction of several permeability-related factors [39]. The role of pericytes is less clear in the retina, but pericyte loss occurs early in diabetic retinopathy [21]. Our data thus suggest that pericyte loss is also implicated in uveitis-related BRB breakdown.

We also observed contamination of EC samples by photoreceptor genes. The retina is mainly composed of rods (80%) and many research teams have shown retinal cell transcriptome contamination by photoreceptor genes, even with extremely draconian sorting methods relying on transgenic mouse lines selectively expressing fluorescent proteins in different retinal cell types [29, 40, 41]. In this context, interestingly, McKenzie et al. studied the modification of retinal EC gene expression in different mutant models of non-neovascular remodeling by microarray, and also picked up photoreceptor gene expression in retinal vessel fractions [42]. Contrary to pericyte contamination, photoreceptor genes were more strongly expressed by diseased ECs compared to naïve. This could reflect the fact that diseased cells are more adhesive than naïve [43].

To bypass this contamination, we explored 2 approaches: the analysis of the 120 most variant genes across DE (diseased endothelium) and NE (naïve endothelium) samples and a bioinformatics analysis to select genes based on their expression profile. We then systematically eliminated remaining photoreceptor genes. By combining these 2 approaches, we identified 82 genes significantly modulated in DE compared to NE. Among those genes, a few are already known to be implicated in uveitis, such as E and P selectins [43], CD44 [44], IL-33 [45, 46] and Lcn2 [47]. Other genes were already implicated in other inflammatory disease models but not in uveitis, such as Lrg1 [48], Ackr1 [49] and Timp1 [50]. Finally, some of those genes are known to be involved in retinal function but not during uveitis specifically, such as Clu [51], Lrg1 [8] and Fgf2 [52].

Functional analysis of the RNA-Seq data with the DAVID web-based tool allowed to identify different pathways enriched in ECs during EAU. Among those, some enriched GO terms were quite expected, such as those related to the inflammatory response or to cell adhesion processes. Among less expected enriched pathways in diseased retinal ECs, we found upregulation of molecules related to complement activation, extracellular matrix and angiogenesis.

A few studies have already reported involvement of the complement cascade in uveitis development [53‐55], as well as in EAE pathogenesis [56].

Disruption of basement membrane and extracellular matrix components is required for immune cell migration towards inflamed sites. Proteolysis by matrix metalloproteinases (MMPs) is also involved in the regulation of EC barrier function, as well in other processes such as vascular growth and interaction with circulating immune cells [57]. Elevated levels of MMPs were found in the aqueous humor of uveitis patients, in correlation with the inflammatory activity [58, 59], and it was shown that specific inhibition of MMP-2 and -9 ameliorates EAU [60].

Angiogenesis is not typically associated with peak EAU. Interestingly, however, the major angiogenic factor VEGF was shown to be increased in the retina during EAU without neovascularization and involved in induction of vascular permeability, highlighting a less described implication in other processes than angiogenesis [61]. The association of sustained inflammation with angiogenesis is now well established [62] and an interplay is described in the pathogenesis of major retinal diseases. These connections between inflammation and angiogenesis are reflected by the efficacy of steroids in diabetic retinopathy [63] and of anti-VEGF in some uveitis patients [64].

Interestingly, to date, in the literature, the results of gene and protein profiling do not correlate very well [56]. Therefore, we used IF imaging on both retinal cryosections and retinal wholemounts as well as flow cytometry to validate the expression of some attractive regulated genes we identified by RNA-Seq.

Our RNA-Seq data point out an expression of serpina3n in both retinal ECs and total retina during EAU. IF images confirm that expression is strongly induced both at the vascular level and on glial cells during EAU. Serpina3n is a secreted serine protease inhibitor associated to the inflammatory response, whose expression was demonstrated in the ischemic brain and in the liver and pancreas in response to inflammatory stimuli [65]. In agreement with our data, Takamiya et al. showed that serpina3n mRNA is induced at the perivascular level, in retinal astrocytes and uveal epithelial cells during the early phase of endotoxin-induced uveitis in rats [65]. Interestingly, serpina3n was shown to have neuroprotective effects in the EAE model, through inhibition of granzyme B [66]. Genes of the serpin family were recently found to have an altered expression in different neurological disease models, with specific involvement of serpina3n in EAE [67]. Our data suggest the potential interest of exploring serpina3n role in uveitis development.

According to our RNA-Seq data, lcn2 expression is induced in both retinal ECs and total retina during EAU. Our IF images clearly show upregulation of lcn2 expression by perivascular macroglial cells during EAU. However, although retinal wholemounts show vascular lcn2 expression, on retinal cryosections it seems that lcn2 expression is not truly attributable to ECs. Lcn2 was shown to be the most upregulated early stress response gene in the eye exposed to light-induced injury [68]. In agreement with our IF data, in mouse models of retinal degeneration, Müller cells were observed to respond to photoreceptor damage by expression of lcn2 [69]. Furthermore, 2 other studies in the EAE model found astrocytes and monocytes/microglia to be the major cell types expressing lcn2 [70, 71]. ECs engage in close contacts with astrocytes and Müller cells, which participate in the formation and maintenance of the BRB. These tight relationships might explain the dragging of some astrocytes and Müller cells in the EC samples during cell sorting. In the field of uveitis, lcn2 was reported to be highly induced in the equine recurrent uveitis model [72] . As concerns humans, lcn2 serum levels are increased in patients with Behçet’s disease and psoriasis compared to controls [73] . Furthermore, lcn2 concentration in aqueous humor is increased in eyes with idiopathic acute anterior uveitis [47]. At the functional level, Nam et al. found a pathogenic role for lcn2 in EAE development [70], while Berard et al. observed increased EAE severity in lcn2 knock-out (KO) mice [71]. Taken together, these data suggest an important yet to precise role of lcn2 in retinal response to stress.

Our RNA-Seq and IF data both point to ackr1 upregulation in diseased ECs compared to naïve. Ackr1 is an atypical chemokine receptor, that binds different pro-inflammatory chemokines and regulates their activity by transcytosis across the endothelium, from the basolateral to the luminal side, where those chemokines contribute to leukocyte diapedesis [74]. Endothelial expression of ackr1 was shown in other experimental models of inflammatory diseases, such as atherosclerosis [75]. Upregulation of ackr1 was shown at the level of the BBB in EAE and MS.

Our RNA-Seq and IF data agree in showing that lrg1 is upregulated in diseased endothelium compared to naïve. Lrg1 belongs to the ‘Leucine-rich Repeat’ family, whose members are known to be involved in protein-protein interactions, signal transduction, cell adhesion and development [76]. Lrg1 expression in the retina was first described by Wang et al. in 2013 [8] . In this paper, lrg1 retinal expression is shown to be restricted almost exclusively to the vasculature and strongly upregulated in different mouse models of choroidal and retinal neovascularization, where it promotes angiogenesis [8] . Besides, lrg1 expression is upregulated during acute inflammation in mice, and it was proposed that it might even serve as a diagnostic inflammatory biomarker [77] . Lrg1 expression was shown to be associated with disease activity in different human diseases such as rheumatoid arthritis [78], ulcerative colitis [79] and Crohn’s disease [48] . Our data bring to light the potential role of lrg1 in the development of retinal inflammation.

Protein expression analysis for those 4 genes thus globally confirms vascular upregulation during EAU. However, serpina3n and lcn2 are also detected at the perivascular level, mainly on glial cells. Such perivascular expression probably exists for other genes in our list. However, as previously mentioned for pericytes, macroglial cells being part of the inner BRB, study of gene regulation at their level is also clearly relevant in the context of our work.

Unlike the previous genes for which protein validation was performed, RNA-Seq data point to downregulation of lamc3 expression in DE compared to NE. Although downregulation is not clearly observed by IF, flow cytometry data show a tendency of ECs to downregulate lamc3 expression during EAU. Lamc3 is a poorly described member of the laminin family, which are a major component of the vascular basement membrane. Lamc3 was shown to play a role in vascular branching and EC proliferation during angiogenesis, though interaction with microglial cells [80].

Our RNA-Seq data confirm the absence of MHC Class II expression in ECs, even during EAU. The only candidate gene possibly involved in Ag presentation is Cathepsin S (Ctss). However, Ctss was also reported to be implicated in EC dysfunction and in particular in microvascular complications of diabetes [81].

In our data, most genes were upregulated in DE compared to NE. Functional analysis of our list of candidate genes seems to indicate that some ‘permeability’ genes are upregulated in retinal ECs during BRB breakdown. Munji et al. performed vast transcriptome studies on isolated BBB ECs in experimental models of different BBB breakdown-associated diseases: stroke, MS, brain injury and epilepsy. With this approach, they identified a set of genes whose altered expression is shared between models, that was named the BBB dysfunction module [67]. Interestingly, this dysfunction module contained mainly peripheral endothelial genes, suggesting that BBB breakdown rather implies upregulation of ‘permeability’ genes than downregulation of ‘barrier’ genes. No study has investigated yet whether this might also be the case in BRB breakdown.

In conclusion, our work provides the first transcriptomic study on total retinal cells and retinal ECs during EAU. Our data not only corroborate the involvement of molecules known for a pathogenic role in retinal inflammation, but further provide a list of new molecules regulated in inner BRB cells during uveitis. These molecules represent potential novel therapeutic targets for the development of biologic drugs to treat human inflammatory eye disorders.