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

Background Blood-retinal barrier cells are known to exhibit a massive phenotypic change during experimental autoimmune uveitis (EAU) development. In an attempt to investigate the mechanisms of blood-retinal barrier (BRB) breakdown at a global level, we studied the gene regulation of total retinal cells and retinal endothelial cells during non-infectious uveitis. Methods Retinal endothelial cells were isolated by flow cytometry either in Tie2-GFP mice (CD31+ CD45− GFP+ cells), or in wild type C57BL/6 mice (CD31+ CD45− endoglin+ cells). EAU was induced in C57BL/6 mice by adoptive transfer of IRBP1–20-specific T cells. Total retinal cells and retinal endothelial cells from naïve and EAU mice were sorted and their gene expression compared by RNA-Seq. Protein expression of selected genes was validated by immunofluorescence on retinal wholemounts and cryosections and by flow cytometry. Results Retinal endothelial cell sorting in wild type C57BL/6 mice was validated by comparative transcriptome analysis with retinal endothelial cells sorted from Tie2-GFP mice, which express GFP under the control of the endothelial-specific receptor tyrosine kinase promoter Tie2. RNA-Seq analysis of total retinal cells mainly brought to light upregulation of genes involved in antigen presentation and T cell activation during EAU. Specific transcriptome analysis of retinal endothelial cells allowed us to identify 82 genes modulated in retinal endothelial cells during EAU development. Protein expression of 5 of those genes (serpina3n, lcn2, ackr1, lrg1 and lamc3) was validated at the level of inner BRB cells. Conclusion Those data not only confirm the involvement of known pathogenic molecules but further provide a list of new candidate genes and pathways possibly implicated in inner BRB breakdown during non-infectious posterior uveitis.


Background
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 noninfectious 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][7][8][9][10][11][12][13][14][15][16][17][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.

Backcrossing and genotyping
Two FVB/N Tie2-GFP males (Tg(TIE2GFP)287Sato/J, stock number 003658) were paired with wild type (WT) C57BL/6 J female mice. C57BL/6 J-Tie2-GFP mice were then generated by backcrossing genetically selected Tie2-GFP carrier males with WT C57BL/6 J females over 10 generations. The resulting litters from each pairing were genotyped by standard PCR amplification of the transgene on genomic DNA extracted from tail tissue samples, as suggested by Jackson Laboratories.
Adoptive transfer model of experimental autoimmune uveitis EAU was induced by adoptive transfer (AT) of autoreactive lymphocytes following the protocol of Shao H et al. [22]. Briefly, naive C57BL/6 J mice were immunized with a subcutaneous injection in each hind leg of 50 μl of a mixture containing 500 μg/100 μl IRBP peptide 1-20 in a 1:1 emulsion of CFA enriched with 2.5 mg/ml of heat-inactivated Mycobacterium tuberculosis. All animals received simultaneously an intraperitoneal injection of 1 μg of PTX. Twelve days after immunization, mice were euthanized by cervical dislocation and their spleen and draining lymph nodes dissected and dissociated. Spleen cell suspensions were enriched in T lymphocytes through passage on nylon wool fiber columns, then pooled with total lymph node cells and re-stimulated in vitro with IRBP1-20 (1 μg/ml). After 2 days in culture, cells were injected intraperitoneally into naive C57BL/6 J mice (5 × 10 6 cells/mouse).

Disease grading
A clinical grading was performed 21 days after AT. Mice were anaesthetized by a 50 μl intraperitoneal injection of a Rompun (0,2%) and Ketalar (20 mg/ml) mixture. Pupils were dilated with tropicamid (5 mg/ml) and phenylephrine (1.5 mg/ml) and eyes were examined under the slit-lamp of a surgical microscope (Zeiss, Göttingen, Germany) by using a cover slip coated in a viscoelastic gel (synthetic polymer of acrylic acid 2 mg/g, Vidisic, Tramedico, Belgium) and positioned on the cornea. The clinical grading was performed independently by two ophthalmologists, based on a system adapted from Xu et al. [23]. Briefly, vitritis, optic neuropathy, retinitis and vasculitis were separately scored in each eye, from 0 (no disease) to 4 (highly severe disease) with half point increments and averaged to generate the clinical score of the eye on a scale from 1 to 4. The clinical score attributed to one mouse corresponds to the mean of the scores of the 2 eyes. The animals were euthanized by cervical dislocation.

Preparation of retinal single cell suspensions
At day 21 after disease induction, mice were euthanized by cervical dislocation while still anaesthetized and unconscious (see section 4.4), and the eyes immediately enucleated. Naïve eyes were used as control. Eyes were carefully hemisected in Hank's Balanced Salt Solution (HBSS) buffer containing penicillin/streptomycin 1%, with surgical scissors under a surgical microscope. Retinal tissue was isolated and rinsed in HBSS buffer. The 2 retinas of each mouse were pooled and cut into small pieces before enzymatic digestion in 3 ml HBSS containing 1,6 mg/ml Liberase DL (Roche, Vilvoorde, Belgium) and 0,1 mg/ml DNase I (Sigma-Aldrich) at 37°C for 45 min. Cell dissociation was stimulated by pipetting every 15 min. Cells were washed with Dulbecco modified eagle medium (DMEM)/10% fetal bovine serum (FBS) and filtered through a 40-μm cell strainer to obtain a single cell suspension [24]. The yield was approximately 2 to 3 million retinal cells per mouse.

Fluorescence-activated cell sorting (FACS) analysis
Retinal single cell suspensions were tested by flow cytometry for the expression of GFP, endoglin (CD105, endothelial cell marker), CD31 (PECAM-1, endothelial cell marker), CD45 (pan-leukocyte marker), ackr1, lrg1 and lamc3 using specific antibodies (BD Biosciences) coupled to different fluorochromes, except for GFP whose native fluorescence was observed. Cells were incubated with the relevant antibodies for 20 min at 4°C, washed and re-suspended in FACS buffer. Live cells were gated with Hoechst 1/4000 and debris and doublets were excluded. Up to one million total cells per sample were analyzed on an LSR-Fortessa flow cytometer using the CellQuest Software (BD Biosciences). Isotypes and Fluorescence minus one (FMO) controls were used for accurate gating. Compensations were performed using BD CompBeads (BD Biosciences).

Analysis of retinal cell gene expression Purification of total retinal cells and retinal endothelial cells
Three weeks after AT, mice were sacrificed, and retinal single cell suspensions prepared as described above. Naïve eyes were used as controls. Cells were stained with APC-labeled anti-CD31, PECy7-labeled anti-CD45 and when applicable PE-labeled anti-endoglin antibodies. Total live retinal cells (DAPI-) and then retinal ECs (either GFP + CD31 + CD45or endoglin+CD31 + CD45-cells) were separately sorted by preparative flow cytometry using a FACSAria with the FACSDiva Software (BD). Due to the low cell number obtained from each mouse, at least 3 mice were pooled to generate each EAU sample and at least 2 mice for naïve samples. Cells were sorted directly in lysis buffer, vortexed for 30 s and flash frozen in liquid nitrogen.

RNA extraction
RNA extraction was performed using the MiRNeasy MicroKit (Qiagen) according to the manufacturer's recommendations and a DNase step to avoid DNA contamination. RNA quality was assessed using the Agilent 2100 Bioanalyzer with RNA 6000 Pico kit (Agilent Technologies).

RNA processing and RNA sequencing
Indexed cDNA libraries were prepared using the Ovation Single Cell RNA-Seq system (Nugen) (for samples isolated from transgenic mice) or the Ovation SoLo RNAseq system (Nugen) (for samples isolated from WT mice, since the Ovation Single Cell RNA-Seq system was then no longer available from Nugen). The multiplexed libraries were loaded, and sequences were produced using a TruSeq PE cluster and SBS-kit on a HiSeq 1500 (Illumina). Approximately 25 million paired-end reads/ sample were mapped against the mouse reference genome (NCBI Build 37/UCSC mm9) using STAR software to generate read alignments for each sample. Expression levels were quantified using the featureCounts [25] tool and the UCSC RefSeq gene annotation as a reference (exons only, genes as meta features). Differential analysis between groups was performed using the EdgeR package (quasi-likelihood F-tests). Normalized expression levels were estimated using the EdgeR rpm function and converted to log 2 FPKM (fragments per kilobase of exon per million mapped reads) after resetting low FPKMs to 1.
To perform blind clustering analysis, genes were selected based on the overall variance between samples (independently of their category), by keeping only the 120 most variant ones. Functional analysis was performed using the Database for Annotation, Visualization, and Integrated Discovery (DAVID) web-based functional annotation tool [26].

Results
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.

Purification and gene profiling of naïve retinal endothelial cells
The first challenge to be addressed pertained to the isolation of retinal ECs, which represent a small percentage of cells to extract from an already very small tissue.
As a first approach to retinal EC sorting, we decided to use transgenic Tie2-GFP mice, which express GFP under the control of the pan-endothelial Tie2 promoter (transgenic strategy). However, FVB/N-Tie2-GFP mice are not susceptible to EAU induction, as shown in Additional file 1: Fig. S1. We thus generated C57BL/6 J-Tie2-GFP mice by crossing founder FVB/N-Tie2-GFP males with WT C57BL/6 J mice and subsequently backcrossing genetically selected Tie2-GFP carrier males with WT C57BL/6 J females over 10 generations. In order to validate the specificity of GFP expression in retinal ECs, we first analyzed comparatively the expression of GFP with that of CD31 and endoglin (2 EC markers) in the retina of naïve heterozygous Tie2-GFP mice by immunofluorescence (IF) on eye cryosections. Figure 1 shows that the expression of GFP matches that of CD31 and endoglin in the retina of naive heterozygous Tie2-GFP mice. Unfortunately, we unexpectedly observed an extremely low transmission rate of the Tie2-GFP allele with successive backcrossing generations (Additional file 2: Fig. S2). Indeed, while 100% of F1 mice were carriers, the proportion dropped to 0% for F10 mice.
We thus developed an alternative EC sorting strategy in C57BL/6 J WT mice, based on combined expression of the 2 retinal EC markers, CD31 and endoglin, that were previously tested by IF in Fig. 1 (WT strategy). We first analyzed CD31 and endoglin expression on AT EAU eye cryosections from WT C57BL/6 J mice by IF (Fig. 2). During EAU, retinal vessels are well identified by CD31 and endoglin stainings (Fig. 2, thin arrows).  However, some cells infiltrating the retina and vitreous show CD31 expression, which is clearly not endothelial (Fig. 2, thick arrows), while endoglin expression seems to remain purely endothelial during EAU. This indicates a lack of specificity of CD31 expression to firmly identify retinal ECs, enhancing the necessity to combine different markers for accurate retinal EC sorting.
Healthy murine retinas were then processed into single cell suspensions by enzymatic digestion. ECs were further sorted by FACS relying on either the expression of GFP and CD31 (transgenic strategy) or the expression of CD31 and endoglin (WT strategy), both combined with the absence of CD45 expression. We have indeed detected, as previously reported in the literature, a few cells coexpressing GFP, CD31 and CD45 which may enlighten the presence of Tie2-expressing hematopoietic cells (data not shown) [27,28]. Each sample was generated from a pool of at least 4 retinas from 2 naïve mice. As shown in Additional file 3: Fig. S3A, ECs represent about 0,1% of the retinal cell population (excluding dead cells, debris and doublets) in both strategies. The WT sorting strategy was tested on Tie2-GFP mice, as illustrated in Additional file 3: Fig. S3B. Among pre-gated CD31 + CD45-cells, the majority of cells express both endoglin and GFP. Among cells gated as endothelial with the WT strategy (endo-glin+CD31 + CD45-, blue gate), the vast majority would also be selected as ECs with the transgenic sorting strategy (GFP + CD31 + CD45-, orange gate).
We also compared the purity of the EC samples sorted with the 2 sorting strategies by RNA-Seq. For transcriptome studies, retinal ECs and total live retinal cells were separately sorted, both in C57BL/6 J-Tie2-GFP mice and in C57BL/6 J WT mice. Given the low expected number of cells, each sample was sorted from a pool of 3 mice. RNA was extracted and its quality tested before being processed for RNA-Seq analysis. RNA-Seq data were first analyzed for the expression of endothelial (Tie2, Cldn5, CD31), pericyte (Pdgfrb, Abcc9, Lama2), photoreceptor (Rho, Pde6b, SAg), microglia/macrophage (Ptprc (CD45), itgam (CD11b), Cx3cr1), macroglia (Aqp4, Aldh1l1, Slc1a3) and RPE (RPE65, Rgr, Rdh10) cell markers. Figure 3a illustrates comparative expression of those retinal cell markers by ECs versus total retina (green bars represent samples obtained with the transgenic strategy; blue bars represent samples obtained with the WT strategy). EC markers are highly enriched, and all non-endothelial markers are depleted in the EC fractions, whatever the sorting strategy, with the exception of pericyte markers. We next investigated the expression of more numerous EC markers, previously identified as specifically enriched in BBB ECs [1]. As illustrated in Fig. 3b, all those barriertype EC markers are significantly enriched in the EC fractions, compared to total retina, with a similar pattern between the 2 sorting strategies.

Analysis of retinal cell gene regulation during EAU
We then induced AT EAU in C57BL/6 J WT mice and sorted live total retinal cells and retinal ECs in order to identify regulated transcripts compared to naïve mice. Each sample was sorted from a pool of at least 3 mice. We analyzed a total of 5 AT EAU samples and 4 naïve samples.
The transcriptional profiles of replicates (samples of the same nature isolated from the same pool of mice), biological duplicates (samples of the same nature isolated from different pools of mice) and samples of different nature were compared, in order to identify the source of variation. As shown in Additional file 4: Fig. S4, the dispersion is lowest for replicates, intermediate for biological duplicates and highest for samples of different nature.
A principal component analysis (PCA) was performed with all diseased and naïve retinal cell samples (Fig. 4a). PCA provides an unsupervised 2-dimensional graphic representation of the similarities and differences between samples. The PCA data show a tendency of diseased retinal cell samples (DR) to segregate from naïve retinal cell samples (NR). However, the transcriptome of 2 diseased retinal cell samples (DR4 and DR5) are closer to the cluster of naïve retinal cell samples. Those 2 samples are isolated from the only 2 pools of mice whose mean clinical grade was inferior to 2. When those 2 samples are eliminated, clearer segregation between diseased and naïve retinal cells is brought to light (Fig. 4b). Since the aim of the project was to target the differences between EAU and naïve cells, those 2 samples were dismissed for further analysis, which was thus performed on 3 diseased samples and 4 naïve samples.
Only genes with a Log fold change (FC) superior to 2 (absolute value) and a false detection rate (FDR) inferior to 0,05 were retained. With those filters, we identified 182 genes significantly regulated between diseased and naïve retinal cells, which were ranked according to their foldchange top down (Additional file 5: Fig. S5).
We then tried to classify those genes into families ( Table 2).
In order to identify the different pathways in which those genes are involved, we used DAVID, Pubmed and Genecards. Four main groups were identified: 52 genes playing a role in antigen presentation to T cells in association with MHC class II, 45 genes implicated in cytokine and chemokine signaling pathways, 39 genes taking part in cell mobility and 18 genes that contribute to innate immunity. However, 33 genes could not be attributed to one of these categories. Genes whose function was previously described during uveitis appear in blue, genes whose function was not previously described in uveitis appear in orange. Figure 5 graphically illustrates the role of the genes classified in group 1 (antigen presentation).
RNA Seq analysis of total retinal cells mainly brought to light upregulation of genes involved in antigen presentation and T cell activation during EAU. This approach thus reflects activation of local antigen-presenting cells and invasion of the retina by immunocompetent cells but does not allow specific detection of BRB cell gene modulation. This comforted us to adopt a more specific strategy targeting ECs.

Identification of retinal endothelial cell gene regulation during EAU
Retinal ECs were isolated from the same EAU and naïve samples as total retinal cell samples, with the WT sorting strategy described above.
A principal component analysis (PCA) was performed with EAU endothelial samples (DE, diseased endothelium) and naïve endothelial samples (NE, naïve endothelium) (Fig. 6). This PCA shows that DE and NE tend to cluster separately.
As illustrated in Additional file 6: Fig. S6, mRNA levels are accurately enriched or depleted for all markers used for cell sorting. As previously observed or mentioned in the literature, contamination of the transcriptome of EC samples by pericytes and photoreceptor genes was observed. Figure 7 indeed illustrates the expression of In order to select relevant genes, we first ranked by false detection rate (FDR) the list of significantly regulated genes (FDR < 0,05) between the 2 conditions. Unfortunately, as expected given the photoreceptor contamination, many of the highly regulated genes were photoreceptor genes, rendering the analysis tedious. In an attempt to bypass this issue, we then looked at the 120 most variant genes between DE and NE (Fig. 8). Among those 120 genes, 65 were eliminated because they did not reach statistical significance. In order to identify different profiles of gene expression, we selected a list of genes by using the bioinformatics formula 'DE significantly different from NE AND DE ≠ NR'. Among those, we selected 3 expression profiles in which the expression in naïve retina is low, hoping to minimize any photoreceptor gene contamination. Those 3 profiles are illustrated in Fig. 9. The first profile corresponds to genes expressed only in DE (DE+). The second profile corresponds to genes expressed in DE and in DR (DE+/ DR+). The third profile corresponds to genes expressed in NE only (NE+). Among significantly regulated genes, we identified 14 genes with the DE+ profile, 15 genes with the DE+/DR+ profile and 19 genes with the NE+ profile.
Research in the literature was conducted for the most variant genes as well as for genes selected for their profile, in order to exclude genes associated with photoreceptors. 21 genes known to be expressed by photoreceptors, among which 6 genes implicated in the photo-transduction cascade, were eliminated (Additional file 7: Fig. S7). Finally, 82 candidate genes were retained, among which 10 were selected both through their variance and through their expression profile (Additional file 8: Fig. S8).
Functional analysis with the David web-based tool was performed with the list of candidate genes. A selection of gene ontology (GO) terms enriched in DE compared to NE samples is illustrated in Table 3, and includes items related to inflammation, cell adhesion, complement activation, extracellular matrix and angiogenesis.

Analysis of the expression of candidate gene products
The expression of many of the identified candidate genes was already validated in the literature, especially those with the expression profile 'DE+'. Based on the availability of antibodies, we decided to analyze the expression of 5 of the most strongly regulated candidate genes at the protein level: serpin family A member 3 (serpina3n), lipocalin 2 (lcn2), atypical chemokine receptor 1 (ackr1), laminin γ3 (lamc3) and leucine rich alpha-2-glycoprotein 1 (lrg1). We looked at protein expression on both retinal cryosections and wholemount retinal preparations, as well as by flow cytometry when applicable.
According to RNA-Seq data, serpina3n is expressed by both DE and DR (Fig. 10). Protein expression is visualized in naïve eye cryosections at the level of the ciliary body and very discretely on some optic nerve and retinal vessels (Fig. 10b). In naïve retinal wholemounts, discontinuous serpina3n expression is observed at the vascular level (Fig. 10c). Stronger expression develops in inflamed retinas, particularly at the level of vasculitis (thick arrows), both in ECs and perivascular glial cells (Fig. 10d). However, since expression is found both at the vascular and perivascular levels, it is difficult to differentiate purely endothelial expression from perivascular expression, e.g. by infiltrating immune cells. In EAU retinal wholemounts, serpina3n expression is found on most retinal vessels, colocalized with CD31 and is particularly intense at the level of inflamed vessels (thick arrows) (Fig. 10e).
Ackr1 belongs to the same DE+/DR+ profile. Ackr1 expression is purely endothelial and appears only in Table 2 Classification of the 182 candidate genes into 4 groups Legend: 182 genes were found to be significantly regulated between diseased and naïve retinal cells. Those genes were classified into 4 main groups. Genes whose function was previously described during uveitis appear in blue, genes whose function was not previously described in uveitis appear in orange EAU cryosections (Fig. 12c) and wholemounts (Fig. 12d). This upregulation of ackr1 expression during EAU is confirmed by flow cytometry data ( Fig. 12e and f). On average, 2% of naïve retinal ECs express ackr1, while 30, 3% of retinal ECs express ackr1 during EAU. This upregulation is statistically significant (n = 3 independent pools of 3 mice, p-value for unpaired t-test: 0,0003).
According to RNA-Seq data, lrg1 is expressed only by DE (Fig. 13a). In naïve retinal sections, lrg1 is faintly expressed at the level of the ciliary body and at the ora serrata. No clear staining is observed in central retina (Fig. 13b). In naive retinal wholemounts, peripheral irregular expression of lrg1 appears at the vascular level, co-stained with CD31 (Fig. 13c). During uveitis, lrg1 upregulation is observed at the level of inflamed vessels on cryosections (Fig. 13d), while wholemounts show more diffuse lrg1 expression towards central retina than in naïve retina (Fig. 13e).
Lamc3 corresponds to the 3rd profile, with expression higher in NE than in DE (Fig. 14a). No clear difference is detected by IF on EAU versus naïve cryosections (data not shown). In retinal wholemounts, staining appears less organized and more dot-shaped in diseased retina (Fig. 14c) compared to naïve retina (Fig. 14b). However, flow cytometry data show a tendency of ECs to downregulate Lamc3 expression during EAU (Fig. 14d and e), which does not reach statistical significance (mean percentage of naïve retinal ECs expressing lamc3 45%, mean percentage of EAU retinal ECs expressing lamc3 30,7%, n = 3 independent pools of 3 mice, p-value for unpaired t-test 0,15).

Discussion
The BRB allows regulated transport of nutrients to the retina while preserving visual function by restricting access to toxins, pathogens and immune cells. During noninfectious 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 Fig. 5 Schematic representation of the interaction between an antigen-presenting cell and a T cell. In orange: genes whose expression or function was not previously described in the context of uveitis. In blue: genes whose expression or function was previously described in the context of uveitis. In black: genes that do not belong to the list of 182 candidate genes MHC Class II and T cell activation. This mainly reflects local engagement of the adaptive immune response, whose function in EAU development is already wellknown [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 cellsurface 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 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 barriertype 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 Fig. 9 Selection of candidate genes based on their expression profile. Genes were first filtered with the bioinformatics formula DE vs NE + DE ≠ NR. Two profiles of expression were then selected, for which the risk of contamination by photoreceptor genes is minimal. a. The first profile corresponds to genes expressed only in DE (DE+, e.g. Lrg1). b The second profile corresponds to genes expressed in DE and in DR (DE+/DR+, e.g. Lcn2). c The third profile corresponds to genes whose expression is down-regulated in DE versus NE (DE-, e.g. Lamc3) EC transcytosis and induction of several permeabilityrelated 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][54][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 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, serpi-na3n 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 Fig. 11 Analysis of lcn2 protein expression. Lcn2 expression was analyzed by immunofluorescence on retinal cryosections and retinal wholemounts 3 weeks after adoptive transfer. Naive eyes were used as control. Cryosections and wholemounts were prepared and stained for lcn2 (magenta) and CD31 (green) detection. Cell nuclei were stained with Hoechst (blue). a Expression profile of lcn2 at the mRNA level. b Analysis of lcn2 expression by immunofluorescence on naive retinal cryosections. c Analysis of lcn2 expression by immunofluorescence on naive retinal wholemounts. d Analysis of lcn2 expression by immunofluorescence on EAU retinal cryosections. e Analysis of lcn2 expression by immunofluorescence on EAU retinal wholemounts. f Analysis of lcn2 expression by immunofluorescence on EAU retinal wholemounts (× 40 objective). Arrows point to possible lcn2 and CD31 co-staining. Scale bars represent 20 μm 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 Fig. 12 Analysis of ackr1 protein expression. Ackr1 expression was analyzed by immunofluorescence on retinal cryosections and retinal wholemounts 3 weeks after adoptive transfer, as well as by flow cytometry. Naive eyes were used as control. Cryosections and wholemounts were prepared and stained for ackr1 (red or green) and CD31 (magenta) detection. Cell nuclei were stained with Hoechst (blue). a Analysis of ackr1 expression by immunofluorescence on naive retinal cryosections. b Analysis of ackr1 expression by immunofluorescence on naive retinal wholemounts. c Analysis of ackr1 expression by immunofluorescence on EAU retinal cryosections. d Analysis of ackr1 expression by immunofluorescence on EAU retinal wholemounts. e FACS analysis of ackr1 expression by naïve retinal endothelial (CD31+ Endoglin+ CD45-) cells. f FACS analysis of ackr1 expression by EAU retinal endothelial (CD31+ Endoglin+ CD45-) cells. Scale bars represent 20 μm 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 proinflammatory 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 Fig. 13 Analysis of lrg1 protein expression. Lrg1 expression was analyzed by immunofluorescence on retinal cryosections and retinal wholemounts 3 weeks after adoptive transfer. Naive eyes were used as control. Cryosections and wholemounts were prepared and stained for lrg1 (green) and endoglin (magenta) detection. Cell nuclei were stained with Hoechst (blue). a Expression profile of lrg1 at the mRNA level. b Analysis of lrg1 expression by immunofluorescence on naive retinal cryosections. c Analysis of lrg1 expression by immunofluorescence on naive retinal wholemounts. d Analysis of lrg1 expression by immunofluorescence on EAU retinal cryosections. e Analysis of lrg1 expression by immunofluorescence on EAU retinal wholemounts. Arrows point to lrg1 and endoglin co-staining. Scale bars represent 20 μm 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, ser-pina3n 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 Fig. 14 Analysis of Lamc3 protein expression. Lamc3 expression was analyzed by immunofluorescence on retinal cryosections and retinal wholemounts 3 weeks after adoptive transfer. Naive eyes were used as control. Cryosections and wholemounts were prepared and stained for Lamc3 (green) and endoglin (magenta) detection. Cell nuclei were stained with Hoechst (blue). a Expression profile of Lamc3 at the mRNA level. b Analysis of Lamc3 expression by immunofluorescence on naive retinal wholemounts. c Analysis of Lamc3 expression by immunofluorescence on EAU retinal wholemounts. d FACS analysis of Lamc3 expression by naïve retinal endothelial (CD31+ Endoglin+ CD45-) cells. e FACS analysis of Lamc3 expression by EAU retinal endothelial (CD31+ Endoglin+ CD45-) cells. Arrows point to possible Lamc3 and endoglin co-staining. Scale bars represent 20 μm 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.

Conclusions
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.