Gene expression profiling in a mouse model of retinal vein occlusion induced by laser treatment reveals a predominant inflammatory and tissue damage response

Purpose Retinal vein occlusion (RVO) has been investigated in several laser-induced animal models using pigs, rabbits and rats. However, laser-induced RVO has been rarely reported in mice, despite the impressive number of available mutants, ease of handling and cost effectiveness. The aim of this study was to further assess the feasibility of a RVO mouse model for gene expression analysis and its possible use to investigate effects of hypoxia. Methods C57Bl/6J mice were injected with eosin Y for photo-sensitization. Subsequently, large retinal veins were laser-treated in one eye to induce vascular occlusion. Contralateral control eyes received non-occlusive retinal laser treatment sparing large vessels. The animals were followed for up to eight days and assessed by funduscopy, angiography, hypoxyprobe staining, histopathology and gene expression analysis by qPCR and RNA sequencing (RNAseq). Another group of mice was left untreated and studied at a single time point to determine baseline characteristics. Results Laser-induced RVO persisted in half of the treated veins for three days, and in a third of the veins for the whole observation period of 8 days. Funduscopy revealed large areas of retinal swelling in all laser-treated eyes, irrespective of vascular targeting or occlusion status. Damage of the outer retina, retinal pigment epithelium (RPE), and even choroid and sclera at the laser site was observed in histological sections. Genes associated with inflammation or cell damage were highly up-regulated in all laser-treated eyes as detected by RNAseq and qPCR. Retinal hypoxia was observed by hypoxyprobe staining in all RVO eyes for up to 5 days with a maximal extension at days 2 and 3, but no significant RVO-dependent changes in gene expression were detected for angiogenesis- or hypoxia-related genes. Conclusion The laser-induced RVO mouse model is characterized by a predominant general inflammatory and tissue damage response, which may obscure distinct hypoxia- and angiogenesis-related effects. A non-occlusive laser treatment control is essential to allow for proper data interpretation and should be mandatory in animal studies of laser-induced RVO to dissect laser-induced tissue damage from vascular occlusion effects.


Introduction
Retinal vein occlusion (RVO) is a common vascular eye disease that causes vision loss due to macular edema, retinal bleeding and ischemia [1,2]. Rare clinical studies examined concentrations of different proteins in vitreous samples of RVO patients [3]. Appropriate animal models to study tissue regeneration following RVO are still being developed.
Earlier reports on RVO animal models provided mixed results. Direct ligation of retinal veins is difficult and only feasible in large animals. In mice, ligation of both, central retinal artery and vein just outside the eye was reported to result in a stroke-like phenotype of whole eye malperfusion [4]. Injection of thrombin (F2a) near a retinal vein in rabbits resulted in a rather brief vascular occlusion [5], and injection of endothelin-1 (ET-1) induced occlusions lasting a few hours at best [6]. The most commonly reported method to occlude retinal veins is the intravenous injection of a photosensitizier such as rose bengal, fluorescein, erythrosin B, or neutral red and subsequent localized activation by an appropriate laser treatment. This strategy appears to work well in pigs [7][8][9][10][11], rabbits [12,13], and rats [14][15][16][17].
In studies with murine RVO models [18][19][20][21][22][23][24][25][26], vascular occlusion was observed by funduscopy as interrupted blood flow at the laser site and as edema detected by fluorescence angiography. Some days after laser treatment, non-perfused retinal areas and vascular leakage were detected. Retinal swelling was detected in OCT images and dilated and tortuous vessels in retinal flatmounts. Structures resembling vascular tufts were reported 2-3 weeks after laser treatment. Laser-induced damage was evaluated in histological sections, but reports characterizing hypoxia were missing.
The aim of our present study was to investigate distinct hypoxia-dependent gene expression changes in the murine laser-induced retinal vein occlusion model. However, employing a non-occlusive control laser treatment, our data revealed a general laser-induced upregulation of inflammatory and wounding-related genes rather than distinct alterations of hypoxia-or angiogenesis-related genes. This would have been missed using untreated control eyes for comparison as suggested in earlier reports. 12 weeks old C57BL/6J mice were used. They were housed at 24˚C with a 12 h light / dark rhythm.

Animals
All mice except the d0 group (the d0 group was designated as untreated) were laser-treated at d0 followed by fundus imaging and angiography. In one eye, large veins were occluded while in the other eye, laser treatment was applied to the retina sparing large vessels as a control. At each of the time points d0, d1, d2, d3, d5, and d8, 18 mice were used for fundus imaging, angiography, and RNA preparation for RNAseq and qPCR, and two additional mice were used for flat mounts stained for hypoxyprobe and Col4 at each time point. Three additional mice were used at d2 for paraffin sections. Five mice were used at d2 or d3 for fundus imaging, angiography, and flatmounts stained for hypoxyprobe and lectin in a preliminary experiment.
All animal procedures adhered to the animal care guidelines of the Institute for Laboratory Animal Research (Guide for the Care and Use of Laboratory Animals) in accordance with the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research and were approved by the local animal welfare committee (Tierschutzkommission).

Laser treatment
During anesthesia with a mixture of 100 mg ketamine and 5 mg xylazine / kg, mice were injected i. p. with 200 mg/kg of a 100 mg/ml solution of eosin Y in 0.9% NaCl to enhance the laser effect. Eosin Y proved to be effective and has several advantages over rose bengal which was usually applied in other studies. Eosin Y has a higher solubility in water (300 g/l), a high singlet oxygen quantum yield [27], and its absorption maximum at 525 nm is much closer to the wavelength of the irradiating laser (532 nm). In addition, eosin is non-toxic and enabled angiography due to its red fluorescence. The settings of the laser (Visulas 532s, Carl Zeiss Meditec, Jena, Germany) were a power of 50 mW, a spot size of 50 μm, and a duration of 2.5 s. 3-6 laser applications were necessary to occlude a vein as detected by funduscopy. All large veins of one eye of each mouse were occluded while the other eye served as control and received the same laser application scheme (number of laser sites, intensity, and distance from the optic disc) in retinal areas between the large vessels, thus avoiding vascular occlusion.

Imaging
Fundus images were taken with a Micron III system (Phoenix Research Labs, Pleasanton, CA, USA). Angiography was performed with rhodamine B (100 mg/ml, 200 mg/kg mouse, i.p.) that nicely fits to the TRITC filter of the Micron system, or with fluorescein and a SLO (HRA1, Heidelberg Engineering, Heidelberg, Germany).
Paraffin sections were prepared for histological examination by standard methods after formalin fixation, paraffin embedding and staining with hematoxylin and eosin (HE).

qPCR
Following fundus imaging and rhodamine B angiography, mice were sacrificed by decapitation. Six retinas from 6 mice were combined to one sample. Three replicate retinal tissue samples were obtained at each time point in each treatment group (laser treatment of all veins or equivalent laser treatment between the large vessels). Veins of one eye of each mouse were occluded while the other eye served as control receiving the same laser application scheme in retinal areas between the large vessels. RNA was isolated from the retinas with the RNeasy kit (Qiagen, Hilden, Germany).
qPCR was performed using standard methods. cDNA was generated with Superscript III reverse transcriptase (Invitrogen, Karlsruhe, Germany) and amplified with the SYBR Premix Ex Taq kit (Takara Bio Europe, Saint-Germain-en-Laye, France). The primers used are presented in Table 1. PCR products were checked on agarose gels and by dissociation curve analysis for single bands. The raw data were analyzed in R (http://www.r-project.org) with the qpcR package [28] that starts with fitting a 6 parameter curve to the raw data. Ct values (from the second derivative) and efficiencies were derived from this curve. Significance was determined by a permutation approach in the propagate function of the qpcR package that was more robust than the Monte Carlo simulation or the error propagation of the same function. PCRs with a replication efficiency of < 1.5 were excluded from further analysis.

RNAseq
The same RNA samples that were used for qPCR were used for RNAseq. Library preparation and RNAseq were carried out as described in the Illumina TruSeq Stranded mRNA Sample Preparation Guide, the Illumina HiSeq 1000 System User Guide (Illumina, Inc., San Diego, CA, USA), and the KAPA Library Quantification Kit-Illumina/ABI Prism User Guide (Kapa Biosystems, Inc., Woburn, MA, USA). In brief, 300 ng of total RNA was used for purifying the poly-A containing mRNA molecules using poly-T oligo-attached magnetic beads. Following purification, the mRNA was fragmented to an average insert size of 200-400 bases using divalent cations under elevated temperature (94˚C for 4 min). The cleaved RNA fragments were copied into first strand cDNA using reverse transcriptase and random primers. Strand specificity was achieved by replacing dTTP with dUTP in the Second Strand Marking Mix (SMM), followed by second strand cDNA synthesis using DNA Polymerase I and RNase H. The incorporation of dUTP in second strand synthesis quenches the second strand during amplification, because the polymerase used in the assay is not incorporated past this nucleotide. The addition of Actinomycin D to First Strand Synthesis Act D mix (FSA) prevents spurious DNA-dependent synthesis, while allowing RNA-dependent synthesis, improving strand specificity. These cDNA fragments then had the addition of a single 'A' base and subsequent ligation of the adapter. The products were purified and enriched with PCR to create the final cDNA library. The libraries were quantified using the KAPA SYBR FAST ABI Prism Library Quantification Kit (Kapa Biosystems, Inc., Woburn, MA, USA). Equimolar amounts of each library were pooled, and the pools were used for cluster generation on the cBot with the Illumina TruSeq SR Cluster Kit v3. The sequencing run was performed on an HiSeq 1000 instrument using the indexed, 50 cycles single-read (SR) protocol and the TruSeq SBS v3 Reagents according to the Illumina HiSeq 1000 System User Guide. Image analysis and base calling resulted in .bcl files, which were converted into .fastq files with the CASAVA1.8.2 software. Library preparation and RNAseq were performed at the service facility "KFB-Center of Excellence for Fluorescent Bioanalytics" (Regensburg, Germany; www.kfb-regensburg.de). Data were evaluated in R with packages from Bioconductor (http://bioconductor.org). Reads were mapped to the mouse genome (Gencode_vM11) with STAR 2.5.2b (https://github. com/alexdobin/STAR) [29]. Features were counted with the featureCount function of the Rsubread package (Subread 1.5.1 for R [30]), and differential gene expression was analyzed with DESeq2 1.14.1 [31]. DESeq2 applies a set of sophisticated statistical tools including normalisation between samples, empirical Bayes shrinkage for dispersion estimation and for fold-change estimation, and correction for multiple testing. Gene ontology analysis was done with debrowser 1.2.4.2 and GO.db 3.4.0. The data are available under the accession number GSE101398 at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE101398.

Retinal vein occlusion
Retinal veins were occluded by laser treatment after systemic application of eosin Y as a photosensitizer. The occlusion was visible during slit lamp laser treatment as the vein became In a preliminary experiment, two veins were occluded in two mice at d0. Fluorescein angiography by SLO showed the occlusion of the veins, and fundus images taken by SLO are shown for comparison. Retinal flatmounts were stained for vessels with lectin (green) and with hypoxyprobe (red) for hypoxic areas. The squares show the positions of the SLO images while the arrows show the position of laser sites. In one mouse at d2, the vein to the left was re-opened while the lower vein was still closed (arrow) and surrounded by a hypoxic area. In the other mouse at d3, the lower vein was re-opened (arrow) but showed some remnants of the hypoxic area. Typical results from a study of 5 mice are shown.
https://doi.org/10.1371/journal.pone.0191338.g001 narrow at the laser site and blood flow subsided. Subsequent angiography revealed leaking vessels at the laser sites. Two days after laser treatment, the retinal area distal from the treatment site showed reduced perfusion and hypoxic areas as detected by hypoxyprobe staining (Fig 1). A time course analysis of vessel occlusion showed that half of the vessels stayed closed for at least three days (Fig 2). Around one third of the veins were still unperfused after eight days.
At funduscopic examination, all laser-treated retinas showed large whitish edematous areas at the laser sites (Fig 3), irrespective of laser targeting. These areas were somewhat more extended in RVO eyes as compared to controls, but the difference was not statistically significant. (Fig 3 and Fig 4).
Examination of laser sites in serial paraffin sections stained with hematoxylin and eosin (HE) showed damage in outer retinal layers including photoreceptor outer segments and retinal pigment epithelium and, in severe cases, even choroid and sclera (RPE, Fig 5 and supplemental S1 and S2 Fig that show serial sections of two laser sites). Typically, a large  RVO was induced at d0 in two mice per time point. Fundus images were taken by a Micron III camera and flatmounts were prepared at the indicated days. Flatmounts were stained for vessels with an antibody raised against collagen IV (Col4, green) and with hypoxyprobe (red) for hypoxic areas. Flatmounts at d0 were not laser-treated. In one eye of each mouse, the veins were occluded (RVO) while the other eye was laser-treated between the large vessels (control). The circle indicates the area of the corresponding fundus image. Hypoxic areas were found from d1 to d3, diminished at d5 and disappeared completely at d8. fold was generated from parts of the outer retina. The inner retinal layers appeared to be less affected. In addition, serous exsudation was observed within the retinal ganglion cell layer showing detachment of the inner limiting membrane which fits the angiographic and funduscopic images. The inner nuclear layer showed serous exsudation, too. The defects of ocular tissues generated by laser treatment were similar in RVO and in control laser treatments. In cross sections of the vein, a fibrous plug generated by the laser treatment after injection of a photosensitizer was visible. The proximal part of the vein was much smaller than the distal part.

Hypoxia time course
Hypoxia was detected by hypoxyprobe staining in all RVO eyes from d1 to d5 and disappeared afterwards (Fig 3). No hypoxia was found in the control eyes following retinal laser treatment No hypoxia staining was observed after intervenous laser treatment in the control group. The large edematous areas (gray areas of the fundus images) at the laser sites were observed both in the RVO and the control group. They resulted in weaker hypoxia staining (compare the RVO eye at d2).
https://doi.org/10.1371/journal.pone.0191338.g003  sparing large vessels. While the hypoxic areas in RVO eyes were rather homogeneous during the first days, they became disjunct during reperfusion so that increasingly smaller hypoxic areas remained within the capillary regions until hypoxia was no longer detectable.

Gene expression profiling
Next, we investigated the effect of hypoxia on global gene expression patterns by RNAseq at d2. RVO eyes were compared to control laser-treated eyes as well as to untreated controls.   There were large differences between RVO eyes and untreated controls, as well as between control laser-treated eyes and untreated controls. The differences between RVO eyes and control laser-treated eyes were much smaller (Fig 6). Similarly, while the maximal fold change was 175 for comparisons between RVO and untreated eyes it was only 8 for comparisons between RVO and control laser treatment ( Table 2). This indicates that it is important to use an appropriate laser-treated control group. To gain further insight into possible functional implications of the gene expression alterations, a gene ontology analysis was performed ( Table 3). The most significantly affected ontology groups in the comparison of RVO and untreated eyes were related to inflammation or wounding. In contrast, in the comparison of RVO and control laser-treated eyes, some ontology groups related to angiogenesis were affected. Ontology groups related to hypoxia did not show up among the first 100 groups with the highest significance score.  In contrast, expression of hypoxia-responsive genes was not increased. Factors like Eno1, Mmp9, Glut1, Pfkl, Pgk1, Pkm, Ldha, Met, Tgfb3, Aldoa, Glut3, and Angpt2 that have hypoxia responsive elements within their promoters, as well as Hif1a and Hif2a were up-regulated not more than 2-fold (Figs 8 and 9). Some of them, Aldoa, Glut3, Tgfb3, and Ldha were downregulated.
Many factors showed an increased expression at d8 which was in contrast to the reduced expression of inflammatory factors at d8. This may indicate reduced inflammation and increased repair of the damaged tissue.
The factors Vegfa, Vegfr1, Angpt2, and Nrp2 are not only hypoxia responsive genes but also angiogenesis factors. Vegfa showed a peak at d1 and d2, however, the peak was small and its physiological relevance is not clear.

Discussion
Retinal vein occlusion by laser treatment of retinal veins has been described in several species including pig, rabbit, and rat. But there are only few reports in mice.
We used eosin Y as a photosensitizer and subsequent laser treatment to induce RVO in mice. Our finding of persistent hypoxia for up to 5 days after retinal vein occlusion is in line with other reports [19,25]. Nevertheless, expression of genes typically altered in hypoxia were not upregulated in whole retina samples obtained in the hypoxic period (Fig 7). Instead, an intense inflammatory response dominated the gene expression changes. Earlier reports described similar increases of inflammatory factors like Ccl2, Il1b, Il6, and Tgfb1 [18,21,32,33].
This intense inflammatory response corresponds to the large edematous areas found in fundus images as well as in histological sections. Other studies described these areas as hemorrhage and cystoid edema [21] which are useful for investigating inflammatory processes. The laser treatment additionally induced severe damage to the outer retina, the RPE and often to the choroid at the laser site. Similar destructions including damage of the outer retina were reported earlier [20,21].
The large laser damage and the intense inflammatory response might have obscured angiogenesis-or hypoxia-related gene expression changes. While the size of the laser lesion in our study is comparable to the lesion size of studies with larger animal species like rat or pig, the resulting hypoxic area of the occluded vein is much smaller in mice. That means that the ratio of the area of laser damage to the hypoxic area is disadvantageous in mice and the influence of the damage-induced inflammatory response is much greater than in larger animals rendering mice a rather unattractive model for RVO. Indeed, such an intense inflammatory response was not found in proteomic studies of RVO in pig [10,11]. Our observation might have been missed when using untreated controls only and stresses the importance of appropriate control laser treatment groups.
As a consequence, it would be desirable to reduce the laser damage to a minimum. In our experiments, we used a laser radiant flux of 50 mW and a spot size of 50 μm as in similar reports of other groups, and this resulted in an irradiance of 25 MW / m 2 . In the first description of photochemical induction of thrombosis with rose bengal [34], an irradiance of 6.4 kW / m 2 was applied for 20 min. This irradiance is 4000 times less than ours what mRNA expression analysis (qPCR) of genes strongly up-regulated in RVO. The first 4 factors had an expression maximum at d1 while the others had a maximum at d2 or d3. All of them are involved in inflammation or responses to stress. Most interestingly, the up-regulation found in RVO was also detected in the controls that were laser-treated between the large vessels. Note the logarithmic expression scale that shows equal distances for factors that are up-or down-regulated. Asterisks at the right of the corresponding data point indicate significant differences as compared to d0. Mice at d0 were not laser-treated.
https://doi.org/10.1371/journal.pone.0191338.g007  may explain the extent of retinal damage. Therefore, less irradiance should be used in further experiments to reduce damage-induced inflammatory response. 735 W / m 2 were shown to occlude retinal vessels in rat [35], and this light intensity is only slightly above the physiological range [36,37].
In summary, the effect of laser wounding dominates the response in mice while the relation of the size of the laser spot to the whole retina may be more favorable in larger animals like rat or pig. It is therefore essential to include the appropriate laser wounding control when investigating the gene expression profile of laser-induced vascular occlusion models in small animals. Furthermore, conditions would be necessary where the vein occlusion is achieved without direct laser-induced tissue damage.