Topical delivery of a small molecule RUNX1 transcription factor inhibitor for the treatment of proliferative vitreoretinopathy

Proliferative vitreoretinopathy (PVR) is the leading cause of retinal detachment surgery failure. Despite significant advances in vitreoretinal surgery, it still remains without an effective prophylactic or therapeutic medical treatment. After ocular injury or retinal detachment, misplaced retinal cells undergo epithelial to mesenchymal transition (EMT) to form contractile membranes within the eye. We identified Runt-related transcription factor 1 (RUNX1) as a gene highly expressed in surgically-removed human PVR specimens. RUNX1 upregulation was a hallmark of EMT in primary cultures derived from human PVR membranes (C-PVR). The inhibition of RUNX1 reduced proliferation of human C-PVR cells in vitro, and curbed growth of freshly isolated human PVR membranes in an explant assay. We formulated Ro5-3335, a lipophilic small molecule RUNX1 inhibitor, into a nanoemulsion that when administered topically curbed the progression of disease in a novel rabbit model of mild PVR developed using C-PVR cells. Mass spectrometry analysis detected 2.67 ng/mL of Ro5-3335 within the vitreous cavity after treatment. This work shows a critical role for RUNX1 in PVR and supports the feasibility of targeting RUNX1 within the eye for the treatment of an EMT-mediated condition using a topical ophthalmic agent.


RUNX1 expression in human PVR membranes. PVR membranes as shown in
, are usually triggered as a consequence of ocular trauma, and unlike PDR, PVR is not driven by aberrant retinal angiogenesis, but instead is considered as an excessive wound-healing process. We found widespread RUNX1 expression in human surgical PVR samples from a total of four donors using immunohistochemistry (Fig. 1b). This was surprising because PVR membranes are devoid of blood vessels and known instead to consist of cells at various stages of the EMT continuum 31,[55][56][57] . RUNX1 expression was identified in areas of cell proliferation identified by Ki67 expression, a marker of active cell division (Fig. 1b).

RUNX1 inhibition limits PVR proliferation ex-vivo.
We developed an ex-vivo model of PVR by growing fragments of freshly isolated human PVR membranes in Matrigel to examine a potential role for RUNX1 in growth of PVR membranes. Phase contrast microscopy of Matrigel-embedded untreated control samples showed long and robust outgrowths from a freshly isolated human PVR membrane over a period of 3 to 7 days (219.9 ± 31.7 μm and 326.2 ± 52.1 μm). In contrast, we observed almost no outgrowths in samples treated with Ro5-3335 (150 μM), a small molecule RUNX1 inhibitor 54 (Fig. 1c), and the few outgrowths present were significantly shorter after 7 days (136.9 ± 18.9 μm and 180.3 ± 27.9 μm) (**p < 0.01). This finding was confirmed in PVR membranes from two additional donors suggesting a consistent link between RUNX1 function and PVR membrane growth.

RUNX1 regulates proliferation and TGF-β2-induced EMT in C-PVR.
We analyzed the expression of RUNX1 in primary cultures from PVR membranes (C-PVR) obtained from three donors generated using our previously published protocols to determine whether RUNX1 expression was retained upon cell culture. We confirmed robust RUNX1 expression in these proliferating cultures under basal conditions in approximately 99% of cells from all donors (Fig. 1d). Inhibition of RUNX1 function using Ro5-3335 54 , strongly inhibited the proliferation of C-PVR in a dose-dependent manner, as measured by the CyQUANT Direct Cell Proliferation Assay (Fig. 1e), suggesting a functional role for RUNX1 activity in PVR. Very low levels of cell death were detected by lactate dehydrogenase (LDH) analysis suggesting a direct effect in proliferation by RUNX1 inhibition (see Supplementary Fig. S1). We sought to establish an in vitro model of EMT using our C-PVR cultures to further examine a potential functional role of RUNX1 in PVR. We found that TGF-β2 treatment strongly induced the expression of EMT markers including α-SMA and N-cadherin in C-PVR using immunofluorescence (Fig. 2a,b) and Western blot analyses (Fig. 2c,d). RUNX1 expression was also strongly induced by TGF-β2 treatment (Fig. 2c,d). These effects were detected at 3-days after treatment and significantly increased over time by day 7. TNF-α or IL-6 treatments failed to induce EMT markers or RUNX1 expression and the combination Scientific Reports | (2020) 10:20554 | https://doi.org/10.1038/s41598-020-77254-0 www.nature.com/scientificreports/ treatment with all three growth factors did not appear to be more effective than TGF-β2 alone suggesting lack of additive or synergistic effects. We used siRNA to knockdown RUNX1 expression in C-PVR to determine whether RUNX1 expression was necessary for TGF-β2-induced EMT. We found that RUNX1 knockdown effectively blunted the response of C-PVR to TGF-β2 treatment by preventing morphological changes associated with EMT and by reducing N-Cadherin expression (Fig. 2e). A significant reduction in the induction of the EMT marker α-SMA by TGF-β2 treatment was also observed using immunofluorescence (Fig. 2f). Under our experimental conditions siRNA effectively reduced RUNX1 expression by 70% as determined by qRT-PCR (Fig. 2g). Downregulation of RUNX1 via siRNA was validated by Western blot analysis using mouse anti-RUNX1 and rabbit anti-RUNX1 antibodies (Fig. 2h).
Development of a rabbit model of PVR using human C-PVR cells. We injected 1 × 10 6 C-PVR cells per eye inside the vitreous cavity in nine adult rabbits to develop a novel model of PVR using human PVR primary cultures (Fig. 3a). Placement of the cells over the optic nerve area was confirmed in live rabbits using fundus imaging and optical coherence tomography (OCT) immediately after cell injection. We evaluated the progression of the disease one-week after injection and performed indirect ophthalmoscopy, fundus, and OCT imaging at two-and four-weeks. Rabbits were euthanized prior to tissue collection after imaging at the 4-week time point.
We observed robust growth of intravitreal and epiretinal membranes at the 2-week time point in all animals. In 3 out of 9 rabbits these membranes caused focal traction of the retina, a hallmark of PVR. Most common findings identified were the formation of vitreous floaters, vitreous strands, intravitreal and epiretinal membranes Large preretinal membranes are observed creating retinal folds (left). Representative macroscopic appearance of human specimen obtained during PVR surgery (right). Note high degree of pigmentation denoting the presence of RPE cells within the pathological excised tissue. Scale bar: 400 μm. (b) RUNX1 positive cells are clearly identified in human PVR specimens obtained from four different donors. Positive Ki67 labelling was found in those cells with an active proliferative state within the tissue. Scale bar: 400 μm. (c) Cell proliferation and sprouting in human PVR explants were evaluated. A reduction in sprouting distance from the specimen is observed 7 days in explants treated with 150 μM Ro5-3335 (**p < 0.001, two-way ANOVA; n = 3 represented as mean ± SEM). Scale bar: 400 μm. (d) C-PVR cells from three different donors showed positive staining with RUNX1 antibody. Scale bar: 400 μm. (e) RUNX1 inhibition with Ro5-3335 inhibitor reduces C-PVR cells proliferation in vitro in a dose-dependent manner (25 μM, 50 μM and 100 μM), at 48 and 72 h (**p < 0.01, ***p < 0.001, ****p < 0.0001, one-way ANOVA; n = 4 represented as mean ± SEM).

Scientific Reports
| (2020) 10:20554 | https://doi.org/10.1038/s41598-020-77254-0 www.nature.com/scientificreports/ with or without retinal traction, which were seen at 2-and 4-weeks (Fig. 3b,c). We developed a scoring system using a combination of clinical and imaging characteristics of PVR including presence of vitreous floaters, membranes, and traction of retinal tissue to grade PVR severity (see Supplementary Table S1). Inclusion of imaging features is important because prior grading systems, such as the Fastenberg scale, traditionally used to grade PVR, only assessed clinical criteria as it was developed before OCT became widely available 58 . Using our PVR score system, we observed a progression of the phenotype from the time of injection to the 2-week time point, but we observed no additional progression beyond this time point after 4 weeks of follow-up (Fig. 3d). Therefore, we used the two-weeks time point for preclinical efficacy experiments because there was no significant change afterwards. A high degree of reliability was found between graders when performing PVR Score measurements. The average intra-class coefficient was 0.976 with a 95% confidence interval from 0.957 to 0.986. Histological analyses of whole eye sections showed PVR-like findings that accurately correlated with the live imaging observations including intravitreal and epiretinal membranes with or without traction (Fig. 3e). We also demonstrated that PVR-like membranous tissue inside the rabbit eye was of human origin using human anti-HLA antibody (Fig. 3e). Detection of Proliferating Cell Nuclear Antigen (PCNA) positive staining confirmed the presence of proliferating cells within pathological tissues (Fig. 3f). Fibronectin staining, a marker of extracellular matrix deposition, was also identified in experimental PVR membranes (see Supplementary Fig. S2).

Development of a nanoemulsion formulation of a RUNX1 inhibitor for topical ophthalmic administration.
To test the preclinical efficacy of RUNX1 inhibition using Ro5-3335 in our newly developed rabbit model, we developed a formulation ideally suited for topical ophthalmic administration using eye drops. Ro5-3335 is a highly lipophilic small molecule, thus, we designed a nanoemulsion formulation utilizing surfactants as the encapsulation matrix within an aqueous phase. We used lecithin and isopropyl myristate, as the oil phase, non-toxic matrices which have been previously used in the eye 59 . For the aqueous phase we used phosphate buffered saline (PBS). Analysis of the nanoemulsion, which we called eNano-Ro5, using dynamic light scattering (DLS) showed that our formulation resulted in particles with an average radius of 4 nm with a unimodal distribution (Fig. 4a). The nanoemulsion remained stable in size for at least 11 days increasing to only about 8 nm in size by 17 days (Fig. 4b). We also investigated the release profile of Ro5-3335 from eNano-Ro5. Ro5-3335 released from the nanoemulsion was sustained for 24 h in vitro, a time at which a rapid equilibrium was reached. (Fig. 4c). We confirmed penetration of topical application of eNano-Ro5 within rabbit ocular structures by liquid chromatography-tandem mass spectrometry (LC-MS/MS). Using 100 μL of 7.92 mM eNano-Ro5 administered three times a day over 4 weeks, detectable levels of Ro5-3335 were found in the cornea (0.03 ng/ mg), aqueous humor (13.15 ng/mL) and vitreous (2.67 ng/mL) (  Growth factor-induced EMT in C-PVR cells show upregulation of EMT markers. (a) After stimulation with TGF-β2 or combination treatment (TGF-β2 + TNF-α + IL-6), significant changes in α-SMA staining were identified at day 3. These changes in α-SMA expression were more prominent 7 days after induction. (b) Increased α-SMA expression levels were detected after stimulation with TGF-β2 and combination when compared to control at day 3 (*p < 0.05, ***p < 0.001, one-way ANOVA; n = 6 represented as mean ± SEM) and day 7 (***p < 0.001, ****p < 0.0001, one-way ANOVA; n = 6 represented as mean ± SEM). (c) TGF-β2 and combination treated cells showed increased protein expression of mesenchymal markers (α-SMA and N-Cadherin) at days 3 and 7 after induction. RUNX1 protein levels were also upregulated after induction by TGF-β2 and combination treatment. (d) An increase in N-Cadherin was observed at day 3 and in α-SMA protein levels at day 3 and day 7 after induction by TGF-β2 and combination treatments (*p < 0.05, ***p < 0.001, two-way ANOVA; n = 3 represented as mean ± SEM). Similarly, increased protein expression of RUNX1 was observed at day 3 and day 7 when C-PVR cells were stimulated with TGF-β2, and at day 7 with the combination of growth factors (*p < 0.05, **p < 0.01, ***p < 0.001, two-way ANOVA; n = 3 represented as mean ± SEM). (e) N-Cadherin expression is reduced by RUNX1 knockdown via siRUNX1 in untreated and TGF-β2-induced cells (*p < 0.05, ****p < 0.0001, two-way ANOVA; n = 3 represented as mean ± SEM). (f) TGF-β2-induced α-SMA expression is reduced by RUNX1 knockdown by siRUNX1 (*p < 0.05, one-way ANOVA; n = 12 represented as mean ± SEM). (g) siRUNX1 induced a 70% reduction of RUNX1 expression measured by qRT-PCR 48 h post-transfection (***p < 0.001, two-tailed unpaired T-test; n = 3 represented as mean ± SEM). (h) Validation of siRUNX1 effect on RUNX1 using mouse and rabbit anti-RUNX1 antibodies. Protein levels quantification of RUNX1 showed 60% and 50% reduction of RUNX1 (*p < 0.05, **p < 0.01, two-tailed unpaired T-test; n = 2 represented as mean ± SEM). Samples used for quantitative comparisons derive from the same experiment and blots were processed in parallel.  ▸ RUNX1 inhibition reduces PVR in a rabbit model. We tested the preclinical efficacy of Ro5-3335 to reduce progression of PVR by treating rabbits starting immediately after cell injection with 100 μL of eNano-Ro5 or vehicle, applied as eye drops on the ocular surface three times a day. We found that rabbits treated with eNano-Ro5 had a significant reduction of PVR severity after 14 days of follow-up compared to vehicle-treated controls (*p < 0.05). Animals treated with eNano-Ro5 had a median PVR score of 2 (n = 16) versus animals treated with the vehicle control, which had a median PVR score of 3 (n = 17) (Fig. 4f,g). Due to feasibility we were not able to include side-by-side comparisons of untreated control rabbits therefore an effect of the vehicle alone cannot be assessed.

Discussion
PVR is the leading cause of retinal detachment surgery failure and to date, despite significant advances in vitreoretinal surgery, it remains without an effective prophylactic or therapeutic medical treatment 60 . In this study, we identified RUNX1 as a mediator of EMT in PVR. Using our previously developed primary culture model of PVR and a newly developed PVR in vivo model, we successfully inhibited EMT and PVR progression via RUNX1 inhibition. To our knowledge this is the first report of successful preclinical modulation of a transcription factor to improve outcomes of an ocular condition driven by EMT [46][47][48][50][51][52][53] . RUNX1 was detected in a subpopulation of cells that are in a highly proliferative stage in human PVR surgical specimens. Histopathology confirmed that RUNX1 expression in PVR was more widespread than the vascular pattern we previously reported for RUNX1 in PDR membranes 41 . Additionally, C-PVR cells also displayed high levels of RUNX1 expression in vitro. When these cells were stimulated by growth factors known to promote EMT (TGF-β2), RUNX1 expression levels were significantly increased along with mesenchymal markers including N-Cadherin and α-SMA. This response was effectively prevented by RUNX1 knockdown suggesting a mechanistic link between RUNX1 expression and EMT in PVR.
We generated a new rabbit model of PVR using injection of human primary cultures derived from PVR membranes in an attempt to reproduce some of the complexity of the disease. Rabbit is commonly used to model PVR 21,33 . Ocular anatomic characteristics such as a large globe, vitreous cavity, and relatively small lens support the use of rabbits as a suitable animal model to study PVR 21,33 . Importantly, rabbits are widely used in research to test pharmacologic efficacy, predictability, ocular safety and tolerability of new drugs and drug delivery devices due to anatomic and physiological similarities with human eyes 61,62 . Previous animal models of PVR have used surgical manipulation or injection of dermal fibroblasts or transformed cell lines to identify candidate treatments but to date these have not translated into effective therapies 21,39,60 .
Using our rabbit model, we showed that topical application of a nanoemulsion containing a RUNX1 inhibitor effectively reduced progression of PVR. This suggests that RUNX1 modulation could be a novel therapeutic strategy for complications of retinal detachment and ocular trauma. Unlike other published models, in our approach we injected C-PVR cells intravitreally but avoided surgical manipulation such as lens extraction, induction of retinal breaks or injection of blood or plasma. As a result, we obtained a mild phenotype that does not cause retinal detachment. We reasoned that our system is therefore useful to model early stages of the disease were migration and proliferation of cells is occurring and pharmacological treatments could help. This modality of treatment would be suitable in patients to prevent the onset or progression of PVR before and/or after surgical repair of retinal detachment or ocular trauma, when the disease is mild (i.e. early stages), and medical treatments are still a therapeutic option. PVR animal models that result in a very severe phenotype including retinal detachment with massive retinal traction and retinal tears may not be suitable to practically assess the preclinical efficacy of a pharmacological treatment because, much like in humans, the treatment for these robust phenotypes is surgery.
The use of a noninvasive topical ophthalmic formulation instead of intravitreal injections may offer advantages for prolonged drug administration in situations with limited access to medical facilities or highly trained clinicians or under social distancing conditions, as they can be self-administered. Intravitreal drug injections,  [63][64][65][66] . In addition, albeit very effective, intravitreal injections of therapeutic agents are commonly administered in a monthly regime, impacting patient quality of life 67 . In the context of PVR, dislodgement of RPE cells and invasion of immune cells starts immediately after retinal or ocular trauma. Therefore, preventing PVR as soon as feasible with an efficacious drug could potentially have significant clinical impact. Thus, there is a clinical need for alternative, less invasive drug delivery methods that could be easily administered to temporize the eye and mitigate damage from this devastating disease. This study has some limitations. First, there is a need to elucidate the precise mechanisms of how Ro5-3335 affects RUNX1 transcriptional activity and its downstream targets in PVR. Secondly, in developing our rabbit model we avoided surgical manipulation such as lens extraction, induction of retinal breaks or injection of blood or plasma, and merely injected C-PVR cells intravitreally. As a result, we obtained a mild phenotype and further Comparison between non-extracted and extracted samples at a concentration of 50 μg/mL. (f) Quantification of PVR severity within both study groups adhering to PVR Score grading system. A reduction in PVR Score was found in the group treated with eNano-Ro5 (n = 16) compared to vehicle (n = 17) (*p < 0.05, two-tailed Mann-Whitney test). The line is denoting the median of each studied group. (g) Representative imaging results in both groups. In the vehicle-treated group, multiple cell clumps and deposits can be identified after cell injection (arrowheads). These findings are also observed by OCT (arrows). Increased cell density was observed after 2 weeks of treatment with vehicle. A reduction in cellular density and progression was observed after 2 weeks of treatment with eNano-Ro5 by fundus imaging (arrowheads) and OCT (arrows). Green line depicts OCT scan area, dotted lines represent a detailed area of posterior pole images.

Scientific Reports
| (2020) 10:20554 | https://doi.org/10.1038/s41598-020-77254-0 www.nature.com/scientificreports/ research using models with more severe disease may be warranted. Third, because of feasibility we examined safety of our treatment via ERG in mice after 1 week of treatment. Longer exposures and ERG testing in other animal models need to be considered for future experiments. EMT is a common process involved in multiple ocular diseases including corneal fibrosis, glaucoma, PVR, PDR and age-related macular degeneration. Hence, the link between RUNX1 and EMT identified in this research could be of relevance in the understating of the pathogenesis of these diseases. In addition, the development of drugs targeting RUNX1 may have multiple applications in the treatment of ocular conditions leading to blindness since RUNX1 also plays a critical role in aberrant ocular angiogenesis 41 .

Materials and methods
Sample size. Total animal numbers needed to perform this study were calculated accordingly with the number of animals used in prior in vivo studies, and by sample size power calculation using data from a pilot experiment which was adjusted for non-parametric distribution to detect a difference between the means of 35% with 80% power (and α = 0.05). These numbers represent the number of animal needed to achieve statistical significance and experimental reproducibility. Specifically, the sample size was calculated by using a formula established for parametric variables and then correcting for the non-parametric condition as previously described (Erich L. Lehmann, Nonparametrics: Statistical Methods Based on Ranks, Revised, 1998).

Randomization.
For the purpose of specimens analyzed for immunofluorescence, in vitro and ex vivo experiments, patients undergoing surgery for repair of retinal detachment and removal of PVR membranes were randomly chosen. An ex vivo preliminary experiment was done to determine the time points in which the peak of the outgrowths was consistent throughout the samples. For the proliferation assays, time points were selected to represent the observed growth rate of our cultures and doses were chosen to be consistent with previously reported IC50 data. For in vitro growth factor EMT-induction assays, time points were based on the characteristic phenotype of the cells and on what has been shown in previous studies. In vivo data analysis was performed individually in a randomized and masked manner by two experienced physician observers trained in clinical ophthalmology. An average of both evaluations was obtained.

Data exclusion criteria. Animals with presence of subretinal, preretinal or vitreous hemorrhage after PVR
induction procedure were excluded of the study.
Replicates. All the in vitro experiments were conducted using three biological repeats and were performed in three independent experiments. For the preclinical safety and efficacy evaluation in vivo, three independent experiments were performed and data shown correspond to pooled results. www.nature.com/scientificreports/ Equipment and settings for immunoblots. All the images were acquired and quantified using the Image Studio version 2.1 (LI-COR Biosciences, Lincoln, NE). All the blots were scanned at an intensity of 4.0. The brightness was adjusted uniformly across the entire blot using the Image studio software to visualize the bands for each experiment. Original complete blots are shown in supplementary data and citations for each antibody provided in the methods section.

Study population. This study was performed at the Schepens Eye Research Institute of Massachusetts Eye
and Ear, and research protocols were approved by the Institutional Review Board at Massachusetts Eye and Ear for the collection of surgical specimens and for the retrospective analysis of clinical data. All research protocols adhered to the tenets of the Declaration of Helsinki, signed informed consent form and Health Information Portability and Accountability Act authorization were obtained from all the participants included in the study. Nine patients were recruited from Massachusetts Eye and Ear who had grade C PVR and were undergoing PVR surgery. The demographics of the patients are summarized in Table 1. Patients had to be at least 18 years old and could not be pregnant. Immediately after surgical extraction, PVR membranes were placed in a specimen cup containing calcium-free and magnesium-free balanced salt solution (BSS) for transportation at room temperature (RT). Within 1-2 h after extraction, membranes were processed under sterile biosafety level 2 conditions for immunofluorescence, cell culture or ex-vivo membrane culture. Biospecimen reporting for improved study quality (BRISQ) reporting guidelines for human biospecimens were followed to complete this section 68 .
Immunofluorescence analysis of PVR membranes. Human PVR membranes were fixed and processed for analysis using our previously published protocol 40 . Sections from 4 different human donors were processed for immunofluorescence using the following antibodies: anti-RUNX1 (1:100; LS-B13948; Lifespan Biosciences, Seattle, WA), and anti-human Ki67 antibody (1:100; NB500-170; Novus Biologicals, Ontario, Canada). For heat-induced antigen retrieval the slides were boiled in 10 mM sodium citrate buffer (pH 6.0) and then maintained at a sub-boiling temperature (95-100 °C) for 20 min and subsequently cooled on the bench top for 30 min. Slides were washed with distilled water and permeabilized with 0.5% Triton X-100 in PBS for 5 min and blocked (10% goat serum in PBS) for 1 h at RT. The primary antibody was prepared in antibody dilution buffer (5% goat serum) and samples were incubated overnight with the antibody solution at 4 °C. Membranes were washed with PBS and incubated with goat anti-rabbit Alexa Fluor 594 secondary antibody (1:300; A-11012; Invitrogen, Carlsbad, CA) for 2 h at RT. Slides were mounted and visualized using Prolong Gold Antifade Reagent with DAPI (P36935, Invitrogen). Images were obtained using an EVOS FL automated stage live cell imaging system (Life Technologies, Cambridge, MA).

C-PVR culture and growth factor induction. Immediately after surgery, the human PVR membranes
were processed for isolation and cell culture (C-PVR) using our previously established protocol 40 . C-PVR cells (30 × 10 3 /well) were seeded in 48-well plates for a period of 24-72 h. Then, they were washed with PBS and treated with PVR media alone (control) or in PVR media with TGF-β2, TNF-α, IL-6 or a combination of all three growth factors, using triplicates for each condition. We incubated our C-PVR cells with these growth factors, which have been previously used by others to induce EMT in vitro 19,47,69 . Immunocytochemistry and assessment of EMT using immunofluorescence. Our previously established protocol for immunocytochemistry was used to detect the presence of RUNX1 in culture cells from 3 human donors 40 . Cells were fixed with 4% PFA for 10 min, washed with PBS, permeabilized with 0.5% Triton X-100 in PBS for 5 min, and blocked (10% goat serum in PBS) for 1 h at RT. The cultures were incubated with primary antibody rabbit anti-RUNX1, (1:100; LS-B13948, LifeSpan BioSciences) overnight at 4 °C followed by incubation with goat anti-rabbit Alexa Fluor 594 secondary antibody (1:300; A-11012, Invitrogen). Following induction with the growth factors cells were incubated with the primary antibody mouse anti-α smooth muscle actin (1:250; M0851, Dako, Glostrup, Denmark) overnight at 4 °C followed by incubation with goat anti-mouse Alexa Fluor 488 secondary antibody (1:300; A-11001, Invitrogen) for 2 h at RT. Cell nuclei were labeled with Hoechst 33342 diluted in PBS (1:200, 639, ImmunoChemistry Technologies, Bloomington, MN) and counted for quantification using ImageJ software (National Institutes of Health, Bethesda, MD) using our previously established methods protocol 40 . Cells were washed with PBS and images were obtained using an EVOS FL automated stage live cell imaging system (Life Technologies).

RUNX1 inhibition in an ex vivo model of PVR using Ro5-3335. PVR membranes were divided into
pieces and embedded in growth factor reduced Matrigel (354,230; BD Biosciences, San Jose, CA) in a 24-well plate and placed at 37 °C for 30 min for the Matrigel to solidify. The specimens were treated with either vehicle or 150 μM of a small molecule inhibitor of RUNX1 (Ro5-3335; 219506, Millipore Sigma, Darmstadt, Germany) in 500 μl in PVR growth media. Phase contrast images using an EVOS FL imaging system (Life Technologies) were taken and the distance of growth from the embedded tissue was quantified using Image J (National Institutes of Health). ) were also applied. Oxygen levels and heart rate were continuously monitored until complete anesthesia recovery. Right eye of each animal was used to develop the model and left eyes were used as controls. To develop our model, we first generated a master stock of C-PVR cells from a donor to be used at a similar passage for intravitreal injections to maintain uniformity of the system between experiments. Gas displacement of the vitreous was induced by intravitreal injection of 0.15 mL of perfluoropropane (C 3 F 8 ) (Alcon, Fort Worth, FL) 3 mm behind the limbus. Three days after gas displacement, intraocular pressure (IOP) was reduced by gas withdrawal. Subsequently, 0.2 mL of BSS containing approximately 1 × 10 6 C-PVR cells were intravitreally injected. These cells were obtained and processed following our previously published method 40 . IOP monitoring with Tonopen (Reichert Technologies, Depew, NY) and topical postoperative treatment with 0.5% Timolol (Akorn), 0.3% Ofloxacin (Akorn) and triple antibiotic ointment (Akorn), were performed for 3 days after gas displacement and after cell injection. Rabbits were examined using indirect ophthalmoscopy after cell injection, 1 week, 2 weeks and 4 weeks. For fundus and OCT imaging, a spectral-domain OCT Spectralis (Heidelberg Engineering, Heidelberg, Germany) was used immediately after cell injection, at 2 and 4 weeks of follow-up. A PVR score grading system was developed to assess severity of disease. The score was determined by combination of the most severe phenotypes identified by indirect ophthalmoscopy, fundus imaging and OCT. Representative images of each severity stage identified are showed in Supplementary  Fig. S10. Under deep anesthesia, animals were euthanized and eyes collected after 14 or 28 days of follow-up by injection of pentobarbital (120 mg/kg) (Fatal Plus, Vortech, Dearborn, MI). eNano-Ro5 characterization. Particle size was characterized by DLS equipment with a DynaPro NanoStar (Wyatt Technology, Santa Barbara, CA) with an operation angle of 90°. The data were collected in a series of ten measurements from duplicate at 25 °C. In order to evaluate stability of the nanoemulsion formulation, changes in nanodrop size were evaluated by DLS at different time points. To quantify the amount of Ro5-3335 released from the nanoemulsion, eNano-Ro5 was dispersed in 1 mL of PBS (pH 7.4) to a final concentration of 0.021 mg/ ml. The mixture (1 mL) was placed into a Float-A-Lyzer G2 Dialysis Device (3.5-5 kD) (1210W03, Thomas Scientific, Swedesboro, NJ) and free dialyzed in 200 mL of PBS, to allow sink conditions. The system was kept at 37 °C and 150 strokes. At regular time intervals, aliquots of the release medium were withdrawn and replaced with an equal volume of fresh buffer. Finally, the concentration of Ro5-3335 was measured by LC-MS/MS.

Ocular toxicity assessment.
Quantification of Ro5-3335 by liquid chromatography-tandem mass spectrometry in rabbit ocular tissues. Samples from the cornea, aqueous humor and vitreous were collected after 2 and 4 weeks of treatment. Corneal tissue was thinly ground using a scalpel blade, homogenized in 0.5 mL of Tris-buffered saline (TBS) (pH 9.0, 0.5 M) for 10 s intervals with a PT10-35 GT Kinematica Polytron homogenizer (Kinematica AG) and then subjected to sonication (10 s, 0.5 kW) with a Qsonica XL-2000 sonicator (Qsonica LLC). For the aqueous humor samples, a volume of 100 μL was collected and mixed with 0.5 mL of TBS buffer. Vitreous samples were homogenized for 10 s and 0.5 ml of the mixture was taken before being combined with 0.5 mL of TBS buffer. Subsequently, 0.5 mL of n-butyl chloride was added to each of the cornea, aqueous humor and vitreous samples and incubated at 4 °C for 25 min to allow for mass transfer. Samples were centrifuged for 5 min at 12,000 rpm and the organic phase was collected. This extraction procedure was repeated three times. At the final stage, eluates were dried and reconstituted in 50 μL of dimethyl sulfoxide (DMSO). To verify the extraction efficiency of the method, aliquots of cornea, aqueous humor and vitreous samples were spiked with 50 ng/μL prior extraction. Successively, samples were prepared according to the protocol described above. Non-spiked tissues served as blanks (non-extracted samples). The recovery was analyzed by comparing the response of extracted and non-extracted samples (Fig. 4e). LC-MS/MS was used to quantify Ro5-3335 on each sample. Calibration curves of Ro5-3335 were prepared by serial dilutions in DMSO to produce concentrations between the range of 0.1 to 50 pg/μL. The compounds were separated on an Allure PFPP column (2.1 × 150 mm, 5 μm) using a 6460 LC-MS/MS system (Agilent, Santa Clara, CA). The mobile phase used was 0.1% formic acid in acetonitrile, and the sample injection volume was 10 μL. A constant flow rate of 0.5 mL/min was utilized with a gradient elution method over 10 min to operate the liquid chromatograph. The mass spectrometer portion of the LC-MS/MS was set at a temperature of 350 °C with a gas flow of 12 L/min, sheath gas temperature 400 °C, sheath gas flow 12 L/ min, capillary voltage 3.5 kV and nebulizer 35 psi. The electrospray ionization positive ion mode provided the maximum ionization and optimized retention time of the Ro5-3335 (5.17 min).

Statistical analyses.
Data are shown as mean ± SEM or median. For statistical analysis, two-tailed unpaired