BRD4 Inhibition Mitigates Acute and Chronic Corneal Injury Following Topical Nitrogen Mustard Exposure

Abstract

Corneal chemical injury, including mustard vesicant injury, remains a clinical challenge with limited therapeutic options. To enable effective mechanistic and therapeutic studies, we developed a mouse model of nitrogen mustard (NM)-induced corneal injury using a 3D-molded, cap-shaped filter paper conforming to the corneal surface to generate reproducible injuries. Multimodal in vivo imaging and histological assessment revealed an injury course characterized by acute epithelial disruption, stromal edema, and inflammation, followed by transient partial recovery and chronic stromal remodeling with persistent endothelial loss. After NM exposure, bromodomain (BD)-containing protein 4 (BRD4) expression and activity were upregulated. Short-term topical BRD4 inhibition with JQ1 reduced acute corneal inflammation (e.g., reducing CD45⁺ cells by 86%) and oxidative stress, and conferred long-term preservation of corneal clarity, stromal organization, and endothelial integrity. Selective inhibition of BD1 or BD2 recapitulated JQ1’s effects, indicating that blockade of either bromodomain is sufficient. BRD4 inhibition also significantly attenuated NM-induced corneal pathology in a rabbit model, supporting cross-species efficacy and translational relevance. Mechanistically, BRD4 inhibition selectively counteracted key NM-driven pathogenic programs, including inflammation, oxidative stress, and extracellular matrix remodeling. These findings identify BRD4 as a central epigenetic driver of vesicant-induced corneal injury and position BRD4 inhibition as a promising translational therapeutic strategy.

Graphical Abstract

Introduction

Chemical injury to the cornea remains a major unmet clinical challenge, frequently leading to acute epithelial damage, inflammation, and corneal remodeling, often progressing to chronic scarring and long-term visual impairment. Among the diverse classes of ocular toxicants, alkylating vesicants such as nitrogen mustard (NM) and sulfur mustard (SM) have been studied extensively because of the severity of the respiratory, cutaneous, and ocular injuries they produce, which are often accompanied by painful and prolonged recovery requiring extended treatment and rehabilitation. Historical analyses of chemical warfare exposures, particularly during World War I, estimate that SM accounted for approximately 80% of chemical casualties, establishing mustard agents as some of the most clinically consequential chemical threats encountered in modern history.^1,2 Due to their availability in stockpiles and ease of production, mustard vesicants also pose a potential threat to civilians as they could be used in terrorist attacks.

Although mustard vesicants injure multiple organ systems, the eye is disproportionately affected, with 90% or more of exposed individuals reported to develop acute corneal injury marked by a robust inflammatory response, disruption of the corneal epithelium, edema, dilation of limbal vasculature, and the early onset of neovascularization (NV).^2–5 While many patients recover following mild or moderate exposures, a subset develops delayed pathology including persistent epithelial defects, chronic inflammation, recurrent erosions, stromal fibrosis, and progressive NV, ultimately compromising vision.^2–4 Because there is currently no specific therapy proven to halt or reverse the ocular effects of SM exposure,³ mustard vesicants provide a stringent and clinically relevant framework for evaluating targeted molecular interventions aimed at treating chemical corneal injury and preventing long-term vision-threatening sequelae.

Considering that delayed corneal pathology reflects sustained dysregulation of injury-response pathways, identifying conserved molecular drivers of injury progression is essential for the development of broadly effective therapies for chemical corneal burns. Epigenetic regulators that govern inflammatory and wound-healing transcriptional programs provide a compelling mechanistic framework for targeted intervention. Histone acetylation is a key epigenetic mechanism regulating gene transcription in diseases.⁶ Following histone acetylation, acetyl-lysine (KAc) residues within N-terminal histone tail are recognized by bromodomains, which bind KAc and recruit transcriptional machinery. The bromodomain and extra-terminal domain (BET) family proteins are such epigenetic readers and act as gatekeepers of gene expression across diverse pathophysiological contexts.⁶ The BET family consists of bromodomain-containing proteins (BRDs), including BRD2, BRD3, BRD4, and the testis-restricted BRDT.^6,7 Among these, BRD4 is the most extensively studied because it is ubiquitously expressed and, unlike BRD2 and BRD3, contains a unique C-terminal domain that recruits positive transcription elongation factor b (P-TEFb), enabling regulation of transcriptional elongation in addition to initiation.^6–8 This functional distinction positions BRD4 at super-enhancers, where it coordinates high-level expression of genes governing diverse cellular responses.⁶ BRD4 has been reported to associate with acetylated RelA and facilitate NF-κB-responsive transcription of inflammatory cytokines through P-TEFb recruitment,⁶ and to participate in TGFβ/SMAD-driven fibrotic responses.^9–12 In addition, BRD4 can regulate a broad range of cellular processes, including inflammation, cell death, proliferation, stress responses, and tissue remodeling, by supporting transcriptional programs controlled by growth factor- and cytokine-responsive regulators such as c-MYC, STAT family members, TP53, and FOXO family transcription factors.^6,13–16

Aberrant BRD4 activity has been implicated in a wide range of acute and chronic diseases, as well as in key pathological processes, including viral infection, sepsis, airway inflammation, cardiovascular and renal injury, neurodegeneration, fibrosis, and NV.^6,17–27 In the eye, only a few studies have explored the therapeutic potential of BET inhibition. BET inhibitors have been shown to ameliorate NMDA-induced retinal ganglion cell death,²⁸ prevent photoreceptor degeneration in models of inherited retinal degeneration,²⁹ suppress retinal inflammation in experimental autoimmune uveitis,³⁰ inhibit retinoblastoma cell growth,³¹ reduce lens fibrosis,³² and attenuate TGFβ-induced corneal myofibroblast differentiation while suppressing mechanical injury-induced corneal scarring.³³ Despite these findings, the role of BRD4 in corneal chemical burns remains unknown.

In this study, we established a reproducible NM-induced corneal injury model in mice and extended this platform to rabbits to enable evaluation across small and large animal systems with clinical relevance. Using longitudinal imaging and histological assessment, we delineated the progression of acute and chronic pathological changes following NM exposure and found that BRD4 expression and activity were markedly upregulated after NM injury. Importantly, we demonstrated that topical pharmacologic BRD4 inhibition significantly mitigated corneal pathology in both species by suppressing inflammatory, oxidative stress-related, apoptotic, and extracellular matrix (ECM)-remodeling pathways, thereby providing a strong rationale for further translational development of BRD4-based therapies for chemical corneal injury.

Results

Development and validation of a localized NM corneal injury model using filter paper caps generated with a 3D-printed mold

NM, an analog of SM, induces similar corneal injury and presents a comparable potential exposure hazard; however, it is commercially available and easier to obtain and handle in a laboratory setting. Consequently, it has been used by many laboratories to study vesicant-induced ocular injuries as an alternative to SM.^4,34–40 To create a reproducible and localized NM-induced corneal injury model in mice, we reshaped the commonly used filter paper into a cap-like configuration using 3D-printed molds (Fig. S1A). The radius of curvature of the mold was set to 1.5 mm, matching the central corneal curvature of adult C57BL/6 mice.⁴¹ The resulting cap-shaped filter paper conformed closely to the corneal surface, allowing localized and uniform chemical exposure. To evaluate this design, we compared cap-shaped and flat filter paper using an alkali burn model as a visual readout to assess corneal contact and injury reproducibility. Flat filter paper failed to conform tightly to the corneal surface, producing irregular and poorly defined contact areas, whereas cap-shaped filter paper generated a circular and well-confined exposure zone (Fig. S1B). The cap-shaped filter paper thus consistently produced sharply demarcated circular injuries with reproducible pathological outcomes across animals (Fig. S2).

Using the cap-shaped filter paper model, we next optimized NM concentration and exposure duration to achieve consistent corneal injury without excessive complications. Corneas were treated with 1%, 2%, or 4% NM for 1, 3, or 5 minutes. Based on longitudinal monitoring from the acute phase (day 3) to the chronic phase (day 84), injury phenotypes were classified as mild (punctate erosions with full recovery), moderate (stromal edema evolving into scarring and thinning), or severe (infection, corneal melting or perforation) (Fig. S3A). Injury severity exhibited clear dose and time dependence (Fig. S3B). At 1% NM, 1- and 3-minute exposures caused negligible injury, whereas the 5-minute exposure induced moderate damage in half of the mice. At 2% NM, a 1-minute exposure caused mild injury, 3 minutes produced moderate injury in a subset of animals, and 5 minutes yielded reproducible moderate injury in most mice. At 4% NM, shorter exposures resulted in mild-to-moderate injury, while the 5-minute exposure led to uniformly severe injury, characterized by corneal melting or perforation that necessitated early termination. Based on these findings, 2% NM for 5 minutes was identified as the optimal condition, producing consistent and appropriately robust injury while minimizing complications. This condition was therefore selected for subsequent experiments to study the progression of NM-induced corneal injury.

Longitudinal in vivo imaging reveals temporal progression of NM-induced corneal injury

Using slit-lamp and anterior segment optical coherence tomography (AS-OCT), we assessed the temporal progression of corneal injury in detail following standardized exposure to 2% NM for 5 minutes (Fig. 1A–D). During the acute phase (within 7 days), corneal changes evolved dynamically. Marked edema and reduced transparency were most prominent during the early acute period (within 3 days after injury) (upper panels of Fig. 1B and Fig. 1C), with AS-OCT revealing a significant increase in central corneal thickness (CCT) (lower panels of Fig. 1B and Fig. 1D). Over the following days, corneal edema gradually subsided, and corneal transparency subsequently improved. By day 14, although corneal transparency was not fully restored, CCT returned to baseline, marking a temporary “healing” stage. During the chronic phase (days 14–84), corneal thickness progressively decreased (lower panels of Fig. 1B and Fig. 1D), and slit-lamp imaging demonstrated persistent corneal opacities and stromal scarring, indicating chronic fibrosis and incomplete structural recovery (upper panels of Fig. 1B and Fig. 1C).

Figure 1. Longitudinal assessment of NM-induced corneal injury using in vivo multimodal imaging.

(A) Study design and imaging timeline. Baseline (“pre-injury”) images were acquired 5 days prior to injury. Corneal injury was induced by applying 2% NM for 5 minutes at day 0, and eyes were monitored at days 1, 3, 7, 14, 21, 42, and 84 using non-invasive multimodal imaging, including slit-lamp photography, anterior-segment optical coherence tomography (AS-OCT), and in vivo confocal microscopy (IVCM). Tissues were collected at day 3 or day 84. (B) Representative slit-lamp and AS-OCT images of the cornea across time. (C) Corneal opacity scores over time. Opacity was graded as follows: 0, no corneal haze; 1, iris details visible; 2, pupillary margin visible but iris details not visible; 3, pupillary margin not visible; 4, cornea totally opaque. n = 7–12 per time point, except n = 4 at day 1. (D) Central corneal thickness measured by AS-OCT relative to the pre-injury baseline. n = 4–9 per time point, except n = 3 at day 1. (E) Representative IVCM images of the indicated corneal layers at each time point. Arrows (→) denote examples of infiltrating inflammatory cells. Hash marks (#) denote examples of areas of partial epithelial healing. Triangles (▲) denote examples of endothelial cell-deficient areas. *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001. Epi, epithelium; Str, stroma; Endo, endothelium.

In parallel, high-resolution in vivo confocal microscopy (IVCM) provided layer-specific details of the cornea across the early (acute), intermediate (healing), and chronic phases of NM injury (Fig. 1E). Within the epithelium, basal epithelial cell density declined between days 3 and 7 post-injury, accompanied by a transition from regular to irregular cell morphology. Partial recovery was observed between days 14 and 21 post-injury, although cell borders remained indistinct throughout the chronic phase (days 14–84).

Stromal morphology evolved over time and varied with depth, following a sequence of inflammation, cell loss, repair, and fibrotic remodeling. At day 3 post-injury, the anterior stroma exhibited disrupted lamellar organization with reduced keratocyte density and evidence of cellular fragmentation and dissolution. The posterior stroma was dominated by densely infiltrating inflammatory cells, whose hyper-reflective signals obscured visualization of the underlying lamellae. By day 7, stromal cells transitioned from stellate to irregular or elongated forms with visible intercellular connections, indicative of cell death or keratocyte activation and differentiation into fibroblasts/myofibroblasts, accompanied by decreased anterior stromal cell density. After day 14, normal stromal lamellae were gradually replaced by diffuse hyper-reflective regions suggestive of progressive fibrosis, which intensified toward the end of the observation period.

The endothelium also exhibited pronounced alterations following NM exposure. In normal corneas, endothelial cells form a regular hexagonal mosaic pattern. By day 3 post-injury, extensive endothelial cell loss and cell-deficient regions were evident. At day 7, neighboring cells migrated into denuded areas, partially restoring coverage; however, the presence of guttata-like excrescences indicated ongoing stress and incomplete repair. After day 14, endothelial cell density continued to decline, reflecting both migration-driven redistribution and progressive cell loss. By day 84, although most cell-deficient regions were repopulated, areas of markedly reduced density or persistent cell loss remained. In addition, cell morphology was altered, with indistinct borders and scattered guttata, consistent with persistent endothelial injury.

In addition to en face imaging, section-like side views were acquired by IVCM to visualize the layered corneal architecture. In NM-injured corneas, the acute and intermediate phases were marked by epithelial disruption and inflammatory cell infiltration across multiple layers, followed by the gradual emergence of hyper-reflective bands in the chronic phase. These changes were consistent with, but more intuitively appreciated than, en face observations.

NM exposure induces corneal inflammation and endothelial damage

To validate the live imaging observations of early-phase inflammation in the cornea, we performed immunofluorescent staining to assess immune cell infiltration (CD45⁺) and T-cell involvement (CD3⁺) in corneal samples collected at 3 days post-NM injury. We observed extensive infiltration of CD45⁺ and CD3⁺ immune cells in both central and limbal regions, confirming inflammatory activation (Fig. 2A–B).

Figure 2. NM exposure induces inflammatory cell infiltration and severe endothelial damage in the cornea.

Corneal injury was induced by applying 2% NM for 5 minutes. (A, B) Representative immunofluorescence images showing CD45⁺ leukocytes (red) and CD3⁺ T cells (green) in the central (A) and limbal (B) cornea at day 3 after NM injury. Nuclei were counterstained with DAPI (blue). Graphs show quantification of infiltrating CD45⁺ and CD3⁺ cells. (C-H) Whole-mount ZO-1 staining (green) of the central corneal endothelium in Sham and NM-injured eyes at day 84. (C) Representative immunofluorescence images of ZO-1. (D) Endothelial cell morphometric reconstructions showing alterations in endothelial cell size, shape, and mosaic organization. (E) Quantification of endothelial cell density. (F) Distribution of endothelial polygonal cell types. (G) Density curves showing the distribution of endothelial cell sizes. (H) Summary of endothelial morphometric parameters, including endothelial cell density (ECD), coefficient of variation (CV), hexagonality (HEX), maximum and minimum cell sizes (S_max, S_min), average cell size (S_avg), and standard deviation of cell size (S_sd). Scale bar, 50 μm. n = 4; *p < 0.05, **p < 0.01, and ****p < 0.0001. Epi, epithelium; Str, stroma; Endo, endothelium; Cj, conjunctiva.

Given that persistent corneal endothelial dysfunction is a critical determinant of long-term visual outcomes after chemical injury,⁴² we next investigated whether NM exposure results in sustained endothelial alterations. ZO-1-stained corneal whole mounts were examined at 84 days post-injury. We observed that the central corneas of control eyes displayed a uniform, tightly packed hexagonal endothelial mosaic with intact intercellular junctions, whereas NM-injured corneas exhibited marked disruption of cell borders, endothelial cell polymegathism and pleomorphism, and extensive cell loss (Fig. 2C). Morphometric analysis demonstrated profound endothelial structural alterations in NM corneas (Fig. 2D–H), including a marked reduction in cell density (Fig. 2E), loss of hexagonal predominance with a dispersed distribution of cell apices (Fig. 2F), and a flattening and broadening of cell size distribution, indicating increased size heterogeneity (Fig. 2G). These abnormalities were accompanied by increased coefficients of variation, reduced hexagonality, and elevated cell size metrics, consistent with severe and persistent endothelial damage following NM injury (Fig. 2H).

BRD4 is upregulated and activated during the early phase of NM-induced corneal injury, and local inhibition suppresses its activity

Our imaging studies characterized the temporal and structural features of NM-induced corneal injury, revealing acute inflammation, progressive stromal remodeling, and endothelial loss. To investigate potential mechanisms underlying NM-induced corneal injury, we reanalyzed RNA-seq data from SM-treated rabbit corneas (GSE127708, NCBI/GEO) deposited by Horwitz et al.⁴³ Given the similarities between SM- and NM-induced injury, we reasoned that this approach could provide preliminary insights into the molecular pathways involved. We compared corneas with delayed injury (clinically impaired) to naïve controls to identify SM-responsive transcriptional pathways. Differential expression analysis was performed with DESeq2 (|log₂FC| ≥ 1, p < 0.05). Clustering analysis of differentially expressed genes (DEGs) clearly separated naïve controls from SM-injured corneas, indicating profound injury-associated transcriptional alteration (Fig. S4A). In total, SM injury resulted in 981 significantly upregulated genes and 383 downregulated genes (Fig. S4B). Gene Ontology (GO) enrichment analysis revealed that SM-induced transcriptional changes were dominated by activation of inflammatory and immune responses, ECM remodeling and tissue repair, cell signaling and mediator production, as well as oxidative stress and cell death pathways, while genes involved in ion transport were preferentially repressed (Fig. S4C). Motif enrichment analysis (RcisTarget) of promoter regions identified enrichment for motifs associated with several transcription factors, including NF-κB, STATs (Stat1, Stat2, Stat6), IRFs (Irf1–5, Irf7–9), Jun, Fos, and Smad, likely driving these changes. The simultaneous upregulation of around 1,000 genes under the control of multiple transcription factors highlights the complexity of mustard-induced corneal injury and suggests that targeting single pathways may be insufficient for therapy.

Epigenetic regulators are known to shape transcriptional programs in various injury models through interactions with multiple transcription factors. Among these, BET family proteins are key gatekeepers of gene expression across diverse pathophysiological contexts,⁶ with BRD4 being the most extensively studied member. We therefore investigated whether BRD4 participates in the pathophysiological response to NM-induced corneal injury. Immunohistochemical analysis of corneal sections collected at 3 days post-injury revealed markedly increased BRD4 expression throughout the cornea (Fig. 3A). Elevated BRD4 immunoreactivity was observed in the epithelium, stroma, and endothelium, suggesting broad engagement of this factor during the early phase of injury. To further assess whether upregulated BRD4 was also functionally activated, we examined histone H3K122 acetylation (H3K122ac), a well-established readout of BRD4 histone acetyltransferase activity.²⁰ BRD4 uniquely acetylates H3 at Lys122, destabilizing nucleosomes and promoting more efficient transcriptional elongation.⁴⁴ Corneal sections from NM-injured eyes exhibited a parallel increase in H3K122ac staining at day 3, indicating an increase in BRD4 activity after NM injury (Fig. 3B).

Figure 3. BRD4 is upregulated and activated after NM exposure, and topical JQ1 suppresses its activity.

(A, B) Corneal sections collected at 3 days after NM injury were stained with antibodies against BRD4 (A) or acetylated histone H3K122 (H3K122ac) (B). Representative immunofluorescence images are shown; nuclei were counterstained with DAPI. Bar graphs show relative mean fluorescence intensity (MFI) normalized to Sham. (C) Study design and imaging timeline. JQ1 treatment was initiated 2 hours after NM exposure and administered topically five times daily for the first 3 days in the acute studies or for 14 days in the chronic studies. (D) Effect of local BRD4 inhibition. Corneas collected 3 days after injury from sham or NM-exposed eyes treated with vehicle (Veh) or JQ1 were stained for H3K122ac. Representative images and quantification (relative MFI) are shown. Scale bar, 50 μm. n = 4–5; *p < 0.05, **p < 0.01 and ***p < 0.001. Epi, epithelium; Str, stroma; Endo, endothelium.

BRD4 inhibition mitigates acute corneal injury after NM exposure

Inflammation is a main pathological feature during the acute phase of corneal injury after NM exposure. To determine whether BRD4 plays a role in NM-induced corneal inflammation, we tested whether pharmacologic inhibition of BRD4 activity with JQ1, a selective BET inhibitor that competitively blocks acetyl-lysine binding to bromodomain 1 (BD1) and 2 (BD2), could modulate corneal inflammatory responses to NM injury. Topical JQ1 treatment (1 mM), initiated 2 hours after NM exposure and administered five times daily for 3 days, markedly reduced H3K122 acetylation at day 3, indicating effective inhibition of BRD4 activity (Fig. 3C–D).

We next examined immune cell accumulation in the cornea 3 days after NM exposure in the presence or absence of BRD4 blockade. Immunofluorescence staining revealed substantial infiltration of CD45⁺ leukocytes and CD3⁺ T cells in both the central and limbal cornea following NM injury (Fig. 4). NM-injured, vehicle-treated eyes exhibited dense stromal accumulation of CD45⁺ cells together with a pronounced increase in CD3⁺ T cells. In contrast, JQ1 treatment markedly diminished CD45⁺ leukocyte infiltration by 86% and profoundly suppressed CD3⁺ T-cell presence by 96% in the central cornea (Fig. 4A). A similar reduction was observed in the limbal region, where JQ1 markedly blunted the infiltration of both CD45⁺ and CD3⁺ cells (Fig. 4B). Interestingly, the magnitude of suppression differed between immune cell types: while JQ1 significantly decreased total CD45⁺ leukocytes, its effect on CD3⁺ T cells was substantially greater. Consequently, both central and limbal corneas exhibited a markedly reduced CD3⁺/CD45⁺ ratio compared with vehicle-treated NM-injured groups (Fig. S5). Together, these findings demonstrate that early BRD4 inhibition effectively attenuates the acute inflammatory response to NM injury, with a particularly strong suppressive effect on T-cell recruitment. Consistently, proinflammatory cytokines and chemokines that play a critical role in this process, including Il1b, Cxcl1, Cxcl2, and Cxcl10, were markedly increased in NM-treated corneas, and their expression was attenuated by JQ1 treatment (Fig. S6).

Figure 4. JQ1 alleviates NM-induced inflammatory responses during the acute phase.

Corneal injury was induced by applying 2% NM for 5 minutes. JQ1 treatment was initiated 2 hours after NM exposure and administered topically five times daily for 3 days. Corneas were collected at day 3 after NM injury. (A, B) Representative immunofluorescence images showing CD45⁺ leukocytes (red) and CD3⁺ T cells (green) in the central cornea (A) and limbal region (B). Nuclei were counterstained with DAPI (blue). Graphs show quantification of infiltrating CD45⁺ and CD3⁺ cells. Scale bar, 50 μm. n = 4; *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001. Epi, epithelium; Str, stroma; Endo, endothelium; Cj, conjunctiva.

BRD4 inhibition selectively reverses NM-induced transcriptomic changes

To gain deep insight into the molecular mechanisms by which BRD4 inhibition attenuated corneal inflammation, we performed bulk RNA sequencing of central corneas collected at day 3 post-injury from four experimental groups (Sham + Vehicle, Sham + JQ1, NM + Vehicle, and NM + JQ1). All 24 samples were processed in a single batch to minimize technical variability. Principal component analysis (PCA) demonstrated that the two sham groups clustered closely, indicating minimal transcriptional impact of JQ1 on uninjured corneas, whereas NM-injured samples separated distinctly from shams (Fig. S7A). JQ1 treatment shifted NM-injured transcriptional profiles, suggesting broad modulatory effects on injury-induced gene expression (Fig. S7A). Differential expression analysis revealed that NM exposure elicited broad transcriptional alterations, with 1,666 genes upregulated and 899 downregulated relative to sham control (Fig. S7B). To evaluate the effect of BRD4 inhibition on NM-induced DEGs, we compared NM + JQ1 with NM + Vehicle samples and found that 784 NM-altered genes (30.6%) exhibited partial reversal following JQ1 treatment, including 702 recovering from NM-induced upregulation and 82 recovering from NM-induced downregulation (Fig. S7C–D). Conversely, a substantial portion of NM-responsive genes remained unchanged, indicating that the early transcriptional response to NM comprises both JQ1-sensitive and JQ1-resistant components.

To place these gene-level changes in a biological context, we next performed GO enrichment analysis on NM-responsive DEGs (NM + Vehicle vs. Sham + Vehicle) and JQ1-regulated genes (NM + JQ1 vs. NM + Vehicle) (Fig. 5A–B). NM injury strongly enriched inflammatory and immune pathways, including leukocyte migration, chemotaxis, lymphocyte proliferation, T cell activation, and cytokine production (Fig. 5A). Consistent with this, NM induced pro-inflammatory and immune-related genes such as Cxcl5, Cxcl12, Il1a, Il1b, C3, C4b, Ifi44, and Cd81 (Fig. S8). NM also upregulated wound-healing and ECM-remodeling pathways, including extracellular matrix organization, connective tissue development, and fibroblast proliferation, accompanied by increased expression of Col2a1, Col3a1, Col18a1, Sparc, Acta2, Mmp9, Mmp13, and Tgfb3 (Fig. 5A, Fig. S8). Angiogenesis-related pathways were transcriptionally enriched, with induction of Vegfa, Pdgfb, Flt1, and Mcam. In addition, NM activated oxidative stress and apoptotic pathways, including reactive oxygen species (ROS) metabolism and responses to hypoxia and oxidative stress (Fig. 5A), with elevated expression of Gpx2, Gpx3, Gpx7, Alox5, Alox12b, and Alox15 (Fig. S8). In contrast, metabolic pathways related to fatty acid and small molecule catabolism were suppressed (Fig. 5A). GO analysis of JQ1-regulated genes revealed broad attenuation of NM-induced pathological programs (Fig. 5B). JQ1 reduced enrichment of inflammatory and immune pathways, ECM-remodeling pathways, and oxidative stress-related pathways, with corresponding suppression of key inflammatory and remodeling genes (Fig. S8). Notably, JQ1 uniquely enriched pathways involved in tight junction assembly, suggesting potential restoration of epithelial barrier integrity. Although angiogenesis-related pathways were transcriptionally enriched, pathological neovascularization was not observed in the mouse cornea after NM exposure, possibly because transcriptional activation alone is insufficient to overcome the cornea’s intrinsic anti-angiogenic barriers in this model.

Figure 5. Pathway-level regulation and functional enrichment analysis of NM-induced corneal injury and JQ1-mediated response.

(A, B) GO enrichment analysis of DEGs in NM + Vehicle versus Sham + Vehicle corneas (A) and DEGs in NM + JQ1 versus NM + Vehicle corneas (B). Bar plots display significantly enriched GO terms, with colors indicating whether the enrichment was derived from upregulated (red) or downregulated (blue) DEGs following NM injury and JQ1 treatment. (C) GSVA-based pathway activity. GSVA scores represent pathway activity changes relative to the Sham + Vehicle group. Pink and green dots indicate [(NM + Vehicle) − (Sham + Vehicle)] and [(NM + JQ1) − (Sham + Vehicle)], respectively. Percentages denote relative changes in pathway activity. Pathways related to inflammation and immune response, ECM remodeling and repair, angiogenesis, and oxidative stress are shown. (D) Protein-protein interaction (PPI) network generated from DEGs in the NM + Vehicle versus Sham + Vehicle comparison. Nodes are grouped into functional modules by community clustering, including inflammation and immune response, ECM remodeling and angiogenesis, oxidative stress and lipid metabolism, cell cycle, interferon signaling, and nucleotide and glucose metabolism. Hub genes within each module are distinguished by larger node sizes. n = 6.

To further quantify pathway-level dynamics across all groups, we performed Gene Set Variation Analysis (GSVA). NM injury caused a broad displacement of pathway activity distributions relative to sham corneas, whereas JQ1 treatment shifted these distributions toward baseline levels (Fig. S7E). Consistent with gene-level analyses, pathway response classification revealed that among the 4,127 GO terms altered in the NM + Vehicle versus Sham + Vehicle comparison, 1,450 pathways (35.2%) exhibited reversal toward sham levels following JQ1 treatment, reflecting early and coordinated BRD4-dependent modulation of NM-responsive biological processes (Fig. S7F–G). GSVA scores further enabled quantitative assessment of the magnitude of pathway reversal induced by JQ1. Analysis of pathway activity scores across all samples identified the most significantly reversed GO terms following JQ1 treatment (Fig. 5C). Specifically, JQ1 markedly attenuated inflammatory pathways (including IL-6 receptor binding, IL-17 signaling, and activated T cell proliferation) as well as ECM disassembly processes, with reversal rates ranging from approximately 30% to greater than 100%. Collectively, these findings demonstrate that JQ1 exerts its therapeutic effects by coordinately suppressing the multidimensional pathological networks activated by NM-induced corneal injury.

Finally, to identify key network-level regulators underlying NM injury and JQ1 treatment, we mapped NM-induced DEGs onto the STRING database to construct a protein-protein interaction (PPI) network. Using the Louvain algorithm for community detection, we identified multiple densely connected gene clusters, each enriched for distinct biological functions, such as inflammatory responses, oxidative stress and lipid metabolism, ECM remodeling and angiogenesis, cell cycle regulation, and metabolic processes (Fig. 5D). To evaluate how BRD4 inhibition influenced these network structures, we overlaid the genes that were reversed by JQ1 onto the PPI network and highlighted their hub nodes (Fig. S9). This analysis revealed multiple central hub genes whose NM-induced dysregulation was selectively corrected by JQ1, indicating that BRD4 inhibition modulates specific regulatory modules rather than uniformly suppressing all NM-driven responses.

Together, these analyses show that NM injury activates a coordinated network of inflammatory, oxidative stress-related, and ECM-remodeling processes, and that BRD4 inhibition via JQ1 reshapes interconnected transcriptional networks to selectively counteract key NM-driven pathogenic programs during the acute phase.

BRD4 inhibition mitigates NM-induced ocular oxidative stress

To validate our transcriptomic findings indicating that BRD4 modulation alters oxidative stress responses, we performed in situ dihydroethidium (DHE) staining to assess superoxide accumulation during the acute phase of injury (day 3 post-NM exposure) (Fig. S10A, C). Topical NM exposure induced a marked increase in DHE fluorescence intensity within the central cornea compared with sham controls, indicating substantial superoxide accumulation. This response was markedly attenuated by topical JQ1 treatment. The specificity of the DHE signal was confirmed by superoxide dismutase (SOD) pre-incubation, which abolished the enhanced fluorescence. Consistently, immunofluorescence analysis of 8-hydroxy-2'-deoxyguanosine (8-OHdG), which is formed through oxidative modification of guanine bases in DNA and is widely used as a biomarker of oxidative stress, also revealed extensive immunoreactivity throughout the corneal tissue in the NM + Vehicle group (Fig. S10B, D), which was significantly reduced by JQ1 treatment. Together, these findings provide direct evidence that local BRD4 inhibition suppresses NM-induced oxidative stress during the acute phase of corneal injury.

Early BRD4 inhibition attenuates long-term corneal remodeling and improves wound healing after NM-induced injury

Encouraged by the broad inhibitory effects of JQ1 on multiple pathological pathways in NM-injured corneas, we further assessed whether early JQ1 treatment conferred functional consequences for tissue protection. We assessed the effects of BRD4 inhibition on corneal structure and opacity after NM exposure and further examined whether this early intervention provided sustained protection throughout the chronic phase of injury. NM-injured eyes were treated with JQ1 or vehicle five times daily for 14 days after the injury, and they were monitored by longitudinal live imaging through day 84 post-injury. Slit-lamp imaging demonstrated that JQ1-treated eyes maintained greater corneal transparency, with fewer focal opacities than vehicle-treated NM-injured corneas (Fig. 6A, 6C). AS-OCT measurements revealed significant attenuation of edema in the acute phase and greater structural preservation of the stroma in the chronic phase, as indicated by preservation of central corneal thickness and more uniform stromal reflectivity (Fig. 6B, 6D). Of note, JQ1 treatment produced no detrimental effects in uninjured corneas, as evidenced by comparable transparency and central corneal thickness in JQ1-treated and vehicle-treated sham corneas (Fig. S11).

Figure 6. Early BRD4 inhibition provides acute and long-term protection against NM-induced corneal injury.

Corneal injury was induced by 2% NM for 5 minutes at day 0. JQ1 was initiated 2 hours after NM exposure and applied five times daily for the first 14 days. (A) Representative slit-lamp photographs at the indicated time points from NM-injured eyes treated topically with vehicle or JQ1. (B) Representative AS-OCT B-scans across time. (C) Corneal opacity scores over time. n = 9–11. (D) Central corneal thickness measured by AS-OCT. n = 6–8. (E) IVCM images across corneal layers at day 84. Epi, epithelium; Str, stroma; Endo, endothelium. *p < 0.05, **p < 0.01, and ***p < 0.001.

IVCM imaging at 84 days post-NM injury demonstrated that JQ1 treatment partially preserved corneal microstructural organization across multiple layers (Fig. 6E). In the basal epithelium, JQ1-treated corneas exhibited a more uniform reflectivity pattern with fewer irregular hyper-reflective areas. Both the anterior and posterior stroma displayed reduced hyper-reflective fibrotic bands and preserved keratocyte morphology compared with NM + Vehicle corneas, suggesting attenuation of stromal remodeling. In the endothelium, JQ1-treated corneas showed a more continuous cellular pattern with fewer large acellular or highly disorganized regions relative to NM + Vehicle corneas. Side-view imaging further revealed improved continuity and stratification of corneal layers in JQ1-treated corneas, consistent with enhanced overall corneal structural integrity.

Complementary histological analysis at 84 days post-NM injury further supported these findings. Corneal flat mounts showed that early JQ1 treatment significantly attenuated the NM-induced accumulation of CD45⁺ cells with dendritic morphology in both the central and peripheral epithelium, indicating alleviation of chronic inflammatory responses (Fig. 7A). Corneal whole mounts and tissue sections revealed decreased stromal cell loss (Fig. 7B) and reduced fibrotic remodeling, as evidenced by reduced Col1a1 expression in JQ1-treated NM-injured corneas (Fig. 7C), demonstrating that early BRD4 inhibition effectively limits ongoing stromal degeneration and fibrosis. Moreover, endothelial analyses at the chronic stage showed that early JQ1 treatment partially rescued NM-induced endothelial damage, as reflected by improved endothelial continuity and organization of intercellular junctions, increased endothelial cell density, and a more normalized cell size distribution (Fig. 7D). These changes were accompanied by partial restoration of endothelial morphometric parameters (Fig. S12), supporting a sustained protective effect of early BRD4 inhibition on corneal endothelial integrity.

Figure 7. Early JQ1 treatment mitigates chronic epithelial immune activation and preserves stromal and endothelial integrity after NM injury.

Corneal injury was induced by applying 2% NM for 5 minutes. JQ1 treatment was initiated 2 hours after NM exposure and administered topically five times daily for the first 14 days. Corneas were collected at day 84 after NM injury. (A) Corneal whole mounts were stained for CD45⁺ cells (red) and nuclei (DAPI, blue), and images were acquired in the central and peripheral cornea. Bar graphs represent the quantification of CD45⁺ cell density in each region. n = 6–9. (B) Stromal cell density was assessed by DAPI staining. n = 4–5. (C) Representative corneal sections stained for Col1a1 (red) and DAPI (blue) illustrating chronic stromal fibrosis at day 84, and bar graph represents the quantification of Col1a1 fluorescence intensity. n = 5. (D) Corneal flat mounts were stained for ZO-1 (green) and DAPI (blue), with polygon/Voronoi reconstructions illustrating endothelial cell morphology. Bar graphs represent the quantification of endothelial cell density and endothelial cell-size distribution. n = 5–7. Scale bars: 50 μm. *p < 0.05, ***p < 0.001 and ****p < 0.0001.

Collectively, these results indicate that early topical BRD4 inhibition not only mitigates acute corneal injury but also confers long-term structural protection, reducing persistent inflammation, chronic fibrosis, and endothelial loss following NM exposure.

Selective BD1 and BD2 inhibition recapitulates JQ1-mediated protection following NM exposure

BET family members, including BRD4, contain two tandem bromodomains (BD1 and BD2) that recognize KAc residues on histones and transcription factors, with distinct binding preferences.^6,7,45 JQ1 is a pan-BET inhibitor with comparable affinity for BD1 and BD2 and therefore does not discriminate between the functional contributions of these domains.⁴⁶ It has been shown that steady-state gene expression primarily relies on BD1, whereas inflammatory gene induction requires both BD1 and BD2.⁴⁶ Consequently, BD1 inhibition may interfere with tissue homeostasis, while BD2 inhibition could more selectively target pathological processes. Therefore, we further investigated the effects of novel selective inhibitors for BD1 (GSK778) and BD2 (GSK620) ⁴⁶ on NM-induced corneal injury.

At 3 days post-injury, slit-lamp imaging showed that both GSK778 and GSK620 substantially reduced acute corneal haze compared with that in vehicle-treated NM controls (Fig. S13A). Quantitative opacity scores confirmed significant reductions in corneal opacity with either BD1 or BD2 inhibition, demonstrating that selective blockade of either domain was sufficient to mitigate early tissue damage. By day 84, long-term imaging revealed that all three inhibitors (JQ1, GSK778, and GSK620) maintained improved corneal transparency relative to vehicle-treated NM eyes, with minimal residual focal opacities. Structural assessment by AS-OCT further supported these findings (Fig. S13B). At day 3, vehicle-treated NM corneas displayed pronounced stromal thickening and irregularity, whereas treatment with JQ1, GSK778, or GSK620 markedly reduced corneal thickening and improved stromal reflectivity. Quantitative CCT measurements confirmed significant reductions in edema across all three inhibitor groups. By day 84, corneas treated with BD1- or BD2-selective inhibitors exhibited long-term preservation of CCT comparable to JQ1-treated eyes and markedly superior to vehicle-treated NM controls.

Together, these results demonstrate that selective inhibition of either BD1 or BD2 confers substantial protection against both acute and chronic NM-induced corneal damage. Both BD1- and BD2-selective inhibitors achieved long-term structural and optical preservation comparable to JQ1, suggesting that both bromodomains contribute to NM-driven pathology and that targeted inhibition of either domain is sufficient to provide sustained therapeutic protection.

Post-exposure BRD4 inhibition mitigates corneal pathology after NM injury in rabbits

To enhance translational relevance, we further evaluated the therapeutic efficacy of BRD4 inhibition in a rabbit model of NM-induced corneal injury. The rabbit cornea closely approximates the human cornea in size, thickness, stromal organization, and epithelial turnover, and its wound-healing and inflammatory responses are highly predictive of clinical outcomes. Consequently, this model is widely used for assessing ocular drug delivery, dosing regimens, safety, and therapeutic efficacy in a clinically meaningful context.

In this rabbit model, central corneal exposure to 2% NM induced a robust and progressive injury characterized by severe acute corneal edema, disruption of the epithelial-stromal interface, loss of corneal transparency, and subsequent neovascularization (Fig. 8). Slit-lamp imaging revealed that corneal edema was most pronounced during the acute phase, particularly within the first two weeks following injury, with partial resolution over time. However, despite a reduction in edema, corneal opacity persisted, and progressive NV became evident during the early chronic phase and continued into the chronic phase, indicating sustained pathological remodeling and angiogenic activation (Fig. 8A–C). AS-OCT imaging corroborated these observations, demonstrating marked corneal thickening during the acute phase, consistent with stromal edema, followed by gradual thinning at later time points as edema resolved. Additionally, vehicle-treated NM corneas exhibited persistent disruption of the epithelial-stromal interface at all time points, indicating incomplete structural recovery despite partial normalization of corneal thickness (Fig. 8D–E).

Figure 8. BRD4 inhibition mitigates NM-induced corneal injury in rabbit.

Corneal injury in rabbits was induced by 2% NM for 5 minutes at day 0. JQ1 or vehicle was initiated 2 hours after NM exposure and applied topically five times daily for the first 14 days. (A) Representative slit-lamp photographs at the indicated time points from NM-injured eyes treated with vehicle or JQ1. Yellow arrowheads denote corneal neovascularization. (B, C) Quantification of corneal opacity and neovascularization area over time. (D) Representative AS-OCT B-scans across time. Yellow arrows point to the epithelial-stromal interface, appearing as a hypo-reflective band in sham corneas. This boundary was lost following NM injury across all time points but was restored progressively in the JQ1-treatment group over time. (E) Central corneal thickness measured by AS-OCT. n = 4–5; *p < 0.05, **p < 0.01 and ***p < 0.001.

Topical treatment with the BRD4 inhibitor JQ1, initiated 2 hours after NM exposure and administered five times daily for 14 days, conferred significant protection against NM-induced corneal injury. JQ1-treated corneas exhibited markedly improved transparency throughout the observation period, with significantly lower corneal opacity scores compared to vehicle-treated NM eyes (Fig. 8A–B). In parallel, JQ1 significantly attenuated acute corneal edema, as evidenced by reduced central corneal thickness during the early post-injury period (Fig. 8D–E). Importantly, during the later phases of injury, JQ1 promoted restoration of corneal layer organization and structural integrity, as evidenced by reappearance of the epithelial-stromal boundary on AS-OCT imaging, and effectively abolished the development of corneal NV (Fig. 8A, C, D).

Additionally, rabbit sham eyes treated with JQ1 showed no detectable changes in corneal transparency, thickness, or morphology, indicating that repeated topical administration of JQ1 during the treatment window was well tolerated and did not induce overt ocular toxicity (Fig. 8). Together, these cross-species findings demonstrate that the protective effects of BRD4 inhibition are conserved in a clinically relevant large-eye model, supporting its therapeutic potential for mustard-induced corneal injury.

Discussion

Mouse models enable efficient mechanistic studies and optimization of pharmacologic interventions, as reagents such as antibodies are widely available, and mice are relatively low-cost and easy to handle. Recently, mouse models of NM-induced ocular injury using either direct corneal application of NM solution or NM-soaked filter paper have been successfully employed to identify novel mechanisms underlying vesicant-induced keratopathy.^47–51 However, due to anatomical characteristics of the mouse eye, including a relatively small and mobile globe within the orbit, a shallow conjunctival fornix, and a comparatively thin Tenon’s capsule, liquid-applied agents may disperse beyond the corneal surface. Under certain experimental conditions, this dispersion could potentially extend posteriorly or periocularly, introducing relatively uncontrollable injury and complicating interpretation of cornea-specific outcomes.

To address this limitation, we developed a mouse model of NM-induced corneal injury using a 3D-molded, cap-shaped filter paper and a minimal volume of NM solution (1 μL). This approach represents a refinement of established exposure methods by conforming closely to the corneal curvature and spatially confining NM exposure to a defined central corneal region. Such localized delivery reduces potential off-target effects and experimental variability, thereby facilitating more precise assessment of corneal-specific injury mechanisms. By integrating multiple non-invasive imaging modalities including slit-lamp examination, AS-OCT, and IVCM, together with histological analyses, we provided a comprehensive and layer-resolved view of corneal injury kinetics. The acute phase (days 0–7) was dominated by epithelial disruption, marked edema, and intense inflammatory infiltration, consistent with alkylating agent-induced acute inflammation. During the pseudo-recovery phase (days 7–14), corneal transparency appeared to improve, but IVCM detected persistent hyper-reflective stromal bands, suggesting ongoing remodeling. In the chronic phase (days 14–84), fibrosis, epithelial irregularities, and progressive endothelial loss became the defining features, indicating that the initial injury set in motion irreversible structural changes. This pattern mirrors clinical observations in mustard-exposed patients, who may appear to recover initially but later develop delayed or chronic corneal disease.^52,53

Of interest, acute epithelial bullae, clearly documented by AS-OCT and IVCM (Fig. S14), emerged as early as day 3 following NM exposure and were most prominent in paracentral regions adjacent to the primary injury site. This spatial pattern, together with pronounced early endothelial cell loss, suggests that bulla formation results from endothelial pump failure, which generates the hydrostatic pressure required to elevate the epithelium. This process appears to occur preferentially in paracentral regions, where the barrier remains sufficiently intact to trap fluid, in contrast to the central region, where epithelial defects permit fluid leakage. Collectively, our findings suggest a potential role for endothelial integrity in shaping deep-tissue injury and raise the possibility that early endothelial assessment could inform estimates of long-term disease severity and therapeutic efficacy.

Using this model, we identified BRD4 as a previously unrecognized mediator of mustard-induced corneal injury. BRD4 expression and activity were rapidly upregulated following NM exposure, and early topical BRD4 inhibition markedly attenuated inflammation, stromal remodeling, fibrosis, and endothelial cell loss. Notably, transient inhibition during the acute phase was sufficient to confer long-term structural and optical preservation, indicating that early BRD4-dependent transcriptional programs play a critical role in determining chronic disease progression. The broad protective effects observed across multiple corneal layers suggest that BRD4 functions as a central regulator of vesicant-driven pathology rather than acting through a single downstream pathway. Importantly, the therapeutic benefit of BRD4 inhibition was conserved across species, extending from a mouse model to a rabbit model that exhibited more severe corneal pathology, including neovascularization. The efficacy observed in rabbits, whose corneal anatomy and wound-healing responses more closely resemble those of humans, strengthens the translational significance of this approach. Furthermore, BRD4 inhibition was achieved through topical ocular delivery, offering a practical therapeutic route that minimizes systemic exposure and enhances the safety profile for future clinical application of BRD4 inhibitors for the treatment of chemically induced corneal burns.

JQ1 is a pan-BET inhibitor that targets the BD1 and BD2 bromodomains with comparable affinity, but accumulating evidence indicates functional divergence between these domains: BD1 primarily sustains steady-state gene expression, whereas BD2 is preferentially engaged during inflammatory induction.^6,7,45,46 Accordingly, BD1 inhibition may disrupt homeostatic functions, while BD2-selective targeting could more specifically suppress pathological responses. In our study, selective inhibition of either BD1 or BD2 recapitulated the protective effects of pan-BET inhibition, indicating that both bromodomains contribute to NM-induced pathology. Although JQ1 treatment was well tolerated in uninjured corneas, BD2-selective inhibitors such as GSK620 may provide an improved safety profile for long-term or clinical use.

Mechanistically, our study suggests that BRD4 acts as a key epigenetic regulator amplifying the acute corneal injury response to NM exposure. NM-induced chromatin acetylation likely promotes BRD4-dependent transcription of inflammatory and immune genes, driving robust leukocyte recruitment with a pronounced bias toward T-cell infiltration. RNA-seq analysis showed that NM exposure broadly activates coordinated gene programs associated with inflammation, ECM remodeling, and oxidative stress. JQ1 treatment selectively reversed components of these programs rather than globally suppressing gene expression, consistent with BET inhibition functioning as a transcriptional “tuner” rather than a blanket repressor. In addition, JQ1 uniquely enhanced tight-junction assembly pathways, suggesting reinforcement of epithelial barrier integrity. These findings highlight the capacity of BET inhibition to target multiple converging pathological pathways simultaneously, an especially advantageous property given the complex, multifactorial nature of mustard-induced corneal pathology.

To our knowledge, only one previous study has investigated the role of BRD4 in corneal injury, demonstrating that JQ1 attenuated mechanical injury-induced corneal scarring through suppression of oxidative stress and myofibroblast formation.³³ Consistent with those findings, we observed that BRD4 inhibition reduced oxidative stress-associated responses and attenuated corneal fibrosis following NM exposure, suggesting that BRD4 may serve as a common regulator of pathological corneal remodeling across distinct injury settings. However, the pathological context of NM injury differs substantially from that of mechanical injury. Whereas mechanical injury primarily elicits wound-healing and fibrotic responses, NM exposure induces a complex, multiphasic disease process characterized by acute epithelial disruption, marked stromal edema, intense inflammatory infiltration, transient partial recovery, and subsequent progression to chronic stromal remodeling with persistent endothelial loss. These findings suggest that the pathogenic role of BRD4 extends beyond fibrosis and includes regulation of inflammatory, oxidative stress, and ECM remodeling pathways that contribute to both acute and chronic stages of vesicant-induced corneal injury. Importantly, whether BRD4 contributes to chemically induced corneal injury has remained unclear. Our findings fill this gap by demonstrating that BRD4 inhibition confers both acute and long-term protection against NM-induced corneal injury in mouse and rabbit models. Taken together, findings from mechanical and chemical corneal injury models support BRD4 as a promising therapeutic target for ocular surface injuries driven by distinct pathogenic mechanisms.

This study has several limitations. First, the present study focuses on the establishment and characterization of NM-induced corneal injury and the investigation of the therapeutic effects of BRD4 inhibition on anterior-segment pathology. However, increasing evidence suggests that ocular chemical injuries, including NM exposure, may affect not only the cornea but also the posterior segment of the eye.^50,51,54 Future studies employing dedicated posterior-segment assessment approaches may provide important insights into the effects of BRD4 inhibition on the broader ocular consequences of chemical injury. Second, pharmacokinetic studies were not performed. Consequently, corneal exposure and the retained concentrations of JQ1, GSK778, and GSK620 were not quantified. Characterization of ocular pharmacokinetic profiles through dedicated liquid chromatography-tandem mass spectrometry (LC-MS/MS) analyses will be required to determine their retention within corneal tissue and optimize therapeutic dosing strategies, thereby supporting future translational development. Third, although short-term topical BRD4 inhibition was well tolerated in the cornea under the conditions tested and theoretical systemic exposure was substantially lower than doses commonly used in systemic studies,^55–58 comprehensive evaluation of long-term ocular and systemic safety will be important for future clinical translation. Fourth, our findings demonstrate that JQ1 attenuates NM-induced corneal inflammation and identify multiple inflammatory genes and pathways suppressed following BRD4 inhibition, suggesting that BRD4 promotes NM-induced corneal inflammation through coordinated regulation of multiple inflammatory transcriptional programs rather than a single downstream effector, consistent with the established role of BRD4 at super-enhancers in broadly amplifying injury-responsive gene expression across multiple pathways. However, future studies incorporating chromatin occupancy analyses (e.g., ChIP-qPCR or ChIP-seq) and functional perturbation of candidate downstream pathways will be necessary to define the mechanisms by which JQ1-mediated BRD4 inhibition regulates NM-induced corneal inflammation and to verify the interactions among downstream target genes and related pathways identified by our transcriptomic analyses. Finally, although the therapeutic effects of BRD4 inhibition were consistently observed in both murine and rabbit models, the absence of clinical data remains a limitation, warranting further validation in human corneal tissues and clinical studies.

Conclusion

We established a controlled and reproducible murine model of NM-induced central corneal injury and systematically characterized its temporal and spatial pathology. Using this model, we identified BRD4 as a key epigenetic regulator driving inflammatory, fibrotic, and endothelial injury responses. Post-exposure inhibition of BRD4 not only reduced acute inflammation and edema but also preserved corneal transparency and endothelial integrity in the long term. Importantly, the protective effects of BRD4 inhibition were replicated in a clinically relevant rabbit model, strengthening the translational relevance of this approach. The ability to deliver BRD4 inhibitors topically, thus limiting systemic exposure, further enhances the clinical feasibility of this strategy. Our findings highlight the pathological role of BRD4 in corneal inflammation, neovascularization, fibrosis, and endothelial dysfunction, and support continued translational development and evaluation of BRD4/BET inhibitors and degraders,^59–61 including formulations designed to improve drug retention and tissue penetration,^62,63 as therapeutic candidates for corneal diseases.

Materials and methods

Experimental animals

To characterize the mechanisms underlying corneal injury and define the role of BRD4 in corneal inflammatory responses and remodeling, we employed complementary small and large animal models. A mouse model was used for mechanistic and discovery analyses, whereas a rabbit model was selected for its corneal size, wound-healing characteristics, and clinical relevance to human ocular injury, thereby enabling translational validation of BRD4 inhibition as a molecular therapy.

C57BL/6J mice were originally obtained from Jackson Laboratory (Bar Harbor, ME) and housed under standard laboratory conditions with controlled temperature and humidity and a 12-hour light/dark cycle. Adult mice of both sexes (15–20 weeks old) were used in this study. For cross-species validation experiments, young adult New Zealand White rabbits of both sexes (approximately 9–10 weeks old at the time of experimentation) were obtained from Charles River Laboratories (Wilmington, MA), acclimated for two weeks, and housed under the same standard laboratory conditions.

All animal procedures involving mice and rabbits were conducted in accordance with the Association for Research in Vision and Ophthalmology (ARVO) Statement for the Use of Animals in Ophthalmic and Vision Research and in compliance with the Animal Welfare Act and the U.S. Department of Agriculture (USDA) regulations, and were approved by the Institutional Animal Care and Use Committee (IACUC) of the University of Texas Medical Branch (UTMB).

Establishment of a localized corneal chemical injury model

Based on corneal bioparameters of adult C57BL/6J mice,⁴¹ a hemispherical base and rod module with a curvature radius of 1.5 mm were designed and 3D-printed. This module was used to mold 2-mm-diameter filter paper into a cap shape that precisely conformed to the curvature of the central cornea (Fig. S1). Anesthesia in mice was induced via intraperitoneal injection of ketamine (100 mg/kg body weight) and xylazine (10 mg/kg body weight). Topical anesthesia was achieved by applying 0.5% proparacaine hydrochloride (Sandoz, Alcon Laboratories, TX) to the eyes and allowing it to take effect for 30 s. Residual eye drops and tears were gently removed using a sterile ophthalmic sponge spear moistened with phosphate-buffered saline (PBS). Chemical injury was induced by placing a dry filter cap onto the designated eye, followed by application of the chemical agent onto the cap. After exposure, the injured eye was thoroughly rinsed with saline for at least 2 minutes. The control eye was treated with saline using identical time intervals and procedures. Mice were monitored daily. Corneal melting, perforation, or severe infection was defined as humane endpoints, and animals exhibiting these signs were euthanized immediately.

For alkali burn injury, 1 μL of 1.0 M NaOH was applied to the filter cap. For nitrogen mustard (NM) experiments, mechlorethamine hydrochloride (122564–5G; MilliporeSigma, Burlington, MA) powder was freshly dissolved in pure water to yield concentrations of 1%, 2%, or 4%. Subsequently, 1 μL of the NM solution was administered to the eye. Unless otherwise specified, a standardized protocol consisting of 2% NM exposure for 5 minutes was used to induce reproducible corneal injury.

For cross-species validation in rabbits, the model was adapted to accommodate larger corneal dimensions. Filter caps with an 8-mm diameter and a curvature radius of 7 mm were used.⁶⁴ Anesthesia was induced via intramuscular injection of ketamine (30 mg/kg body weight) and xylazine (3 mg/kg body weight), supplemented with topical 0.5% proparacaine and a retrobulbar injection of 1.5 mg bupivacaine (Hospira Inc., Lake Forest, IL). NM exposure parameters were adjusted to achieve injury severity comparable to that observed in mice, using 16 μL of 2% NM applied for 5 minutes. Peri-procedural analgesia with sustained-release buprenorphine (Wedgewood Pharmacy, Swedesboro, NJ) was administered 2 hours before and 72 hours after NM exposure to minimize pain and discomfort. All other procedures, including chemical application and ocular irrigation, followed the same protocols as those used in mice.

BET inhibitor treatment

The pan-BET inhibitor JQ1 (HY-13030), the BD1-selective inhibitor GSK778 (HY-136570), and the BD2-selective inhibitor GSK620 (HY-137892) were purchased from MedChemExpress (Monmouth Junction, NJ). Stock solutions were prepared in dimethyl sulfoxide (DMSO) and diluted in PBS containing 10% hydroxypropyl-β-cyclodextrin (HPβCD) immediately before use. JQ1, GSK778, and GSK620 were administered at final concentrations of 1 mM, 1 mM, and 3 mM, respectively. These concentrations were selected based on the following considerations. Due to the well-recognized limitations of topical ocular drug delivery, including rapid precorneal clearance and restricted tissue penetration, ophthalmic formulations are often administered at concentrations that are orders of magnitude higher than their biochemical half-maximal inhibitory concentration (IC₅₀) values. Accordingly, millimolar concentrations of potent signaling inhibitors are commonly used in topical ophthalmic applications.^65–70 Because JQ1 inhibits BRD4-BD1 and BRD4-BD2 with IC₅₀ values of 77 nM and 33 nM, respectively,⁵⁵ a topical concentration of 1 mM was selected to maximize the likelihood of achieving effective BRD4 inhibition in corneal tissues following administration. GSK778 was administered at 1 mM because its potency toward BRD4-BD1 (IC₅₀ = 41–75 nM) is comparable to that of JQ1.⁴⁶ In contrast, GSK620 exhibits lower potency toward BRD4-BD2 (IC₅₀ = 72–285 nM);⁴⁶ therefore, a higher concentration (3 mM) was selected to provide a comparable level of target engagement. DMSO was used solely as a vehicle to solubilize the BRD4 inhibitors. We aimed to keep the DMSO concentration in the formulation as low as possible while maintaining adequate compound solubility and achieving the target drug concentration for ophthalmic administration. Based on the solubility characteristics of the compounds, a final concentration of 1% DMSO was selected. This concentration is substantially lower than those reported in previous ophthalmic studies, in which topical eye-drop formulations containing higher concentrations of DMSO demonstrated acceptable ocular tolerance.^71,72 Working solutions were prepared fresh immediately before each administration. Topical treatment with a BET inhibitor or vehicle (PBS containing 1% DMSO and 10% HPβCD) (5 μL per application for mice and 50 μL per application for rabbits) was initiated 2 hours after NM injury and administered five times daily for 3 days in the acute studies or for 14 consecutive days in the chronic studies.

Multimodal, longitudinal, non-invasive live imaging

Eyes were monitored from baseline through defined endpoints using slit-lamp biomicroscopy, anterior segment optical coherence tomography (AS-OCT), and in vivo confocal microscopy (IVCM). Both white- light and fluorescein-stained images were obtained using a Topcon SL-D701 slit-lamp biomicroscope (Topcon, Oakland, NJ). AS-OCT imaging was performed using the Bioptigen Envisu R2200 SD-OCT system (Bioptigen, Durham, NC) with an 18-mm telecentric lens. IVCM imaging was conducted using a Heidelberg HRT3 RCM system (Heidelberg Engineering, Franklin, MA) with a 300 × 300 μm field of view. Images were acquired en face at the center of the ocular surface throughout the full thickness of the cornea, as well as in a lateral orientation to obtain a cross-sectional profile. In alkali burn experiments, fluorescein staining images were captured immediately after rinsing to visualize the injured areas. Subsequent imaging was performed using slit-lamp and AS-OCT on days 7 and 14 post-injury. For NM injury experiments in mice, pre-injury images were captured 5 days prior to exposure to establish baseline measurements. Following NM exposure, live imaging, including slit-lamp biomicroscopy, AS-OCT, and IVCM, was conducted on days 3, 7, 14, 21, 42, and 84 post-injury to reveal longitudinal pathological changes. For rabbit experiments, imaging was performed on days 3, 7, 14, 28, and 42 post-injury to evaluate the effects of early JQ1 intervention.

Tissue collection and histology

Mouse tissues were collected on days 3 and 84 post-injury to evaluate pathological changes at early and late stages, respectively. For the 3-day samples, eyeballs were harvested following euthanasia, embedded in optimal cutting temperature medium, immediately frozen, and sectioned for immunohistochemical staining. For the 84-day samples, eyeballs were collected in two ways: some were fixed in 4% paraformaldehyde (PFA) at 4 °C for 2 hours and then dissected as corneal flat mounts, while the remaining samples were processed into frozen sections using the same method described for the 3-day samples.

To prepare corneal flat mounts, corneas were rinsed with PBS and blocked in a solution containing 5% normal goat serum and 0.5% Triton X-100 for 3 hours. The tissues were then incubated overnight at 4 °C with rat anti-mouse CD45 (550539; BD Pharmingen, San Diego, CA) and ZO-1 polyclonal antibody (61–7300; Invitrogen, Waltham, MA). After rinsing, the corneas were incubated with Alexa Fluor-conjugated secondary antibodies (Thermo Fisher Scientific, Waltham, MA) for 4 hours at 4 °C, followed by DAPI staining (Thermo Fisher Scientific) for 40 minutes. The prepared corneas were mounted on slides and imaged using a confocal microscope (LSM800, Zeiss, Thornwood, NY).^73,74

Frozen sections were air-dried and fixed in 4% PFA at room temperature for 10 minutes, followed by blocking in a solution containing 5% normal goat serum and 0.1% Triton X-100 for 1 hour. Primary antibodies against CD45 (550539), CD3 (11–0031-82; eBioscience, San Diego, CA), Col1a1 (72026;Cell Signaling Technology, Danvers, MA), BRD4 (ab128874; Abcam, Waltham, MA), acetylated histone H3 (K122) (ab33309; Abcam), and 8-OHdG (BS1278R; Bioss, Woburn, MA) were applied to designated sections and incubated overnight at 4 °C. Alexa Fluor 488- or 568-conjugated secondary antibodies were used for detection. All sections were mounted using Fluoroshield mounting medium with DAPI (F6057; MilliporeSigma) and imaged using a confocal microscope.

Measurement of superoxide production

To evaluate superoxide production in situ, dihydroethidium (DHE) staining was performed as previously described.⁷⁵ In brief, fresh-frozen corneal sections were pre-incubated with 10 mM HEPES buffer (J61275; Thermo Fisher Scientific) in the presence or absence of superoxide dismutase (SOD; 400 U/mL; S9549, MilliporeSigma) at 37 °C for 10 minutes in a humidified chamber. Subsequently, DHE (D11347; Thermo Fisher Scientific) was applied and incubated at 37 °C for 15 minutes in the dark.⁷⁵ After three washes with HEPES buffer, the sections were imaged using a confocal microscope.

Image processing and quantitative analysis

Slit-lamp images were evaluated using a standardized five-point scale based on the degree of corneal opacification.⁷⁶ The grading criteria were defined as follows: Grade 0, no corneal haze; Grade 1, iris details visible; Grade 2, pupillary margin visible but iris details not visible; Grade 3, pupillary margin not visible; and Grade 4, cornea totally opaque. OCT images were analyzed using ImageJ/Fiji (NIH) to measure central corneal thickness (CCT).^77,78 The area of corneal neovascularization (NV) in rabbits was quantified using a standardized geometric formula involving a combination of intraoperative recording and post-hoc image analysis.⁷⁹ The number of clock hours ($C$) of limbal involvement was recorded manually during examination, while the vessel length ($L$) was acquired from slit-lamp images. The NV area (mm²) was calculated using Equation (1):

$$Area=\frac{C}{12}\times \pi \times [r^2-(r-L{)}^2]$$ (1)

where measurements were standardized using the corneal radius ($r=6.0mm$) as a biological reference. For eyes in the injury group in which vessels reached the central optical zone, $L$ was defined as equal to the corneal radius ($r$).

For the histological analyses, cell counts and fluorescence intensity were quantified using ImageJ. Corneal endothelial images were analyzed using the built-in functions of ImageJ and the MorphoLibJ plugin.⁸⁰ Images were converted to grayscale, background-corrected, binarized, and cell boundaries were delineated using watershed segmentation. Morphometric parameters were extracted using the Analyze Regions tool in MorphoLibJ. Neighbor relationships were determined using connected component labeling and region adjacency graph construction. Cells located at image borders and cells with an area <100 μm² were excluded to eliminate partial or fragmented cells. ImageJ-derived outputs were further processed in R (tidyverse, RImageJROI, EBImage), where regions of interest (ROIs) defined in ImageJ were used for data integration, quality control, spatial neighbor analysis, and generation of summary statistics and colored ROI visualizations.^80,81 Central corneal endothelial parameters included endothelial cell density (ECD), coefficient of variation of cell size (CV), hexagonality (HEX), minimal and maximal cell sizes (S_min, S_max), average cell size (S_avg), and standard deviation of cell area (S_sd).

Quantitative real-time PCR (RT-qPCR)

Total RNA was extracted from mouse central corneas using the miRNeasy Mini Kit (217004; Qiagen, Valencia, CA). Complementary DNA (cDNA) was synthesized from total RNA using the High-Capacity cDNA Reverse Transcription Kit (4368813; Thermo Fisher Scientific). Quantitative real-time PCR was performed on a StepOnePlus Real-Time PCR System (Applied Biosystems, Waltham, MA) using SYBR Green Master Mix (4368708; Thermo Fisher Scientific). Relative gene expression was calculated using the ΔΔCt method, with Hprt serving as the internal reference gene. Primer sequences were as follows: Ccl2 For-5′-CAG GTC CCT GTC ATG CTT CTG-3′; Ccl2 Rev-5′-GAG CCA ACA CGT GGA TGC T-3′; Cxcl1 For-5′-GCA GAC CAT GGC TGG GAT T-3′; Cxcl1 Rev-5′-CCT GAG GGC AAC ACC TTC AA-3′; Cxcl2 For-5′-GCG CCC AGA CAG AAG TCA TAG-3′; Cxcl2 Rev-5′-AGG GTC AAG GCA AAC TTT TTG A-3′; Cxcl10 For-5′-GGA CGG TCC GCT GCA A-3′; Cxcl10 Rev-5′-CCC TAT GGC CCT CAT TCT CA-3′; Il1b For-5′-AGT TGA CGG ACC CCA AAA GA-3′; Il1b Rev-5′-GGA CAG CCC AGG TCA AAG G-3′; Hprt For-5′-GAA AGA CTT GCT CGA GAT GTC ATG-3′; Hprt Rev-5′-CAC ACA GAG GGC CAC AAT GT-3′.

Bulk RNA sequencing and data processing

C57BL/6J mice were randomly assigned to vehicle or JQ1 treatment groups following exposure to NM or saline. After 3 days of treatment, central corneas from the four experimental groups (n = 6 biological samples per group) were dissected and flash-frozen. Total RNA was extracted from each sample using the miRNeasy Mini Kit (217004) and assessed using a NanoDrop 2000 (Thermo Fisher Scientific). High-quality RNA libraries were prepared using the NEBNext Ultra II RNA Library Prep Kit (E7775L; New England Biolabs, Ipswich, MA) and subsequently submitted to the UTMB Next Generation Sequencing Core for RNA sequencing (RNA-Seq). All downstream analyses, beginning with the raw gene-level count matrices provided by the sequencing facility, were performed in R (version 4.4.3).^82,83 Initial quality assessment included evaluation of gene count distributions, sequencing depth, global sample clustering, and visualization using principal component analysis (PCA). Differential expression analysis was conducted using DESeq2 for comparisons of NM + Vehicle vs. Sham + Vehicle and NM + JQ1 vs. NM + Vehicle. Significant genes were defined as those with an absolute log₂ fold change exceeding the median by more than 1.5 times the median absolute deviation (MAD) and an adjusted p-value < 0.05. Gene Ontology (GO) enrichment analysis and Gene Set Variation Analysis (GSVA) were performed to characterize biological processes and pathway alterations associated with NM injury and modulated by JQ1 treatment. Differentially expressed genes (DEGs) were mapped to the STRING database, and protein-protein interaction (PPI) networks were analyzed using the Louvain algorithm to identify functional gene modules and hub genes reversed by JQ1 treatment. Transcriptomic comparisons between the Sham + JQ1 and Sham + Vehicle groups were conducted to assess potential off-target or safety-related effects of JQ1.

Statistics

Sample sizes were determined based on our pilot studies using the NM-induced corneal injury model and on group sizes commonly used in published corneal injury studies. Individual animals were considered biological replicates and served as experimental units for statistical analyses. All analyses were performed using Prism (GraphPad Software, Boston, MA) or R. Data are presented as mean ± SEM. Comparisons between two groups were performed using Student’s t-test or Welch’s t-test depending on variance homogeneity, while comparisons across more than two groups were analyzed using one-way ANOVA followed by appropriate post-hoc testing. For longitudinal observations, including opacity grade, CCT, and NV area, multiple Mann-Whitney U tests or Kruskal-Wallis tests were performed at each time point. To account for multiple testing, raw p values were adjusted using the Benjamini-Hochberg false discovery rate (FDR) procedure with a 5% threshold. Differences were considered significant at p < 0.05.

Lu and colleagues identify BRD4 as a central epigenetic driver of vesicant-induced corneal injury. Using reproducible mouse and rabbit models, they show that short-term topical BRD4 inhibition suppresses acute inflammation and provides durable protection of corneal clarity, stromal organization, endothelial integrity, and neovascularization, supporting translational therapy for chemical eye injuries.

Data availability:

Data presented are available within the article and the supplemental materials. Additional information can be accessed from the corresponding author upon request.