The study protocol adhered to the tenets of declaration of Helsinki and was approved by the Institutional Review Board (LEC02-14-029) of the L V Prasad Eye Institute (LVPEI). Preterm babies referred for further management from the neonatal intensive care units of different hospitals in Hyderabad to the LVPEI between January 2007 and December 2010 were enrolled. Overall, the study cohort comprised 372 preterm babies of GA ≤ 35 weeks and/or BW ≤ 1,700 g with ROP (n = 189) and no-ROP (n = 167). A detailed demographic and clinical history (Table S1 in Supplementary Material) of all the preterm babies enrolled were documented and a written informed consent was obtained from their parents. The diagnosis and categorization of ROP cases from mild to severe form was based on severity (stages 1–5), location (zones I, II, III), amount of disease (clock hours), and presence or absence of “plus” disease following ICROP guidelines (20, 21) (Figure S2 in Supplementary Material). Severe ROP includes progressive disease, which requires prompt treatment. It includes any stage (1–5) Zone I with plus and stages 2–3 Zone II with plus. Mild ROP cases include less severe disease, which does not require any treatment. Although until the regression of the disease completely, babies are under regular follow-up for ROP screening.

Venous blood (0.5–1 mL) was collected from the ROP and no-ROP preterm babies by venipuncture. DNA was extracted from the blood samples using an automated DNA extraction platform (MagNa Pure LC 20, Roche) following the manufacturers guidelines. Likewise, for proteomic studies, the vitreous humor samples (100–500 µL) were collected from preterm babies with stage IV and V ROP (n = 30) who had undergone vitrectomy as a part of their routine clinical management. The controls for the proteomic studies included babies with congenital cataract (<6 months of age) who underwent partial vitrectomy as part of the surgical management (n = 30). The vitreous samples were immediately centrifuged at relative centrifugal force (rcf) of 10,621 g and the supernatant was stored at −80°C deep freezer until further use.

Additionally, crude tears were collected before instilling any drops or drug for pupil dilation in the eyes of preterm babies with ROP (Stage II–V) (n = 27) and no-ROP (n = 13) using a capillary tube and without touching the conjunctiva. The tear samples for ROP subjects were collected during the active disease condition either before or after 3 months following medical intervention.

A customized panel containing 384 single-nucleotide polymorphisms (SNPs) from 26 chosen genes (Table S2 in Supplementary Material) involved in growth and development of the fetal retina, angiogenesis, inflammation, neurodegeneration, and oxidative stress processes were genotyped using a microarray platform (Illumina Inc., golden gate assay). Following hybridization, the fluorescent signals were scanned by a bead array reader and the raw signal intensities were imported to the Genome Studio software (version 1.9) for assessing quality scores. The assay and sample reliability were measured by means of the gen call score and the genotypes were called following clustering. The genotypes of a subset of samples for all the genes were validated by resequencing on an automated DNA sequencer (ABI 3130 XL) using the BigDye chemistry.

The concentrations of 27 cytokines (Bio-Plex Human cytokine 27-Plex, Bio-Rad, Hercules, CA, USA) and 28 different proteins (HMMP1-55K, HMMP2-55K, HNDG1-36K, HNDG2-36K, HTIMP2-54K, TGFB-64K-03, HYCYTOMAG-60K, Merck Millipore, Billerica, MA, USA) involved in ECM remodeling, angiogenesis and inflammatory pathways were screened by multiplex bead immunoassays using the Luminex xMAP technology in vitreous samples that were pre-diluted to concentration 1:3. Similar assay was used for estimating the concentrations of MMPs and cytokines in the tears samples and were quantitated by comparing them with their respective standard curve. All standards and some of the samples (due to less volume of samples) were measured in duplicates.

A part of vitreous sample (50–100 µL) was lysed using a buffer containing 50 mM Tris-HCl (pH = 8), 120 mM NaCl, 0.5% NP40, protease inhibitor cocktail, and precipitated with acetone. The protein pellet was eluted in 50 µL of 1× phosphate buffered saline (PBS) containing the protease inhibitor cocktail and quantified by bicinchoninic acid assay. The normalized vitreous proteins (10 µg) were then subjected to western blotting. Western blotting was done using mouse anti-human C3 antibody (sc-28294, Santa Cruz) and mouse anti-human C5 (MAB2037, R&D Systems) followed by incubation with IRDye^® 680RD secondary antibody. The entire procedure was done according to the manufacturer’s recommended application protocol (https://www.licor.com).

MMP gelatinase activity was measured in the vitreous and tear of ROP babies and controls by zymography as described earlier (22). An equal volume of crude vitreous and tear samples were electrophoresed under non-reducing conditions in 10% SDS-PAGE gels polymerized with 1 mg/mL gelatin. The gel was washed with 2.5% Triton X-100 for 30 minutes at room temperature with gentle agitation, followed by rinsing with distilled water. The gel was then incubated for 30 min in developing buffer containing 50 mM Tris-HCl, pH 7.8, 5 mM CaCl₂, 0.2 M NaCl, 0.02% Brij 35. The gel was incubated with fresh developing buffer at 37°C for 16 h and stained with Coomassie blue (Bio-Rad). This was followed by destaining with 10% v/v methanol, 5% v/v acetic acid in dH₂O. Active MMP (MMP2/MMP9) band was detected in the zymogram in the discovery cohort. The observations of discovery cohort were further validated in tears in an extended cohort of patients at different stages of ROP: no-ROP (n = 9), mild ROP regressed (n = 6), mild ROP_progressed (n = 7), and severe ROP (n = 12) by zymography in order to confirm their role as biomarker in ROP pathogenesis. Zymogram band intensities were calculated with ImageJ software.

Vitreous were subjected to cytospin to separate the cells that were embedded in the parafilm block. Sections were cut, air dried, and stained with H&E (23) for understanding their morphology. Tissue sections were then deparaffinized using xylene and gradually rehydrated with ethanol. Antigen retrieval was done by microwaving the sections at full power for 4–5 min in Tris EDTA buffer (10 mM Tris, 1 mM EDTA, 0.5% tween 20, pH 9.0). Blocking was carried out with 2.5% (w/v) BSA in PBS (10 mmol/L sodium phosphate, pH 7.5, 120 mmol/L sodium chloride) for 30 min at room temperature. Thereafter, the slides were incubated for 60 min with the primary antibody (CD 68 for human 1:100) diluted in 1× PBS, followed by three washings with PBS. Further incubation was carried out with biotinylated anti-mouse immunoglobulin. Sections were then washed in PBS and incubated with avidin DH/biotinylated horseradish peroxidase reagent in PBS for 30 min before final washing. The antigen was localized using 1 mg/mL diaminobenzidene tetrahydrochloride (DAB; Sigma), 0.2% H₂O₂ in 50 mmol/L Tris–HCl, pH 7.6, which appeared as a brown end product. Sections were then counterstained with DAPI for nuclei staining.

The human microglial cell line (CHME3) was cultured in DMEM containing 10% FBS along with antibiotics penicillin and streptomycin. The confluent cells were trypsinized using 0.25% trypsin-EDTA. Hypoxic stress was introduced in the microglia cells by treating them with Cobalt chloride (CoCl₂) at various concentrations from 100 to 250 µm. Briefly, around 15,000 cells were seeded on a six well plate and then serum deprived for 6 h, followed by treatment with CoCl₂ for 24 h on attaining 70–80% confluency. The serum deprived cells in the same duration that were not treated for hypoxia were used as controls.

The cells were washed in HBSS (Thermoscientific, Waltham, MA, USA) and then incubated with the calcium binding dye Flu-4 (diluted with HBSS 1:750) for 30 min. After washing the cells with HBSS three times, live cell imaging was performed for 10 min using an EVOS fluorescent microscope (Thermo Fischer Scientific, Waltham, MA, USA) under 20× magnification. The cytosolic calcium flux was measured using the change in Fluo-4 intensity over time for individual cells (Excitation: 494 nm Emission: 506 nm) (24).

The RNA from untreated and treated cells was extracted by Trizol method (25). The cDNA was prepared using iScript cDNA synthesis kit (Bio-Rad, CA, USA). Semi quantitative PCR was carried out using the specific primers (Table S3 in Supplementary Material) for VEGF165, C3, HIF1α, BAX, and IL-1β while β-actin was used as an endogenous control.

Allele frequencies of all the 384 variants were calculated by gene counting method along with odds ratio and 95% CI. A p value < 0.05 was considered to be significant. The associated allele and haplotype frequencies were further analyzed for statistical correction using Bonferroni and permutations tests (n = 10,000 permutations). Estimates of Hardy–Weinberg equilibrium (p > 0.001), linkage disequilibrium (LD), and haplotype frequencies were calculated using the Haploview software (version 4.2) (26).

Protein and cytokine levels in ROP and control samples were represented as bar plot (the mean ± SE) and box plot (median, interquartile range, and whiskers). Comparison of proteins and cytokines levels between ROP and controls vitreous/tears were calculated using the unpaired Student’s t-test. A p-value < 0.05 was considered to be statistically significant. Since the cytosolic calcium level does not follow a normal distribution, we performed the testing of equality in medians for control and hypoxic condition using Wilcoxon rank-sum test.

Of the 384 variants screened (Table S2 in Supplementary Material), 73 were removed from further analysis as they were either not in Hardy-Weinberg equilibrium in the controls (n = 16), were monomorphic (n = 44), or had a call rate < 97% (n = 13). Thus, 311 SNPs from 26 genes were finally analyzed for association with ROP. Among these 37 SNPs in 14 genes (AGTR1, ANGPT2, C3, CFH, CFB, CXCR4, FBLN5, H2AFX, IHH, MMP2, TGFβ1, CETP, VEGF, and TSPAN12) exhibited significant association (p < 0.05) with ROP (Table 1). Additionally, 5/37 associated SNPs in CFH, CFB, CXCR4, FBLN5, and CETP genes withstood Bonferroni correction. Intriguingly, only the CETP variant (rs891141) conferred significant risk of ROP, while the variants across the other genes were protective (Table 1). Strong LD was observed across all the variants (except rs1831821) in CFH and rs891141 and rs289716 in CETP gene, while moderate LD was observed between rs891141 and rs289713 in CETP and rs2268002 and rs2284340 in FBLN5 (Figure S1 in Supplementary Material).

Likewise, haplotypes generated with the associated and flanking variants of these five genes revealed that only the haplotype C-A-T in CETP conferred significant risk of ROP, while those with CFH and FBLN5 were protective. Haplotypes with the CXCR4 and CFB were not informative (Table 2). Thus, the present study highlights the potential involvement of novel genes (CFH, CFB, CETP, FBLN5, and CXCR4) in ROP based on their allelic and haplotype associations.

Based on strong associations in the CFH and CFB genes, a quantitative assessment of a neurodegenerative panel containing CRP, SAP, MIP-4, Complement C4, apolipoprotein AI, apolipoprotein CIII, apolipoprotein E, Complement Factor H, and Complement C3 proteins was carried out by multiplex immuno-bead assay in the vitreous samples of ROP patients (n = 30) and controls (n = 30). All the complement components and apolipoproteins were detectable in the vitreous samples. Overall, we observed significantly elevated levels of C3 (p = 0.05), C4 (p = 0.001), CFH (p = 2.24 × 10⁻⁵), VEGF (p = 0.0027), apolipoprotein AI (p = 0.0007), and apolipoprotein CIII (p = 0.004) in the vitreous of ROP compared to the control subjects indicating their possible involvement in the disease pathogenesis (Figure 1A).

We validated the differential expression of complement component C3 by western blotting. An intense band of 192 kDa corresponding to C3 molecule was observed in ROP cases compared to controls (Figure 1B). Additionally, we observed the activated C3 fragments; C3b (182 kDa), C3c (145 kDa), and iC3bα (63 kDa) in the ROP vitreous under non-reducing conditions as confirmed by mass spectrometry (data not shown here), but not in the controls (Figure 1B), suggesting a higher activation of C3 in ROP. Likewise, a higher expression of complement component C5 was observed in ROP vitreous compared to the controls (data not shown) suggesting a further activation of the complement pathway.

We further demonstrated that along with increased expression of angiogenic molecules in the vitreous samples of patients, activated macrophages/microglia in turn would also be secreting proinflammatory cytokines that might exacerbate the inflammation, further playing a role in the ROP pathogenesis. We detected activated macrophages/microglia on H&E, further confirmed by immunostaining with CD68 in ROP vitreous but not in the controls (Figure 1C). The results of this experiment supported for shift in the proangiogenic state as demonstrated by a significant increase in the levels of cytokines IL8 (p = 0.0149), G-CSF (p = 0.0099), IL1ra (p = 0.0019), and VEGF (p = 0.0027) (Figures 1A,D) along with marginal increase of IL6, IL12, IL7, RANTES, and MCP1 in the ROP vitreous (data not shown).

To further confirm these results, we subjected the cultured microglial cells to hypoxic condition and checked for the expression of proinflammatory markers. The effect of hypoxia on the activation of macrophages/microglia was observed with an intense calcium staining in cells exposed to hypoxia compared to the unexposed ones. The result shows that there is increase in cytosolic calcium levels in case of hyperactivated cells subjected to 24 h of hypoxic stress (Figure 2A, n = 50). Specifically, there is a significant increase (p < 0.0005) in cytosolic calcium (Ca_max = the maximum Fluo-4 intensity) in microglial cells followed by hypoxia exposure (Figure 2B). Likewise, a higher expression of complement C3, VEGF165, and hypoxia inducing factor-1α (HIF-1α) was also observed in exposed cells (Figure 2C).

A strong association of SNPs in FBLN5 and moderate association of MMP2, TGFβ gene (Table 1) with ROP suggested the role of ECM proteins in ROP pathogenesis. Further, a quantitative assessment of the ECM proteins indicated a significant increase in MMP9 (p = 0.038), TIMP1 (p = 0.004), and α2 macroglobulin (p = 0.0018) in the ROP vitreous (Figure 3A). We also assessed the MMP activation in ROP by gelatine zymography. Our results showed higher levels of both pro and activated MMPs (MMP9 and/or MMP2) in the vitreous of patients suggesting its potential role in disease pathogenesis (Figure 3B).

We explored if increased expression of inflammatory markers (as seen in the vitreous samples of ROP patients) could also be reproducibly detected in tears and further be established as the biomarker for disease progression. A quick multiplex ELISA of tear samples collected from the ROP babies at different stages and no-ROP preterms was performed for some inflammatory markers (interleukins, TNFα, IFNγ, and MMPs). Significantly higher expressions of IL-1ra (p = 0.014), MMP2 (p = 0.0085), and MMP-9 (p = 0.03) were detected in severe ROP cases compared to mild ROP and no-ROP tear samples that was further confirmed by zymography (Figures 3C–E). On the zymogram, the tear samples from no-ROP showed very low expression of MMPs as compared to severe ROP (Figure 3E). These results were confirmed to be reproducible in the extended cohort of ROP with a significant increased expression of activated MMP2 in severe ROP (p = 0.0023) and progressive ROP (p = 0.007) as compared to mild ROP (p = 0.01) and premature controls. Similar pattern of gradual increase in MMP9 expression was also observed in mild ROP (p = 0.02) to progressive (p = 0.001) and severe ROP cases (p = 1.2 × 10⁻⁶) with respect to premature controls (Figure 4).