Two hundred patients along with their parents were enrolled via the Eye and ENT Hospital Biobank from 2004 to 2018. The diagnosis of PCG must agree with the following criteria: (1) expansion of corneal diameter, corneal clouding, or Haab’s striae; (2) IOP >21 mm Hg; (3) optic nerve cupping or asymmetric cup-to-disc ratio; and (4) disease onset before age 3. Other developmental glaucoma or secondary glaucoma, such as Peter’s anomaly, Axenfeld–Rieger syndrome, and aniridia were excluded.

All PCG cases were confirmed by glaucoma specialists before surgery under general anesthesia. Anterior segment examination and corneal diameter measurement were performed with the aid of surgical microscopes. IOP was measured using a Tono-PEN tonometry (Mentor, Norwell, MA, United States). The structures of anterior chamber angle and the optic disc were observed under gonioscopy whenever possible. Five milliliters of venous blood was collected from each PCG patient and his/her parents and preserved in the biobank for genomic DNA extraction.

In this study, a customized panel with 58 pairs of primers was designed to cover the coding regions and 50-bp sequences flanking the splice sites of TEK gene (Supplementary Table S1). DNA samples were amplified by multiplex PCR and sequenced by a Hiseq4000 platform (Illumina, San Diego, CA). Sequencing reads were aligned to the human reference genome assembly (hg19). Variants were called by Genome Analysis Toolkit (version 3.70) and annotated by ANNOVAR (Wang et al., 2010). Exonic and splice site (±5 bp) variants with an allele frequency <0.1% in the Genome Aggregation database (gnomAD_all and gnomAD_eas) were validated by Sanger sequencing and selected for further analysis. Additionally, four probands were exome sequenced using a SureSelect Human All Exon Kit (Aglient, Santa Clara, CA, United States) and the aforementioned Hiseq4000 platform. When available, parental DNA samples were also subjected to Sanger sequencing for co-segregation analysis.

Twelve in silico tools were integrated to predict the pathogenicity of non-synonymous variations, including SIFT (Sim et al., 2012), Polyphen-2 (Adzhubei et al., 2010), LRT (Chun and Fay, 2009), MutationTaster (Schwarz et al., 2014), MutationAssessor (Reva et al., 2011), FATHMM (Shisha et al., 2013), PROVEN (Choi and Chan, 2015), VEST (Carter et al., 2013), Meats (Kim et al., 2017), M-CAP (Jagadeesh et al., 2016), CADD (Kircher et al., 2014), and DANN (Quang et al., 2015). Multiple Tie2 protein sequences from different species were retrieved to evaluate the evolutionary conservation of affected residues. To assess the effect of residue substitution on protein structure and stability, the FoldX plugin for YASARA (Van Durme et al., 2011) was used to calculate the free energy change induced by specific mutations with three repetitions each, and the resulting structures were visualized with YASARA View (Krieger and Vriend, 2014). For variants located within immunoglobulin (Ig)-like domains and the epidermal growth factor (EGF)-like domain, the Tie2 ligand-binding domain crystal structure (PDB ID: 2GY5, 23-445 aa) was utilized as the template (Barton et al., 2006). Additionally, crystal structure of the membrane proximal three fibronectin type III (FNIII) domains of Tie2 (PDB ID: 5UTK, 438-741 aa) and Tie2 kinase domain (PDB: 1FVR, 798-1124 aa) were used as templates for respective missense variants (Moore et al., 2017) (Shewchuk et al., 2000).

The full-length human TEK cDNA (reference sequence: NM_000459.5, 4,683 bp) with 3x Flag (DYKDDDDK) tag was cloned into the pCDNA3.1-ZsGreen vector (Supplementary Figure S1A). Site-directed mutagenesis was performed to introduce mutations into TEK-expressing vectors. Specifically, two plasmids (Y1024X-Flag and TEK1023-Flag, Supplementary Figures S1B, C) were generated to verify the nonsense mutation p.Y1024*. All plasmids were validated by Sanger sequencing.

Wild-type (WT) or mutant TEK-Flag plasmids (1 μg) were transfected into subconfluent 293T cells in six-well dishes using Lipofectamine 2000 (Life Technologies, Carlsbad, CA). Twenty-four hours later, cells were lysed by 1% Triton X-100, centrifuged, and separated into supernatant and insoluble pellets. Radioimmunoprecipitation assay (RIPA) buffer was added to the pellets before sonication. Same amount of proteins (15 μg) was separated by electrophoresis and transferred to polyvinylidene fluoride (PVDF) membranes. After blocking with 5% non-fat milk, antibodies against phosphotyrosine (1:3,000; ab179530, Abcam, Cambridge, United Kingdom) and Flag (1:5,000; 20543-1-AP, Proteintech, Rosemont, IL, United States) were used to detect the phosphorylation and expression of Flag-tagged Tie2, respectively. β-Actin (1:5,000; 66009-1-Ig, Proteintech, Rosemont, IL, United States) was used as a loading control. For proteasomal inhibition assay, transfected 293T cells were treated with 5 μM MG132 (MedChemExpress, Shanghai, China) for 24 h before harvest.

Total RNA was extracted from 293T cells using EZ-press RNA Purification Kit (EZBioscience, Roseville, MN, United States) 24 h after transfection. One thousand nanograms of RNA was used for reverse transcription and subsequent quantitative PCR in CFX96TM real-time system (Bio-Rad, Hercules, CA, United States). The housekeeping gene ACTB was used as the internal control. All mRNA variants were normalized to WT-TEK using the δ-delta Ct method. The primers are listed below: forward primer, 5′-TCC​AGG​CAA​CTT​GAC​TTC​GG-3; reverse primer, 5′-CCT​TGA​ACC​TTG​TAA​CGG​ATA​G-3.

Human umbilical vein endothelial cells (HUVECs) were obtained from ScienCell and cultured using endothelial cell medium (ScienCell, Carlsbad, CA, United States). Cells were transfected with either WT or mutant TEK-Flag plasmids by electroporation with Neon transfection system (Thermo Fisher, MA, United States) following the manufacturer’s instructions. The transfected cells were seeded on glass coverslips in 24-well dishes. After 48 h, cells were either directly fixed with 4% paraformaldehyde for 15 min at room temperature or stimulated with 600 ng/ml recombinant human Angiopoietin 1 (R&D systems, Minneapolis, MN, United States) for another 30 min at 37°C before fixation. Cells were then permeabilized and blocked by 5% goat serum and 0.3% TritonX-100 in phosphate-buffered saline (PBS) for an hour at room temperature and incubated overnight at 4°C with anti-Flag antibody (1:200; 66008-3-Ig, Proteintech, Rosemont, IL, United States) and anti-ZO1 antibody (1:500; 21773-1-AP, Proteintech, Rosemont, IL, United States). Cyanine3- and Cyanine5-conjugated secondary antibodies (Thermo Fisher, MA, United States) were used to detect Flag and ZO-1 signals, respectively. Cell nuclei were stained by 4,6-diamidino-2-phenylindole (DAPI) (1 μg/ml; Sigma-Aldrich, St. Louis, MO, United States) for 5 min. Images were taken with a fluorescence microscope (DM4000 B, Leica, Wetzlar, Germany).

Statistical analysis was performed with Stata software (15.1, StataCorp). Continuous variables were compared using Student’s t-test or paired t-test, while Pearson chi-square test was applied to comparing categorical variables. A p value <0.05 was considered statistically significant.

An average read depth of 1,938.3 was achieved in this study, while 99.3% of the targeted regions had 100x coverage per base. In total, 44 TEK variants were detected in 199 samples. Among these, 11 rare or novel heterozygous nonsynonymous variants were identified in 11 unrelated patients (Table 1 and Supplementary Figure S2). Four variants (p.R1003H, p.H52R, p.H494Y, and p.Y1024*) co-segregated with the glaucoma phenotype. Other six variants were inherited from unaffected parents, consistent with the autosomal dominant pattern with decreased penetrance and variable expressivity suggested by Souma and others (Souma et al., 2016).

One mutation (p.A841V) was previously identified in a large American pedigree by Young et al. (2020). In addition, a different amino acid change in proline residue 244 (p.P244R) was also reported in the same study. The nonsense mutation (p.Y1024*) was confirmed by Western blotting (Supplementary Figure S1D). The truncated protein lacks tyrosine residues for phosphorylation in C-terminus (e.g., 1,102 and 1,108), which are indispensable in mediating cellular signals. Therefore, it was reasonable to speculate that these variants were detrimental to protein function.

The remaining eight missense variants differed in pathogenicity prediction. Half of them (p.C264F, p.L504P, p.L888P, and p.R1003H) involved evolutionarily conserved residues (Supplementary Figure S3) and evaluated as destabilizing by FoldX (Table 1), whereas the other four variants (p.H52R, p.M131I, p.M228V, and p.H494Y) were predicted to be benign by most in silico tools.

Since family 01-07 were previously confirmed negative for CYP1B1 and LTBP2 mutations, probands from family 08-11 were exome sequenced to explore mutations in other known disease-causing genes. C-354 was found to inherit the nonsense mutation of TEK (p.Y1024*) and a missense variant of CYP1B1 (p.L107V) from his affected father. C-350 carried an additional missense variant of LTBP2 (p.T886K) from her mother compared to her unaffected father. No rare or novel variants of MYOC, FOXC1, and ANGPT1 were detected (Supplementary Table S2).

In this study, 5.5% of PCG patients carried rare or novel variants of TEK (Table 2). Of the carriers, 13 out of 19 (68.4%) exhibited clinically identifiable glaucoma phenotypes starting from 9.5 months after birth on average. There appeared to be a gender predilection, as most cases were male. In addition, more patients were bilaterally affected. Eight patients were followed up for disease prognosis (Table 3). Most of them (5/8) remained stable after initial trabeculotomy. Two patients underwent secondary angle surgeries, and one was enucleated for ocular components.

There were no statistically significant difference between TEK-mutated PCG groups with regard to clinical features listed in Table 2. Similarly, the clinical presentations were comparable between Chinese patients with TEK mutations and those with CYP1B1 mutations or unidentified causes (Table 4).

All PCG-related TEK missense variants but p.A841V were tested in cell-based assays to evaluate their impact on protein expression, solubility, auto-phosphorylation, and response to ligand stimulation. First, WT and mutant TEK plasmids were overexpressed in 293T cells. The transcription levels of missense variants were all comparable to WT (Supplementary Figure S4). Three variants (p.P244S, p.C264F, and p.L504P) located in the ectodomain of Tie2 led to a dramatic reduction in protein expression in both soluble and insoluble fractions (Figure 1). After inhibition of proteosomal activity by MG132, the insoluble fraction of p.P244S and p.C264F mutant was remarkably increased, suggesting significant protein degradation, whereas p.L504P was insensitive to MG132 treatment. P244 together with P243 is critical in the structure of a β-turn (Figure 2). Substitution of proline (non-polar) to a serine (polar) is likely to change protein conformation. As C264 forms a disulfide bond with C255, we inferred that the p.C264F variant might lead to the disruption of the disulfide bridge and misfolding of translation product. L504 locates in the E β-strand of fibronectin type IIIa (FNIIIa) domain, which contributes to a classical antiparallel β-sheet motif (Moore et al., 2017). The replacement of leucine by proline disrupted intra-strand hydrogen bond resulting in an energetically unstable structure.

Both protein expression and auto-phosphorylation levels of the other four variants in the ectodomain (p.H52R, p.M131I, p.M228V, and p.H494Y) were comparable to that of WT (Supplementary Figure S5), consistent with the in silico prediction of pathogenicity and structural analysis (Supplementary Figure S6). Notably, p.R1003H mutant was overexpressed in the insoluble component with increased auto-phosphorylation in the soluble fraction, matching the features of activating mutations (Figure 3). Structural analysis indicated that the de novo variant p.R1003H was likely to cause the disruption a hydrogen bond connecting R1003 with M1024 (Figure 2). However, variant p.L888P could not be distinguished from WT in our cell-based assay (Figure 3).

The ectodomain of Tie-2 consists of three Ig-like domains and an EGF-like domain, which are responsible for ligand binding (Yu et al., 2013), and a FNIII domain that is involved in ligand-induced receptor multimerization (Moore et al., 2017). Variants in these regions have the potential to weaken or disrupt ligand-mediated signaling. Thus, we transiently expressed all variants in HUVECs and explored their response to stimulation by potent agonist angiopoeitin-1 (ANGPT1). As anticipated, the WT TEK was diffusely expressed on the cytomembrane of HUVECs and relocalized to cell–cell junctions when treated with ANGPT1 (Figure 4). Surprisingly, the expression of variants p.P244S, p.C264F, and p.L504P was greatly reduced, consistent with the results of Western blotting. However, the other seven mutants did not exhibit significant alteration in response to ligand stimulation (Figure 4). A summary of functional analysis results is provided in Supplementary Table S3.