Systematic Ocular Phenotyping of Knockout Mouse Lines Identifies Genes Associated With Age-Related Corneal Dystrophies

Abstract

Purpose: This study investigates genes contributing to late-adult corneal dystrophies (LACDs) in aged mice, with potential implications for late-onset corneal dystrophies (CDs) in humans.

Methods: The International Mouse Phenotyping Consortium (IMPC) database, containing data from 8901 knockout mouse lines, was filtered to include late-adult mice (49+ weeks) with significant (P < 0.0001) CD phenotypes. Candidate genes were mapped to human orthologs using the Mouse Genome Informatics group, with expression analyzed via PLAE and a literature review for prior CD associations. Comparative analyses of LACD genes from IMPC and established human CD genes from IC3D included protein interactions (STRING), biological processes (PANTHER), and molecular pathways (KEGG).

Results: Analysis identified 14 genes linked to late-adult abnormal corneal phenotypes. Of these, 2 genes were previously associated with CDs in humans, while 12 were novel. Seven of the 14 genes (50%) were expressed in the human cornea based on single-cell transcriptomics. Protein–protein interactions via STRING showed several significant interactions with known human CD genes. PANTHER analysis identified six biological processes shared with established human CD genes. Two genes (Rgs2 and Galnt9) were involved in pathways related to human corneal diseases, including cGMP-PKG signaling, mucin-type O-glycan biosynthesis, and oxytocin signaling. Other candidates were implicated in pathways such as pluripotency of stem cells, MAPK signaling, WNT signaling, actin cytoskeleton regulation, and cellular senescence.

Conclusions: This study identified 14 genes linked to LACD in knockout mice, 12 of which are novel in corneal biology. These genes may serve as potential therapeutic targets for treating corneal diseases in aging human populations.

The human cornea plays a pivotal role in visual acuity and ocular protection.¹ However, it is subject to a spectrum of disorders known as corneal dysmorphologies (CDs), which describe a collection of multifactorial eye disorders that result in progressive vision loss.² These can be categorized into two broad groups: regressive degeneration corneal dystrophies and corneal dysplasias, driven by defective growth and differentiation. CD may present at birth or develop insidiously during various stages of life, resulting in clinical manifestations that range from subtle to debilitating vision loss, pain, photophobia, or foreign body sensations.³ 

The cornea is composed of five layers, each with a unique function in maintaining its integrity and transparency. These layers include the epithelium, Bowman's layer, the stroma, Descemet's membrane, and the endothelium. CD occurs with degradation or accumulation of material within one or more of these layers.⁴ The IC3D classification system, established by the International Committee for the Classification of Corneal Dystrophies (IC3D), proposes seven major categories of CD: Epithelial and Subepithelial Dystrophies, Bowman Layer Dystrophies, Stromal Dystrophies, Descemet Membrane Dystrophies, Endothelial Dystrophies, Unspecified Stromal Dystrophies, and Miscellaneous Dystrophies.⁵ The IC3D has identified and published CD genes within this classification system.⁵ 

The use of knockout mice offers a potent strategy for investigating genes associated with CD. The International Mouse Phenotyping Consortium (IMPC) is a global initiative of 21 centers that produces and phenotypes knockout mice for research purposes. These knockout mice undergo a standardized phenotyping pipeline, where a wide range of traits, including ocular phenotypes, are rigorously assessed.⁶ 

The IMPC utilizes specialized pipelines to assess phenotypes at various developmental stages, encompassing both early-adult and late-adult evaluations. The early-adult pipeline concentrates on the initial phases of mouse development, typically examining mice between 9 to 15 weeks old.⁶ This phase delves into the immediate effects of gene knockouts on traits such as growth, organ development, and behavior manifesting during early life stages. The late-adult pipeline, a subset of the early-adult pipeline, is dedicated to the assessment of phenotypes in mice 49 weeks and older.⁶ It provides insights into the long-term consequences of gene disruptions in mice, shedding light on age-related phenotypic changes, diseases, and characteristics not apparent during earlier stages, with potential implications for age-related CD development in humans. 

In recent years, research has increasingly highlighted various networks of signaling pathways and biological processes crucial for ocular health and corneal disease. For instance, the cyclic guanosine monophosphate (cGMP) pathway regulates collagen synthesis in the eye, while mucin glycoproteins, particularly O-glycans, protect the cornea and conjunctiva from physical, chemical, and microbial damage.^7,8 Oxytocin is essential for maintaining ocular surface homeostasis, and limbal stem cell regulation is critical for corneal epithelial regeneration.^9,10 The MAPK signaling pathway is involved in corneal wound healing and dry eye disease, WNT signaling influences corneal epithelial stratification, regulation of actin cytoskeleton controls tight junction permeability in the cornea, and cellular senescence facilitates the turnover of damaged epithelial cells.^11–14 This study aims to discover additional genes relevant to corneal biology in the aging population through an unbiased screening of systematically phenotyped knockout mouse lines. 

Materials and Methods

Animals and Phenotyping

The IMPC knockout process involves the disruption of protein-coding genes within the mouse genome, followed by rigorous genetic quality control assessment of the mutant mouse lines. Once the genetic quality is confirmed, the consortium generates cohorts of at least seven female and seven male mice for each mutant line. These mice are then subjected to thorough phenotyping, conducted in parallel with age- and sex-matched wild-type (WT) control mice, which are also produced at the same specialized production center.⁶ 

The IMPC employs two methods to produce their mouse lines, CRISPR/Cas9 editing and embryonic stem cell–derived mice, both on the C57BL/6N strain background.^15,16 The phenotypes they identify are systematically described using standardized mammalian phenotyping ontology terms developed by the Mouse Genome Informatics group (MGI).¹⁷ The zygosity of the mutant lines is also determined at this time, distinguishing between homozygous (HOM), heterozygous (HET), and hemizygous (HEM) conditions.⁶ For further information and to access their comprehensive work, visit the IMPC website at http://www.mousephenotype.org. 

This study analyzed Data Release 21.0, which was released on May 7, 2024, and queried on May 10, 2024. In this release, IMPC phenotyped a total of 8901 unique genes, encompassing 9594 mutant lines, and identified 106,561 phenotype hits with a significance level of P < 0.0001.¹⁸ 

All procedures carried out at IMPC centers comply with strict local, state, and national regulatory guidelines and uphold the principles outlined in the Animal Research: Reporting of In Vivo Experiments guidelines, which aim to standardize and enhance the quality and reproducibility of animal research. Additionally, a Housing and Husbandry protocol is followed, encompassing a set of both mandatory and optional procedures that guide international mouse experimentation.^15,16 Furthermore, the consortium ensures that all procedures involving live animals are reviewed and approved by associated institutional animal care and use committees or their equivalent entities, underpinning their commitment to the highest standards of animal welfare. 

Bioinformatics

The exploration of late-adult CD (LACD) phenotypes involved a methodical assessment of genes within the IMPC's online data set to validate the presence of corneal abnormalities. This process entailed querying the IMPC using the term cornea in the phenotype search and filtering to focus only on late-adult mice with a significant (P < 0.0001) CD phenotype. Focusing on late-adult abnormalities, mouse lines with corneal phenotypes in early adulthood (age <16 weeks) were excluded. Genes linked to LACD phenotypes were manually curated to exclude possible false positives and then subjected to a comprehensive literature review, investigating documented mouse models and their associated corneal phenotypic anomalies in both humans and mice. Additionally, human orthologs of all candidate LACD genes underwent analysis of corneal gene expression and predicted protein-protein interactions and functional pathways, both within the set of candidate genes and against a set of 48 previously established human CD genes, identified from the IC3D.¹⁹ This analysis was conducted using established bioinformatics tools: Platform for Analysis of Sceiad (PLAE) was used to identify gene expression of human orthologs in the cornea, the Search Tool for the Retrieval of interacting Genes/Proteins (STRING) within the Cytoscape platform (version 3.10.2) was used to analyze protein–protein interactions, Protein Analysis THrough Evolutionary Relationships (PANTHER) was used to compare and contrast biological processes, and Kyoto Encyclopedia of Genes and Genomes (KEGG) within the Database for Annotation, Visualization, and Integrated Discovery website (DAVID) was used to assess whether candidate and established human CD genes were known to be involved in cellular pathways or signaling cascades.^20–25 

STRING was used to analyze protein–protein interactions within the candidate LACD gene set and protein–protein interactions, including the 48 established human CD genes. STRING allows for the inclusion of a specified number of additional interacting proteins into the specified protein query set. Additional queries were conducted with 10 additional interactor proteins in the analysis within the candidate gene set, as well as between the candidate LACD genes and established human CD genes. Another query was conducted to analyze protein interaction between candidate CD genes and CD genes known to be involved in early-adult CD mice only.¹⁹ The STRING database supplied confidence scores to facilitate comparative assessments of gene interactions, adhering to the thresholds recommended by the database designers: 0.15 to 0.4 for low confidence, 0.4 to 0.7 for medium confidence, 0.7 to 0.9 for high confidence, and >0.9 for the highest level of confidence in alignment with the study's methodology.²¹ Queries were run at all confidence levels, but only interactions found to be medium or higher were investigated further. 

Candidate LACD genes found to be expressed in the human cornea via PLAE were analyzed individually in STRING at the highest confidence level with 10 additional interactor proteins to establish potential functional networks. 

PANTHER biological processes for candidate LACD and established human CD genes were assessed separately and then compared. The gene sets were analyzed by entering corresponding STRING IDs of each gene in the gene set with settings, list type: ID list, organism: Homo sapiens, analysis: Functional classification viewed in gene list. Resulting lists were manually inspected and confirmed. Individual biological processes were recorded from the gene list, and bar graphs depicting biological process categories of the full gene set were generated. 

KEGG pathways were assessed for the candidate CD gene set and the established CD gene set by manual curation on the KEGG website, generating a comprehensive list of all associated pathways. Both gene sets were then processed through DAVID, generating a list of pathways found to be significant based on the provided inputs. Candidate CD genes, established human CD genes, and additional interactor protein genes were then annotated on significant pathways. 

Results

Of the 8901 IMPC knockout lines, 587 (7%) were evaluated in late-adult pipelines. Of those in the late-adult pipeline, 14 (2.4%) presented LACD phenotypes, which were not detected at the early-adult stage. The 14 identified LACD genes (Abca16, Abhd17b, Fsd2, Galnt9, Gtpbp10, Ik, Krt80, Rgs2, Scamp2, Slc30a7, Sprr1a, Tenm4, Trim39, Vwa5a) were characterized by ontology terms, including corneal opacity (8, 57%), increased corneal thickness (1, 7%), abnormal corneal morphology (4, 29%), and sclerocornea (1, 7%). Each of the candidate LACD genes exhibited a single corneal phenotype. Five (36%) of the 14 candidate genes had late-stage CD phenotype images available, and Figure 1 illustrates examples of these CD phenotypes compared to the wild-type cornea. 

Figure 1.

 

External color photography of corneas from late-stage knockout mice with documented cornea abnormalities. Top row: WT, Abhd17b^−/−, Ik^+/−. Bottom row: Slc30a7^+/−. Sprr1a^−/−, Tenm4^−/−.

A total of 13 (93%) of the 14 candidate gene lines were homozygous knockouts, and 1 (7%) was a heterozygous knockout due to embryonic lethality in homozygotes. Moreover, six (42%) of the lines had CD phenotypes in both sexes, while the remaining eight (58%) candidate gene phenotypes achieved statistical significance in only one sex (sexual dimorphism), with six occurring in females and two in males. Nine of the gene lines (64%) displayed bilateral CD in the majority (≥50%) of mice with abnormal cornea phenotypes (Abca16, Fsd2, Galnt9, Gtpbp10, Krt80, Rgs2, Scamp2, Tenm4, Vwa5a), two gene lines (14%) displayed bilateral CD in the minority (<50%) of mice with abnormal corneal phenotypes (Ik and Sprr1a), two gene lines had no mice with bilateral CD (Abhd17b and Trim39), and one gene line did not have laterality data available (Slc30a7). For a comprehensive list of candidate LACD genes, along with their phenotypes, zygosity, sex dependence, bilaterality, and tissue expression, see Table 1. 

Table 1.

 

List of 14 Candidate LACD Genes With Human Orthologs, Gene Location, Full Gene Name, Associated Corneal Phenotypes, Zygosity, Gender Specificity, Life Stage, Phenotyping Center, P Value, Human Cornea Expression, and Previous Cornea Publication PMID

 

Two genes (Fsd2 and Scamp2) found to have LACD in late adulthood exhibited other abnormal ocular phenotypes in early adulthood. Specifically, Fsd2 showed an early-adult “abnormal eye morphology” phenotype, and Scamp2 presented early-adult “cataracts.” 

A thorough literature search revealed two of the candidate genes (Krt80 and Sprr1A) had been previously associated with an existing CD, specifically keratoconus (KCN).^26,27 The remaining 12 candidate LACD genes were novel, given that they had no prior literature associating them with CDs (Abca16, Abhd17b, Fsd2, Galnt9, Gtpbp10, Ik, Rgs2, Scamp2, Slc30a7, Tenm4, Trim39, Vwa5a). 

Seven of the 14 LACD genes (Gtpbp10, Ik, Rgs2, Scamp2, Slc30a7, Tenm4, Vwa5a) (50%) were found expressed in single-cell transcriptomic data sets from the human cornea using PLAE. For comparison, 34 of the 48 (71%) established human CD genes studied in a recent publication were found expressed using PLAE19. These genes were expressed in multiple corneal cell types; GTPBP10 is expressed most in limbal progenitor cells along with 12 additional cell types, Ik is expressed most in T/natural killer cells along with 18 additional cell types, RGS2 is expressed most in corneal endothelial cells along with 9 additional cell types, SCAMP2 is expressed most in conjunctival epithelium along with 17 additional cell types, SLC30A7 is expressed most in T/natural killer cells along with 17 additional cell types, TENM4 is expressed most in corneal progenitor cells along with 15 additional cell types, and VWA5A is expressed most in conjunctival epithelial cells along with 12 additional cell types. For a comprehensive list of candidate LACD gene expression, see Table 2. 

Table 2.

 

List of Seven LACD Genes Expressed in Human Corneal Tissue on PLAE With Cell Type, Cell Expression Count, Cell Count, Percent Expression, and Overall Expression

 

STRING protein analysis within Cytoscape, excluding Abca16 (since it has no human ortholog) and MIR184 (microRNA) from the candidate and established CD gene sets, respectively, found no protein interactions within the candidate gene set (data not shown). When including 10 additional interactor proteins in the candidate gene analysis, four functional clusters emerged in the highest confidence interval (>0.9) containing four candidate genes (Ik, Rgs2, Sprr1a, Tenm4), as seen in Supplemental Figure S1. Analysis at the high confidence interval included the gene Trim39 into one of the four clusters, while analysis at the moderate confidence interval resulted in the inclusion of three more candidate genes (Fsd2, Krt80, Slc30a7) into five total clusters. 

Protein interactions between candidate LACD genes and established human CD genes revealed no protein clusters involving candidate genes at the highest or high confidence level. One protein cluster containing a candidate gene (Krt80) emerged in the moderate confidence interval (Supplemental Fig. S2). When including 10 additional interactor proteins in the candidate and established CD gene analysis, one protein cluster containing three candidate genes (Krt80, Rgs2, Trim39) emerged in the moderate confidence interval (Fig. 2). 

Figure 2.

 

STRING protein–protein analysis between human ortholog proteins of 13 candidate LACD genes (red), 47 established human CD proteins (gold), and 10 additional interactor proteins determined by STRING (green). Candidate LACD gene Abca16 and established human CD gene MIR-184 were omitted from this analysis as they are not available in STRING. Analysis run with modified settings (Organism: Homo Sapiens; Network Type = full STRING network; Confidence cutoff 0.40; Additional interactors 10). Darker edges indicate stronger protein–protein interaction.

Protein interactions between candidate LACD genes and genes that resulted in early-adult CD via the IMPC¹⁹ revealed one cluster containing one candidate gene (Ik) in the highest and high confidence interval. Three clusters emerged containing three additional candidate genes (Rgs2, Slc30a7, Sprr1a) at the moderate confidence interval, as seen in Supplemental Figure S3. A detailed list of genes with corresponding protein–protein confidence levels can be found in Table 3. 

Protein interactions of each of the seven LACD genes expressed in the human cornea returned four genes with protein networks (Gtpbp10, Ik, Rgs2, Tenm4), while three genes (Scamp2, Slc30a7, Vwa5a) did not have any protein networks at the >0.90 confidence level. Gtpbp10 had 10 protein interactions, including mitochondrial ribosomal assembly proteins; Ik had 10 protein interactions, including pre-mRNA splicing proteins; Rgs2 had 5 protein interactions, including G-protein signaling proteins; and Tenm4 had 3 protein interactions, including cell–cell adhesion proteins, as seen in Supplemental Figure S4. 

In PANTHER biological process analysis, 13 of the 14 candidate LACD genes (excluding Abca16) were mapped to 6 biological process categories, and 47 of the 48 established human CD genes (excluding MIR-184) were mapped to 11 biological process categories, as seen in Figure 3. All six biological processes categories were shared between the two gene sets (biological regulation, cellular process, developmental process, localization, metabolic process, and multicellular organismal process), while the established human CD genes had five unique biological process categories (homeostatic process, immune system process, locomotion, pigmentation, and response to stimulus). A detailed list associating candidate and established genes with specific biological processes can be found in Table 4. 

Figure 3.

 

Biological process categories of 13 candidate LACD genes (top) and 47 established human CD genes (bottom) using PANTHER analysis. Candidate LACD gene Abca16 and established human CD gene MIR-184 were not included in the analysis as they were not available on PANTHER. Gold star depicts biological processes found in both candidate LACD genes and established CD genes.

Table 3.

 

Confidence Ranking of STRING Protein-Protein Interactions Among Candidate LACD Genes, and Between Candidate LACD Genes and Established Human CD Genes, Early Adult-Stage CD Genes, and 10 Additional STRING Interactor Proteins.

 

KEGG pathway mapping of the 14 candidate LACD genes, the established human CD genes, and the 10 additional interactor genes was combined, resulting in three pathways containing candidate LACD genes: cGMP-PKG signaling, oxytocin signaling, and mucin-type O-glycan synthesis. The cGMP signaling pathway (Fig. 4) contains one candidate LACD gene (Rgs2), one established gene (Cna1), and two interactor genes (Ins and Akt1). The oxytocin signaling pathway (Supplemental Fig. S4) includes LACD gene Rgs2 and two interactors (Egfr and Actb). The mucin-type O-glycan biosynthesis pathway (Supplemental Fig. S5) includes LACD gene Galnt9. Five additional pathways did not contain candidate LACD genes but contained established CD genes or additional interactor genes found to have protein–protein interactions with the candidate LACD gene set, MAPK signaling, WNT signaling, signaling pathways regulating pluripotency of stem cells, regulation of actin cytoskeleton, and cellular senescence. These can be viewed in Supplemental Figures S6 to S10. A full list of KEGG pathways for candidate and established CD genes can be found in Table 5. 

Figure 4.

 

cGMP-PKG signaling pathway highlighting candidate LACD gene Rgs2 (red star), established CD genes CNA1 (gold star), and two additional STRING interactor genes INS and AKT1 (green star).

Table 4.

 

List of PANTHER Biological Processes Associated With Candidate LACD Genes (Left) and Established Human CD Genes (Right)

 

Table 5.

 

List of KEGG Pathways for Candidate LACD Genes and Established CD Genes

 

Discussion

In this study, we identified 14 mammalian genes required for corneal clarity in the late-adult phase (Abca16, Abhd17b, Fsd2, Galnt9, Gtpbp10, Ik, Krt80, Rgs2, Scamp2, Slc30a7, Sprr1a, Tenm4, Trim39, Vwa5a). Most of these genes, 12 of 14 (bold above), have no reported functional roles in corneal biology. While half of the 14 candidate genes are expressed in human cornea, most of them do not have obvious bioinformatic relationships with established CD genes and may have biological functions that are not well understood. 

The seven candidate genes expressed in human cornea tissue presented in multiple cornea cell types with multiple cell types shared among the genes. Given this overlap in expression, a direct relationship of cell type expressed to observed phenotype is not easily derived. Furthermore, the protein networks generated from the expressed genes do not present a clear relationship of protein network to observed phenotype. 

Examining the functions of these genes outside the cornea provides insight into their potential roles within it. Gtpbp10, a GTP-binding protein, aids in mitochondrial ribosomal RNA folding.²⁸ This function is supported by its high-confidence interactions with mitochondrial ribosomal assembly proteins in STRING, suggesting a similar role in the aged cornea, particularly in limbal progenitor cells, where its expression is highest. Ik, a cytokine, inhibits interferon-gamma–induced major histocompatibility complex class II expression and is a spliceosome component in noncorneal tissues.^29,30 STRING analysis revealed interactions only with other spliceosome proteins. However, its highest expression in corneal T/natural killer cells suggests its immune-modulating functions may contribute to the abnormal corneal morphology observed. Scamp2, a secretory carrier membrane protein, facilitates post-Golgi recycling and regulates cell surface T-type calcium channels in noncorneal tissues.^31,32 STRING did not reveal high-confidence protein interactions, and its highest expression in conjunctival epithelial cells does not indicate a clear corneal function in this analysis. Slc30a7, a zinc transporter, enables cellular zinc efflux and has demonstrated antioxidant effects in high-glucose apoptosis outside the cornea.^33,34 It had no high-confidence interactions in STRING and was most highly expressed in T/natural killer cells. The observed LACD phenotype may stem from disruptions in zinc homeostasis, antioxidant effects, or an unknown mechanism. Tenm4, a teneurin transmembrane protein, promotes focal adhesion kinase activation and interacts with adhesion G protein-coupled receptors.³⁵ It is most highly expressed in corneal progenitor cells, suggesting that impaired cell–cell adhesion in these cells could disrupt corneal architecture, potentially leading to corneal opacity. Vwa5a, a von Willebrand factor A domain-containing protein, may function as a tumor suppressor outside the cornea.³⁶ Protein analysis provided little insight, but its highest expression in conjunctival epithelium raises the possibility that the observed corneal opacity may result from conjunctival overgrowth. 

Rgs2, a regulator of G-protein signaling, emerges as a noteworthy candidate LACD gene with multiple pathways and interactions that could explain its corneal opacity phenotype in knockout mouse lines. The cGMP-PKG pathway has been previously linked to regulation of collagen synthesis in multiple organs, including the eye, but has not yet been implicated in CD.⁷ Due to the importance of collagen in maintaining the integrity and physiological properties of the corneal stroma, it follows that a disruption of the collagen architecture in an aging cornea could compromise its translucent properties, resulting in corneal opacity. Given the direct involvement of Rgs2 in this cGMP-PKG pathway, the knockout of this gene may compromise the pathway enough to result in a CD. Furthermore, Rgs2 is directly involved in oxytocin signaling, a pathway implicated in dry eye disease (DED).⁹ It is known that DED can result in corneal epithelial opacity; therefore, Rgs2 knockout and subsequent phenotypic corneal opacity could also be explained through this relationship. Rgs2 also had indirect involvement in other signaling pathways such as MAPK, WNT, and regulation of pluripotency of stem cells signaling; however, these interactions were of lower confidence. It is unclear if the Rgs2 knockout and emergent corneal opacity phenotype is a result of disruption to one or more of the previously mentioned pathways. 

Several other candidate LACD genes were similarly found to result in corneal opacities. One such gene is Galnt9, polypeptide N-acetylgalactosaminyltransferase 9, found in the mucin-type O-glycan biosynthesis pathway. It is well established that mucin plays a vital role in protecting the cornea, and its disruption can result in epithelial damage.⁸ Therefore, it follows that disruption to the synthesis of the mucin layer would leave the cornea susceptible to damage, resulting in corneal opacity. Trim39, tripartite motif-containing 39, is another candidate LACD gene resulting in a corneal opacity phenotype. Disruption of the p53 pathway has been shown to affect the differentiation and mucin expression of corneal epithelial cells.³⁷ Therefore, Trim39’s moderate interaction with notable interactor gene Tp53 within the cellular senescence pathway could represent a link to corneal mucin production. Compromise in mucin expression and cell turnover, especially in the rapidly replicating corneal epithelial cells, could allow damaged cells to accumulate and disrupt the translucent nature of this tissue layer, resulting in corneal opacities. Both Galnt9 and Trim39’s phenotypic effects on corneal opacity highlight the importance that mucin expression and proper epithelial function have on the overall integrity and function of the cornea. 

The analysis of late-stage CD genes is incomplete due to the consortium's ability to include only ∼7% of IMPC knockout lines for aging phenotypes. Furthermore, our analysis does not span the whole mouse genome, as it is based on data from the IMPC, which includes 8901 protein-coding genes. Only a fraction of these genes underwent late-adult phenotyping. Of the genes labeled with cornea phenotypes, some false positives existed due to incomplete data entry at the IMPC data portal. Since the IMPC data set is dynamic with new data releases every 3 to 4 months, the nature of this research demands independent verification. This study is based on high-throughput ocular examination by expert graders who are masked to the genotype of mice. However, due to limitations in scope and funding, their findings were not always assessed by histopathology. Ik and Sprr1a displayed bilateral corneal phenotypes in less than 50% of mice. Abhd17b and Trim39 had no mice with bilateral phenotypes. However, these genes may be required for the ocular tear film, corneal epithelial integrity, or corneal wound healing. Therefore, these corneas may be less durable in response to an environmental stressor (such as mild trauma or bacterial seeding) and thus vulnerable to injury, infection, or fibrotic wound response, explaining how a germline deletion can lead to unilateral phenotypes. Furthermore, the limited IMPC ontology terms and thin nature of the mouse cornea limit precise identification of the location within the cornea where the observed CD occurs. This, along with the difference in corneal structure between mice and humans, impedes the direct correlation between mouse corneal phenotypes and clinical CDs. This indirect relationship could contribute to the identification of novel human cornea genes, as some mouse cornea genes may not play a role in the human cornea. Narrowing the focus to candidate LACD genes expressed in human cornea tissues derives greater clinical implication from this study. Future analysis should examine precise layers within the cornea so relationships can be more accurately linked to disrupted pathways. 

Conclusions

In this investigation, 14 genes associated with LACD were identified from a pool of 587 IMPC knockout mouse lines that went through late-adult phenotyping, 12 of which are only now implicated in corneal biology and 7 are found to be expressed in the human cornea. These genes may represent underappreciated biological processes important for the aging cornea. Further studies may prove them to be potential therapeutic targets for preventing or treating corneal diseases in aging human populations. 

Acknowledgments

The authors thank all the various funding agencies supporting the International Mouse Phenotyping Consortium and all scientists at each of the production and phenotyping centers. The authors gratefully acknowledge their funding sources, including the Government of Canada through Genome Canada/Ontario Genomics OGI-051 (CM); NIH R03OD032622 (KCKL and AM) and K08EY027463 (AM); NIH U54HG006364, U42OD011175, 5UM1OD02322, and UM1OD023321 (KCKL and CM); Infrafrontier grant 01KX1012; EU Horizon2020: IPAD-MD funding 653961 (MHdA); NCATS UL1 TR001860 and TL1 TR001861 (BY); and MC_UP_2201/1 (SC, RJ, MS, and SW). 

Work of CCP was supported by the Czech Academy of Sciences RVO 68378050 and by the project LM2023036 Czech Centre for Phenogenomics provided by Ministry of Education, Youth and Sports of the Czech Republic. 

The following authors are IMPC members who contributed to this study: Elif Acar, Antonio Aguilar-Pimentel, Elizabeth Axton, Shinya Ayabe, Abdel Ayadi, Lore Becker, Alexandr Bezginov, Marie-Christine Birling, Jin Woong Bok, Raphaël Bour, Vivian Bradaschia, Julia Calzada-Wack, Adam Caulder, Linda Chan, Dave Clary, James Cleak, Gemma Codner, Patricia da Silva Buttkus, Nathalia Dragano, Kyle Duffin, Qing Fan-Lan, Martin Fray, Tamio Furuse, Xiang Gao, Wendy Gardiner, Lillian Garrett, Marina Gertsenstein, Isabelle Goncalves, Leslie Goodwin, Kristin Nicole Grimsrud, Alain Guimond, Sabine Hölter-Koch, Joanne H. Hsu, Mizuho Iwama, Lois Kelsey, Chang-Hoon Kim, Kyoungmi Kim, Markus Kraiger, Tatsuya Kushida, Valerie Laurin, Sophie Leblanc, Ho Lee, Christoph Lengger, Stefanie Leuchtenberger, Lauri G. Lintott, Jimmy Liu, Yang Liu, Aline Lux, Susan Marschall, Matthew McKay, Matthew McKenzie, David Miller, Christophe Mittelhaeuser, Ikuo Miura, Saori Mizuno, Toshiaki Nakashiba, Ki Taek Nam, Clare Norris, Yuichi Obata, Manuela Österreicher, Kristina Palmer, Guillaume Pavlovic, Kevin Peterson, Benoit Petit-Demouliere, Dawei Qu, Birgit Rathkolb, Kyle Roberton, Adrian Sanz Moreno, Claudia Seisenberger, Audrie Seluke, Xueyuan Shang, Hirotoshi Shibuya, Gillian Sleep, Tania Sorg, Nadine Spielmann, Claudia Stöger, Toyoyuki Takada, Nobuhiko Tanaka, Lydia Teboul, Todd Tolentino, Igor Vukobradovic, Hongyu Wang, Brandon Willis, Joshua Wood, Xia Xue Catherine Xu, Ikuko Yamada, and Jing Zhao. 

Author Contributions: AB queried, analyzed, and interpreted the mouse and human genomic data and drafted the manuscript; PV helped gather data; BY helped analyze the data and review manuscript; AM designed and oversaw the project, helped analyze the data, and helped write the manuscript; All other authors provided the mouse data; All authors read and approved the final manuscript. 

Disclosure: A. Briere, None; P. Vo, None; B. Yang, None; D. Adams, None; T. Amano, None; O. Amarie, None; Z. Berberovic, None; L. Bower, None; S.D.M. Brown, None; S. Burrill, None; S.Y. Cho, None; S. Clementson-Mobbs, None; A. D'souza, None; M. Eskandarian, None; A.M. Flenniken, None; H. Fuchs, None; V. Gailus-Durner, None; Y. Hérault, None; M. Hrabe de Angelis, None; S. Jin, None; R. Joynson, None; Y.K. Kang, None; H. Kim, None; H. Masuya, None; H. Meziane, None; K.-H. Nam, None; H. Noh, None; L.M.J. Nutter, None; M. Palkova, None; J. Prochazka, None; M.J. Raishbrook, None; F. Riet, None; J. Salazar, None; R. Sedlacek, None; M. Selloum, None; K.Y. Seo, None; J.K. Seong, None; H.-S. Shin, None; T. Shiroishi, None; M. Stewart, None; K. Svenson, None; M. Tamura, None; H. Tolentino, None; S. Wells, None; W. Wurst, None; A. Yoshiki, None; L. Lanoue, None; K.C.K. Lloyd, None; B.C. Leonard, None; M.J. Roux, None; C. McKerlie, None; A. Moshiri, None