Insights Into the FOXE3 Transcriptional Network and Disease Mechanisms From the Investigation of a Regulatory Variant Driving Complex Microphthalmia

Abstract

Purpose

FOXE3 encodes a highly conserved, lens-enriched transcription factor essential for eye development. Biallelic mutations in FOXE3 are associated with a spectrum of ocular anomalies, ranging from congenital cataracts to complex microphthalmia (CM), with severity and penetrance correlating with genotype. This study aimed to investigate the regulatory landscape of FOXE3 and its contribution to CM.

Methods

In a patient with CM, a truncating FOXE3 variation (p.Cys240*) was identified alongside a second, trans-acting regulatory variant (rv: rs745674596G>A) located 3 kb upstream of FOXE3. To investigate its functional impact, mouse models were generated carrying either the rv or a frameshift (fs) mutation in homozygosity (Foxe3rv/rv, Foxe3fs/fs) or in compound heterozygosity (Foxe3rv/fs). Ocular phenotypes were characterized, and molecular analyses were conducted to assess FOXE3 expression and transcriptional regulation.

Results

Phenotypic severity followed a progressive pattern from Foxe3rv/rv to Foxe3rv/fs, with Foxe3fs/fs consistently exhibiting CM, mirroring genotype-dependent effects observed in humans. Protein levels, but not mRNA levels, correlated with ocular phenotype, with the frameshift mutation leading to pronounced mRNA overexpression in embryos. In Foxe3fs/fs mice, CM resulted from early anterior lens epithelium disorganization, triggering progressive lens degeneration and ocular involution. Transcription factor binding studies identified USF2 as a key regulator of FOXE3 expression, positioning it as a novel candidate in ocular development and disease.

Conclusions

This study highlights the critical role of regulatory variants in ocular pathology, proposes a potentially novel mechanism for microphthalmia through lens degeneration, and identifies USF2 as a potential contributor to the FOXE3-regulatory network that remains largely unknown.

Vertebrate eye development relies on the precise regulation of gene networks and coordinated interactions among tissues of diverse embryonic origins, which are crucial for forming and patterning fundamental ocular structures.¹ The lens, a critical component in this process, requires accurate gene expression for proper development.^1,2 Multiple transcription factors and signaling pathways orchestrate lens development, including cell proliferation, differentiation, and patterning.³

FOXE3, a single-exon gene encoding a 319-amino-acid forkhead family protein, is crucial for lens development.⁴ It is an essential transcription factor expressed in the lens epithelium. In mice, its expression has been shown to be downstream of PAX6 and PITX3. FOXE3 plays a pivotal role in maintaining lens epithelial cell identity and regulating cell proliferation during lens development.^5,6 Pathogenic variants in FOXE3 are linked to various eye disorders, ranging from congenital cataracts and absence of the lens (aphakia) to anterior segment dysgeneses (ASD) and severe bilateral complex microphthalmia with lens anomalies. These mutations exhibit both autosomal dominant and recessive inheritance patterns, with recessive mutations often leading to severe conditions such as complex microphthalmia and dominant mutations resulting in milder phenotypes like cataracts and ASD.⁷

FOXE3 expression is regulated by several mechanisms, including transcription factors, signaling pathways, and epigenetic modifications involving nearby regulatory elements.^7,8 Despite the critical role of these mechanisms, the detailed regulatory architecture of FOXE3 locus and the impact of variants on these regulatory elements are not fully understood.

In this study, we identified a heterozygous pathogenic FOXE3 nonsense variant (p.Cys240*)^4,9,10 in a patient with a severe bilateral ocular phenotype, indicative of autosomal recessive inheritance and suggesting the presence of a second loss-of-function allele. Sanger sequencing of a conserved region located 3 kb upstream of FOXE3 revealed a noncoding variant at a heterozygous state. This region was previously implicated in mice exhibiting complex microphthalmia and cataract.¹¹ Mouse models harboring the noncoding variant, a truncating variation, or both demonstrated a genotype-dependent reduction in FOXE3 protein levels. While single heterozygotes exhibited normal ocular development, homozygosity for the noncoding or null allele, as well as compound heterozygosity, resulted in progressively severe ocular defects. Notably, Foxe3fs/fs mice initially developed normal eyes, but these later regressed to complex microphthalmia, underscoring FOXE3’s critical role in maintaining lens development and stability.

Further investigation revealed that the noncoding variant disrupts USF2 binding, and Usf2 knockdown experiments confirmed its role in downregulating Foxe3. These findings highlight the significance of identifying pathogenic noncoding variants in ocular disorders and provide valuable insights into the regulatory architecture of eye development, with genotype–phenotype correlation in partially resolved cases aiding in the discovery of such noncoding pathogenic variations.

Family, Materials and Methods

Clinical and Genetic Analysis

The Toulouse University Hospital (TUH) Data Protection Officer has ensured this study complies with France CNIL MR-004 standards for medical research and adheres to EU General Data Protection Regulation requirements. The study was included in both the TUH retrospective study registry (RNIPH # 2024-158) and the CNIL MR-004 registry.

Full medical and familial history was collected. The patient underwent a detailed general and ophthalmologic examination with slit-lamp examination, gonioscopy, ocular ultrasound, fundus examination, and optical coherence tomography (OCT). Brain magnetic resonance imaging was performed to investigate visual tract and intracerebral structures. Ultrasound was used to investigate cardiac and kidney structures.

The patient underwent analysis of 119 genes involved in ocular development using a customized next-generation sequencing panel, as previously described.¹² This method enables the detection of both single-nucleotide variants (SNVs) and copy number variations (CNVs). Variants were classified according to American College of Medical Genetics and Genomics guidelines.¹³ Candidate variants and their cosegregation with the disease within the family (Fig. 1A) were confirmed by Sanger sequencing with specific primers, available upon request.

Figure 1.

Clinical and genetic features of the proband, including the position of the regulatory variant in the highly conserved noncoding region upstream of FOXE3. (A) Family pedigree illustrating the segregation of the c.720C>A nonsense (p.Cys240*) and rs745674596G>A variants, along with a picture of the compound heterozygous proband displaying left complex microphthalmia and a right polar cataract. (B) UCSC screenshot showing the highly conserved region approximately 3 kb upstream of the FOXE3 coding sequence, containing the rs745674596G>A variant (highlighted in a square and capitalized). The forward and reverse primers used for PCR amplification are also capitalized. The sequence orthologous to the naturally occurring deletion in the mouse that causes microphthalmia is underlined.

CGH-array (BlueGnome, Cambridge, UK, 44k) was performed on the proband's DNA to identify genomic rearrangements, following previously described protocols.¹⁴

Whole-genome sequencing (WGS) of the proband and her parents was performed by the Centre National de Recherche en Génomique Humaine in Evry, France. Genomic DNA (1 µg) from each sample was used to prepare whole-genome sequencing libraries with the Illumina TruSeq DNA PCR-Free Library Preparation Kit (Illumina, San Diego, CA, USA), following the manufacturer's instructions. Libraries were sequenced on an Illumina NovaSeq 6000 platform as paired-end 150-bp reads, with samples pooled on a NovaSeq 6000 S4 flowcell to achieve a minimum average depth of 30×. After demultiplexing, sequences were aligned to the hg19 reference genome using the Burrows–Wheeler Aligner. Subsequent processing was done with GATK, SAMtools, and Picard following established best practices (http://www.broadinstitute.org/gatk/guide/topic?name=best-practices). Variants were called using GATK Haplotypecaller v4, and large variants (CNVs and structural variants [SVs]) were identified using WiseCondor,¹⁵ Canvas,¹⁶ and Manta.¹⁷

All variants were filtered using an in-house developed annotation software system (Polyweb, platform of Institut Imagine, Paris, France), based on two inheritance modes: recessive and dominant (including de novo mutations, parental mosaicism, and incomplete penetrance), with minor allele frequencies (MAFs) of <1% and <0.1%, respectively. The pathogenicity of the selected variants was assessed using the Alamut Mutation Interpretation Software (http://www.interactive-biosoftware.com), a decision support system for mutation interpretation based on Align DGVD, MutationTaster, PolyPhen‐2, SIFT, SpliceSiteFinder‐like, MaxEntScan, NNSPLICE, GeneSplicer, Human Splicing Finder, ESEfinder, and RESCUE‐ESE. The pathogenicity of the identified noncoding variant was assessed using scores such as GERP, CADD GRCh37-v1.7, and FATHMM-XF.¹⁸

Hi-C Data Visualization

To characterize the chromosome conformation structure of the FOXE3 locus in humans, we analyzed publicly available Hi-C data from the 4D Nucleome Data Coordination and Integration Center. This data, generated from H1-hESC cell lines, were processed using Micro-C and its updated version, Micro-C XL.¹⁹ These methods provide nucleosome-level resolution, with Micro-C XL offering improved recovery of higher-order interactions through enhanced cross-linking. Data from both methods were combined into contact matrices for each cell line, which were further processed into binary heatmap files, similar to the .hic files used by the UCSC Genome Browser.

For chromatin conformation analysis of the Foxe3 locus in mice, we used mouse embryonic stem cell (mESC) Hi-C data from Bonev et al.,²⁰ generated from intact cell nuclei using restriction enzyme-mediated DNA cleavage prior to proximity-based ligation.²¹ Hi-C reads were trimmed and quality-checked using Trim Galore (v0.6.5, Cutadapt v2.6, FastQC v0.11.9).²² They were mapped to mm10 and filtered for Hi-C artifacts using HiCUP (v0.7.2, Bowtie2 v2.3.5, R v3.6.0_3.9).²³ Contact maps were generated and analyzed with Juicer (v1.22.01)²⁴ using the parameter -s DpnII and resolution determined via the “calculate_map_resolution.sh” script. HOMER was used to compute observed matrices, adjusting for linear distance and sequencing depth. The resulting Hi-C contact maps were visualized as UCSC custom tracks on the UCSC Genome Browser.

Luciferase Reporter Assay Analysis of rs745674596 G and A Alleles

Two complementary single-stranded 81-base oligonucleotide pairs spanning GRCh37/hg19 chr1:47,878,388–47,878,468 and representing the rs745674596 G (WT) and A (Mut) alleles, with three additional nucleotides at their 3′ ends to introduce BglI cohesive ends, were designed and synthesized by Sigma-Aldrich Technologies (St. Louis, MO, USA) (Supplementary Table S1). The WT and Mut oligonucleotide pairs were hybridized separately and then cloned into the BglI-digested pGL4.24 expression plasmid (Promega, Madison, WI, USA), positioning them upstream of the PGK1 promoter and the firefly luciferase reporter gene. T4 DNA ligase in 1× reaction buffer (NEB, Milton Park, Oxfordshire, UK) was used according to the manufacturer's protocol. The resulting WT and Mut pGL4.24 constructs were transformed into competent Escherichia coli (OneShot TOP10; Thermo Fisher Scientific, Asnières-sur-Seine, France) and propagated to amplify the constructs. The plasmids were extracted and purified using a Miniprep kit (Qiagen, Hilden, Germany), according to the manufacturer's protocol. The insertion and orientation of the mutant and wild-type inserts were confirmed by Sanger sequencing of the plasmids.

Undifferentiated feeder-free embryonic stem cells (CCE)-Rx mESCs expressing the retina and anterior neural fold homeoboX (Rax), a key gene in ocular formation,²⁵ were cultured in gelatin-coated (Biocoat, Thermo Fisher Scientific, Waltham, MA, USA) 10-cm dishes with Dulbecco's modified Eagle's medium supplemented with pyruvate, glucose, and Glutamax (Thermo Fisher Scientific); β-mercaptoethanol (1%) and MEM nonessential amino acids (Thermo Fisher Scientific) (1%); penicillin-streptomycin and embryonic stem (ES) cell culture medium supplemented with fetal calf serum (Dutscher, Sigma-Aldrich, St. Louis, MO, USA) (15%); and leukemia inhibitory factor (LIF; Sigma-Aldrich). Cells were kept at 37°C and 5% CO₂, with daily medium changes and subculturing every 48 hours. For transfection, the cells were trypsinized and seeded in nonadherent dishes to promote embryoid body formation (1.2.10⁶ of cells in a 6-cm plate containing LIF-free medium) and incubated at 37°C/5% CO₂. Trypsinized cells were cotransfected using a recombinant vector (1 µg of either unmodified pGL4.24 vector or the mutant or wild-type constructs), Renilla reporter vector (100 µg; 100:1 ratio), and enhancer DNA in Effectene reagent, following the manufacturer’s protocol (Qiagen). Nontransfected CCE-Rx cells served as a negative control for background noise. Forty-eight hours posttransfection, embryoid bodies were harvested, washed with 1× PBS, and lysed with Passive Lysis Buffer (Promega). Lysates were stored at −20°C. Luciferase activity measurements were conducted in 96-well plates, with 20 µL per well, based on three independent transfections for each condition, with each transfection performed in triplicate. A luminometer (NOVOstar; BMG Labtech, Offenburg, Germany) recorded absolute luminescence emitted by both Firefly and Renilla luciferase activities, and the Firefly/Renilla luminescence ratio was calculated for each sample. The normalized ratios were analyzed using a one-way ANOVA, followed by Tukey's multiple comparisons test to identify significant differences between means (***P < 0.0001). Data are presented as mean ± SD on the graph.

Immortalization of Human Lens Epithelial Cells

Human lens epithelial cells (HLEpiCs; Innoprot, Derio, Bizkaia, Spain) were cultured in 6-well plates using Epithelial Cell Medium (EpiCM; Innoprot) supplemented with 10% fetal bovine serum (FBS) at 37°C in 5% CO₂. The cells were immortalized by transfection with 1 µg of the pLAS plasmid encoding the SV40 large T antigen,²⁶ using Lipofectamine 2000 (Thermo Fisher Scientific) according to the manufacturer’s protocol. The cells were transferred to a T25 flask and propagated to confluence, and the SV40 large T antigen-expressing immortalized HLEpiCs were isolated based on morphology through cell sorting into 96-well plates for subsequent expansion and experimentation.

DNA Pull-Down and Mass Spectrometry

DNA pull-down experiments were conducted using nuclear extracts from undifferentiated CCE-Rx and CCE cells, along with whole-cell lysates from immortalized HLEpiCs. We selected CCE cells for their broad gene expression profile and CCE-RAX for their relevance to eye development, given that RAX is essential for the formation of ocular structures during embryogenesis. CCE-Rx and CCE cells were cultured as previously described, harvested by trypsinization, and washed with PBS. Nuclear extracts were then prepared from both cell types following established protocols.²⁷ For whole-cell lysate preparation, immortalized HLEpiCs were cultured in EpiCM supplemented with 10% FBS, harvested by trypsinization, washed with PBS, and then resuspended in a buffer containing 150 mM NaCl, 50 mM Tris (pH 8.0), EDTA-free complete protease inhibitors (CPI) (Roche, Basel, Switzerland), and 1% NP40.

The biotinylated 81-bp oligonucleotides encompassing the rs745674596 G and A alleles (FOXE3-WT_F and FOXE3-Mut_F sequences; see Supplementary Table S1) were sourced from Integrated DNA Technologies (Leuven, Belgium). DNA pull-down experiments were conducted using 150 µg of nuclear extract or 1.5 mg of whole-cell lysate per reaction, following a previously described protocol.²⁷ Each pull-down was performed in technical duplicate, in which label-swapping of the dimethyl labels was performed between the replicates to eliminate labeling bias. Proteins were digested on-bead into tryptic peptides, which were subsequently cleaned and eluted using StageTips. The light- and medium-labeled samples were then combined in both forward and reverse reactions. Peptides were then separated on an Easy-nLC 1000 (Thermo Fisher Scientific, Waltham, MA, USA) connected online to an LTQ-Orbitrap Fusion Tribrid mass spectrometer (Thermo Fisher Scientific) or an Exploris 480 mass spectrometer (Thermo Fisher Scientific), using a 2- or 1-hour linear gradient of acetonitrile, respectively. For samples measured on the Fusion, scans were collected in data-dependent top-speed mode of a 3-second cycle with dynamic exclusion set at 60 seconds. For samples measured on the Exploris, scans were collected in data-dependent top 20 mode with a 45-second dynamic exclusion. Peptides were searched against the UniProt mouse or human proteome with MaxQuant v1.5.1.0,²⁸ using default settings, the appropriate dimethyl labels, and requantify enabled. Data were analyzed with Perseus version 1.4.0.0 and R. Outliers were identified in the forward and reverse reactions independently, using boxplot statistics with a threshold of 1.5 times the interquartile range. Significant proteins were those that were called outliers in both the forward experiment and the reverse experiment.

Generation of Foxe3 Mouse Models

Mouse lines harboring the regulatory rs745674596G>A variant (Foxe3^em1Gophallele, hereafter referred to as the rv allele) upstream of Foxe3 or a premature termination codon in the gene (c.82_92del; p.Gly28Argfs*112; Foxe3^em2Goph, referred to as the fs allele) were created using the CRISPR/Cas9 system by the Transgenesis platform at the LEAT Facility (Imagine Institute) (Supplementary Fig. S1). For the rv allele, the guide RNA (single guide RNA [sgRNA]: 5′-ATGCTCAGCCGCATCACGTC-3′; designed with CRISPOR [http://crispor.tefor.net/]) and a phosphorothioate-modified ssODN (5′-A*T*CCAGGCCCATGAGAAAGGGGCCACCTTCACTGGCCGTCTTATGCCCGGATGCTCAGCCGCATCACATCCGGCCCAGGGCCTGTGAAAAGAGGGCCCAGCCACGCTGAAAACGCGGA*T*T-3′) centered on the rs745674596 variant (underlined) were used. For the fs allele, the CRISPOR-designed sgRNA CTCCGGTTCGCGCCCCGGTT was employed. The CRISPR/Cas9 ribonucleoprotein complex was microinjected into C57BL/6J mouse zygotes’ pronuclei, following the method of Ucuncu et al.²⁹ Offspring were genotyped using PCR amplification and Sanger sequencing with appropriate primers (Supplementary Table S2). To minimize potential off-target mutations, mutant mice were backcrossed to C57BL/6J mice for several generations before intercrossing to obtain Foxe3+/+, Foxe3rv/+, Foxe3rv/rv, Foxe3+/fs, Foxe3fs/fs, and Foxe3rv/fs lines (Supplementary Fig. S1). Animal procedures were approved by the French Ministry of Research and adhered to the guidelines of the French Animal Care and Use Committee of Paris Descartes University (APAFIS# 31491), in full compliance with the ARVO Statement on the Use of Animals in Ophthalmic and Vision Research.

Ocular Phenotype Assessment in Adult Mice

Adult mice were examined for ocular structure under mild anesthesia with a ketamine–xylazine mixture using a slit lamp and OCT (Envisu R-class; Leica Microsystems, Wetzlar, Germany). Following euthanasia and enucleation, eye size was measured with a caliper, and histologic analysis was performed on 4-µm-thick hematoxylin and eosin (HE) stained sections from paraffin-embedded eyes fixed in Davidson’s solution. Ophthalmoscopic and histologic ocular anomalies, including eye size shortening, cornea and lens abnormalities, retinal defects, and other abnormalities, were systematically recorded.

To assess the impact of different genotypes on the penetrance and severity of ocular anomalies, statistical analyses were conducted using R software (version 4.4.1). Logistic regression model using Firth's bias reduction method (logistf package of R software) evaluated the association between genotype and phenotype, with “phenotype” as a binary variable (absent/present) and “genotype” as a factor with six levels (Foxe3+/+, Foxe3+/fs, Foxe3rv/+, Foxe3rv/rv, Foxe3rv/fs, Foxe3fs/fs). Odds ratios and 95% confidence intervals were calculated. Statistical significance was set at P < 0.05, and additional analyses were performed to evaluate the effects of age and gender on phenotype occurrence and include them as covariates if necessary (Table 1).

Table 1.

Univariate Analysis of the Potential Association of the Phenotype With Age and/or Gender

Variable	Odds Ratio	95% Confidence Interval	P Value
Intercept	0.01	0.00–0.06	4.6E-07
Age	1;19	0.97–1.47	0.1
Genotype
Foxe3+/+	—	—	—
Foxe3 rv/+	15.9	1.65–2131	0.013
Foxe3+/fs	19.6	2.01–2629	0.007
Foxe3 rv/rv	22.9	2.5–3053	0.002
Foxe3 rv/fs	63	7.81–8181	5.2E-07
Foxe3 fs/fs	1305	60.4–413122	2.6E-10

Age (in months) was significantly associated with the presence of a phenotype (P < 10⁻⁴), while gender showed no significant association; therefore, age but not gender was included in the logistic regression model with Firth's correction, using Foxe3+/+ as the reference genotype.

To assess the association between the phenotype and both the number and nature of mutated alleles, a logistic regression adjusted for age (in months) was employed. The severity of the genotype was scored based on the number of mutated alleles and the severity of the mutation (allele + = 0, allele rv = 0.75, and allele − = 1; Table 2), while phenotypic severity was classified into four levels for analysis: no phenotype (0), mild (cataract; 1), moderate (anterior segment dysgenesis; 2), and severe (significant developmental anomalies with eye growth defects; 3) (Table 2). An ordinal regression model using cumulative link models (ordinal package in R software) was applied to analyze the severity of the phenotype in mice based on the number and severity of mutated alleles (Table 3).

Table 2.

Distribution of Analyzed Mice by Genotype Across Varying Classes of Genotypic and Clinical Severity

		PSS
	GSS	0	1	2	3
Foxe3+/+	0	30	0	0	0
Foxe3rv/+	0.75	23	2	3	0
Foxe3+/fs	1	17	2	3	0
Foxe3rv/rv	1.5	18	8	2	0
Foxe3rv/fs	1.75	19	14	7	0
Foxe3fs/fs	2	0	0	0	12

The scoring system quantifies genetic contributions to phenotypic outcomes. GSS (Genotype Severity Score): Sum of scores from both alleles, calculated from allelic scores of + = 0, rv = 0.75, and − = 1, representing overall genotype severity. PSS (Phenotype Severity Score): 0, no phenotype; 1, mild; 2, moderate; 3, severe.

Table 3.

Correlation Between Mutated Alleles and Phenotype Severity: Analysis of Odds Ratios From Statistical Regression Models

Variable	OR	95% CI	P Value	OR (CLM)	95% CI (CLM)	P Value (CLM)
Intercept	0.01	0–0.06	3.00E-07
Age	1.1	0.93–1.31	0.3	1.09	0.93–1.29	0.3
Genotype	10.6	4.62–28.4	3.00E-07	16.6	6.53–50.7	6.59E-08

Analysis of the correlation between the number and severity of mutated alleles and the presence of a phenotype reveals statistically significant results, with ordinal regression analysis using cumulative link models (CLMs) demonstrating a significant association between the nature and severity of the mutated alleles and the severity of the phenotype (P < 0.01). CI, confidence interval; OR, odds ratio.

Assessing Ocular Abnormalities in Developing Mouse Embryos

For analysis of ocular development during embryonic life, mice were bred at 6 PM and checked for a vaginal plug 16 hours later. Females with a plug and a weight gain of more than 3 g over 7 days were euthanized at embryonic (E) day 12.5, 13.5, 14.5, or 18.5 or at birth (P0) to collect embryos. Embryos were fixed in 10% buffered formalin at 4°C for 24 hours, heads were separated and embedded in paraffin, and 4-µm-thick eye tissue sections were subjected to HE staining. Ocular anomalies were systematically recorded.

Quantitative Real-Time PCR Analysis of Gene Expression in Mouse Lens Epithelium

Mice aged 2 to 5 months were euthanized, and their eyes were enucleated and dissected on ice to isolate the lens epithelium. The procedure involved opening the eyes at the ora serrata, removing the lens, and carefully separating the capsule from the fiber mass. The lens epithelia from both eyes were pooled, and total RNA was extracted using the RNeasy Mini Kit (Qiagen), following the manufacturer’s protocol. RNA purity and concentration were assessed using a spectrophotometer (Nanodrop; Thermo Fisher Scientific). mRNA was reverse transcribed from 35 ng of total RNA using the cDNA Verso Kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. The abundance of Foxe3 mRNA, along with the housekeeping genes Tbp, Gusb, Gapdh, and Hprt1, was quantified by real-time quantitative PCR using specific primers (Supplementary Table S3) and the SsoAdvanced Universal SYBR Green Supermix (Bio-Rad, Hercules, CA, USA) in a Realplex2 Mastercycler (Eppendorf, Montesson, France).

The mean quantification cycle (Cq) value for each cDNA sample was calculated from technical triplicates, with a standard deviation below 0.3 Cq to ensure reproducibility. Expression levels of target genes were normalized using a normalization factor generated by the geNorm algorithm (Microsoft Excel add-in),³⁰ which identifies the most stable reference genes across samples. Negative controls, including no reverse transcriptase (non-RT) and no template control (NTC) reactions, were included in each quantitative PCR run (Cq for NTC = undetermined; non-RT > 40).

Gene expression data are presented as the mean ± SEM from three independent experiments, expressed as relative mRNA expression ratios. Statistical significance was assessed using Fisher's protected least significant difference test following ANOVA, performed with Statview software (version 5; SAS Institute, Cary, NC, USA). P < 0.05 was considered statistically significant.

Western Blot Analysis of FOXE3 in Mouse Lens Epithelium (Capsule)

Mouse lens epithelium was prepared as described above. Proteins (70 µg) were prepared using RIPA lysis buffer (Thermo Fisher Scientific) and resolved on a Mini-ProteanTGX Stain Free 4% to 15% gel according to the supplier's recommendations (Bio-Rad, Marne la Coquette, France). Proteins were transferred to a PVDF membrane (Bio-Rad) using an RTA transfer kit (Bio-Rad), which was probed with FOXE3 (1:100; Santa Cruz Biotechnology, Dallas, Texas, USA, sc-377465) and vinculin (1:2000; Abcam, Cambridge, MA, USA, ab91459). Secondary antibodies, either mouse IgG or rabbit IgG HRP (Thermo Fisher Scientific), were used at a 1:8000 dilution. Membranes were developed with Clarity Western ECL substrate (Bio-Rad) and visualized using a Chemidoc MP Imaging System (Bio-Rad). Protein quantification was performed using the gel analysis tool in Fiji (v. 1.53t) software.

For each individual sample, the integrated signal of the target protein was normalized to that of the loading control. Each experiment was independently repeated three times. Densitometric values from all replicates were pooled per individual animal (n = 18). Statistical comparisons between groups were performed using one-way ANOVA. Analyses were conducted using GraphPad Prism software (version 8.0.2; GraphPad Software, La Jolla, CA, USA), and differences were considered statistically significant at P < 0.05.

Analysis of the Spatiotemporal Expression Patterns of Foxe3 and Three Candidate Genes Identified Through Pull-Down Assays

Coexpression of Foxe3 with Usf2, Cnbp, and Gabpa during ocular development was analyzed using RNAscope to visualize cell-specific expression and tissue morphology. For each developmental stage (E12.5 to E18.5), two embryos were analyzed, and three tissue sections per embryo were processed. Embryos were fixed in 4% paraformaldehyde (PFA) at 4°C for 24 hours, followed by washing in PBS. After embedding in OCT compound, the heads were rapidly frozen in liquid nitrogen and stored at −80°C. For sectioning, heads were equilibrated to −20°C for 1 hour and sectioned at a 14-µm thickness using a Cryostat set to −20°C for the object and −18°C for the chamber. Sections were air-dried for 1 to 2 hours and stored at −80°C.

Slides were prepared for RNAscope analysis according to the manufacturer’s instructions for frozen sections (Advanced Cell Diagnostics, Biotechne, Rennes, France). In brief, the sections were dehydrated in ethanol baths, dried, and treated with H₂O₂ for 10 minutes at room temperature. Proteinase digestion was performed the following day with Protease IV for 20 minutes at room temperature. Head sections were incubated for 2 hours at 40°C with a probe mix targeting Foxe3 and the candidate genes Usf2, Cnbp, and Gabpa. Positive controls included probes for Polr2a, Ppib, and Ubc, while DapB served as the negative control for the experimental groups. RNAscope signal amplification, including a series of amplifiers, HRP activators, blockers, and fluorophore incubations at 40°C, was conducted following the manufacturer’s protocol. Opals (Akoya Bioscience, Menlo Park, California, USA) were diluted 1:1500 in TSA buffer and assigned to channels C1 (Opal 570), C2 (Opal 690), and C3 (Opal 520). Nuclei were stained using DAPI, and the sections were mounted with Fluoromount and stored at −20°C for at least 12 hours before imaging.

Images were acquired with a Zeiss Spinning Disk Confocal (Zeiss, Oberkochen, Germany). Whole sections were scanned using a 10× objective, and detailed images of lens epithelium were captured with a 40× or 63× oil objective as maximal projections of 10 or 30 stacks covering 6.09 µm (frozen sections). Images were saved as .czi files and converted to TIFF format for analysis. Quantitative image analysis was performed using Fiji (v. 1.53t) and CellProfiler (v. 4.2.1) software following the analysis method suggested by the manufacturer. In brief, the presumptive lens was manually defined as the region of interest (ROI). The integrated intensity of the ROI was measured for each channel, along with the area covered by the DAPI signal. For each channel, 25 spots were measured to determine the average intensity value of one mRNA spot, and 25 nuclei were delimited to measure the average area of one nucleus. Dividing the integrated intensity of the ROI by the average dot intensity estimates the total dots of mRNA in the region, and this value can be reported as the number of nuclei, resulting in a mean value of dots/nuclei. An unpaired bilateral Student's t-test was used for two-group analyses. Data are presented as the mean ± SEM; P < 0.05 was considered significant.

Small Interfering RNA-Mediated Knockdown of Usf2, Cnbp, and Gabpa Expression in Cultured Epithelial Lens Cells From Foxe3 Mouse Lines

Foxe3+/+ and rv/rv mice between 2 and 4 months of age were euthanized, and their eyes were enucleated. Eyes were briefly washed with 70% ethanol and incubated in PBS under sterile conditions. Lens capsules were recovered as described above, and anterior epithelial cells were incubated in P12 cell culture plates with EpiCM at 37°C with 5% CO₂ until the first cells adhered to the plate, typically within 48 hours. Once the cells reached confluence, they were maintained in EpiCM on plates coated with poly-L-lysine (Sigma-Aldrich. Saint-Quentin-Fallavier, France) and passaged using 0.05% trypsin-EDTA (Thermo Fisher Scientific).

For small interfering RNA (siRNA) knockdown, cells at 70% confluence were transfected with Usf2, Cnbp, and Gabpa (50 nM; FlexiTube GeneSolution; Qiagen, Courtaboeuf, France, Gene ID: GS22282, GS12787, and GS14390, respectively) or no-target siRNAs (AllStars Neg. Control siRNA, cat. 1027281; Qiagen) using Lipofectamine 2000 reagent (Invitrogen, Illkirch, France). Transfected cells were harvested after 48 hours, and total RNA was extracted for real-time quantitative PCR analysis as described above using specific primers (Supplementary Table S3). Foxe3 data represent the mean ± SD of four independent experiments, normalized to the no-target control siRNA, with statistical significance evaluated using a two-way ANOVA followed by Dunnett's post hoc test.

Results

Targeted and Whole-Genome Sequencing Revealed a Likely Disease-Causing Noncoding Variant Upstream of FOXE3 in a Patient

The proband, an 11-year-old girl, presented with left complex microphthalmia with Peters anomaly and a right polar cataract (Fig. 1A). She had no significant medical history and was the first child of healthy, young Caucasian parents with no notable family medical history.

Targeted sequencing using a 119-gene panel identified a heterozygous FOXE3 c.720C>A (NM_012186.3) pathogenic variant, resulting in a p.Cys240* protein change and a premature stop codon. Sanger sequencing–based familial segregation analysis revealed that the pathogenic variant was inherited from the asymptomatic father (Fig. 1A).

The FOXE3 c.720C>A nonsense mutation has been reported multiple times, typically with a second mutation in trans in individuals with ocular anomalies, while heterozygous relatives remain asymptomatic.^4,9 This led us to investigate a potential maternally inherited noncoding variant affecting FOXE3.

Array-CGH (44k) analysis was performed to detect potential rearrangements near FOXE3 that could impact its expression, but no such rearrangements were found. Sequencing of an approximately 600-bp evolutionarily conserved region, ∼3 kb upstream of FOXE3 (GRCh37/hg19 chr1:47,878,072–47,878,696), revealed a rare heterozygous G>A variant at chr1:47,878,428 (rs745674596; with C as a possible alternative nucleotide) shared by the proband, her mother, and her sister (Figs. 1A, 1B). The ultra-rare G>A substitution (with a minor allele frequency of 0.0001 in GnomAD v4.1, only observed in the heterozygous state) occurs at a nucleotide with high pathogenicity scores, including GERP (3.5799) and CADD GRCh37-v1.7 (18.83), indicating a potential deleterious effect. Notably, this variant is positioned 16 bp downstream of a 22-bp sequence, the deletion of which has been associated with complex microphthalmia and cataracts in mice (Fig. 1B).¹¹ In addition, analysis of the variant using FATHMM-XF,¹⁷ a predictive tool for evaluating the functional consequences of noncoding SNVs, yielded a noncoding score of 0.519910, strongly suggesting the variant’s pathogenicity. These findings collectively support a potential role of this noncoding variant in the proband’s disease. A homozygous G>C change at chr1:47,878,626 (rs10399673), inherited from both heterozygous parents, was also identified. This variant, with an MAF of 0.444 in GnomAD v4.1, is likely benign. WGS was subsequently performed to rule out other potential disease causes. It did not identify any additional candidate variants or rearrangements (including copy number and structural variants) in noncoding regions within FOXE3 and its surrounding areas (Supplementary Fig. S2), especially in the topologically associating domain (TAD) regulatory unit comprising the gene (Supplementary Fig. S3). Additionally, after analyzing the data set for both recessive and dominant inheritance patterns (the latter including de novo mutations, parental mosaicism, and incomplete penetrance), no other pathogenic variations were found that could explain the patient's ocular phenotype.

In Vitro Luciferase ASSAY INDICATES THAT the FOXE3 rs745674596 Noncoding G>A Change Acts as a Regulatory Variant

To assess whether the G>A change at position chr1:47,878,428 (rs745674596) affects FOXE3 expression, we cloned into a BglI-digested pGL4.24 expression plasmid (Promega) two fragments of 81 bp encompassing the rs745674596 G and A alleles (Supplementary Table S1) upstream of a luciferase reporter gene for functional analysis. Luciferase assays showed that the mutant (A allele) construct led to a ∼45-fold increase in reporter activity compared to the empty vector (P < 0.0001), while the wild-type (G allele) construct induced a ∼10-fold increase (not statistically significant, P = 0.2338). When directly comparing the mutant and wild-type constructs, the A allele exhibited a 4.5-fold higher expression than the G allele (Supplementary Fig. S4). This result supports the classification of the rs745674596 variant as a regulatory variant (rv), capable of modulating reporter gene transcription in vitro.

Mice With the Regulatory Variant, Frameshift Variation, or Both Show Genotype-Correlated Increased Eye Anomalies

To assess the impact of the rs745674596 noncoding G>A variation (rv) in vivo, we used CRISPR/Cas9 to edit the C57BL/6J mouse genome, generating mice with single heterozygosity, homozygosity, and compound heterozygosity for this regulatory variant and a premature termination codon (p.Gly28Argfs*112) (Supplementary Fig. S1). Although the mouse frameshift mutation (p.Gly28Argfs112) leads to the loss of the entire forkhead domain and is thus likely a null allele, this strategy was consistent with our aim to model the severe phenotype observed in humans carrying homozygous truncating mutations. In patients, such mutations, including the p.Cys240* nonsense variant, typically result in complex microphthalmia, regardless of whether the forkhead domain is partially or entirely disrupted.⁴

The resulting genotypes (Foxe3+/rv, Foxe3+/fs, Foxe3rv/rv, Foxe3rv/fs, and Foxe3fs/fs) and wild-type controls (Foxe3+/+) were evaluated and compared for ocular globe size and structure using caliper measurements, slit lamp (Supplementary Table S4), and OCT examinations. The Foxe3+/+ mice (n = 30) exhibited no ocular anomalies (Fig. 2A and Supplementary Figs. S5 and S6), with anteroposterior ocular diameters (APODs) ranging from 3.5 to 4 mm at 3 months of age, thereby establishing the normal range. The APODs of heterozygous Foxe3+/rv (28 animals, 56 eyes) and Foxe3+/fs (22 animals, 44 eyes) mice remained within the normal range. However, some of these mice exhibited cataracts and anterior segment anomalies (Fig. 2A, Supplementary Figs. S5 and S6, and Supplementary Table S4). The prevalence of ocular anomalies increased more strikingly in animals with these variants in homozygosity (Foxe3rv/rv) or compound heterozygosity (Foxe3rv/fs), although their APODs also remained within the normal range. Specifically, 36% of Foxe3rv/rv (28 animals, 56 eyes) and 53% of Foxe3rv/fs (40 animals, 80 eyes) mice exhibited cataract and anterior segment anomalies (corneal opacities, lens and iridocorneal adhesions, iris coloboma) (Fig. 2A, Supplementary Figs. S5 and S6, and Supplementary Table S4). Foxe3fs/fs null mice (12/24 animals/eyes) exhibited a more severe and consistent phenotype, characterized by microphthalmia (APOD ≤3 mm) with significant anterior segment anomalies, including corneal opacities, central pits, iridocorneal and lens adhesions, athalamia, cataracts, and microphakia. This distinctive phenotype, present in 100% of Foxe3fs/fs null mice, aligns with the diagnosis of complex microphthalmia (Fig. 2A, Supplementary Figs. S5 and Supplementary S6, and Supplementary Table S4). Many of these mice also showed a small pyramidal pigmentary and vascular formation, suggestive of persistent fetal vasculature. Of note, the retina and its layers remained unchanged in most cases.

Figure 2.

Comparative histologic analysis of ocular development in Foxe3 mutants and control samples. (A) Representative live view, OCT images, and histologic sections and from adult wild-type mice (a–c) and mutant mice (n ≈ 25 per genotype) illustrate anomalies of increasing severity from a focal anterior subcapsular cataract lesion in Foxe3rv/rv eyes (*) (d–f) to cataract (**) (i) with extensive irido-lenticular and irido-corneal adhesions and focal fibrotic thickening of the anterior lens capsule (*) (h, i) in Foxe3rv/fs (g–i) to complex microphthalmia, with corneal clouding (*) (j, k), central pit (arrow) and athalamia (**) (j, k), extensive uveo-corneal adhesions (**) (l), and a small, vacuolated, triangular-shaped cataractic lens (*) (l) in Foxe3fs/fs. (B) Hematoxylin and eosin–stained eye sections of Foxe3+/+, Foxe3rv/fs, and Foxe3fs/fs mouse embryos from E12.5 to birth (P0) illustrate ocular development (a–t). (a–e) In Foxe3+/+ animals, early lens development is seen at E12.5, with the lens vesicle detaching from the surface ectoderm, a defined optic cup, and a neuroepithelial layer forming in the retina (a). By E13.5, the lens vesicle is rounder, with early fiber cell differentiation and thickening retina, marking early stratification (b). At E14.5, primary lens fibers elongate, the retinal ganglion cell layer becomes visible, and the optic nerve head connection develops (c). By E18.5, the lens and retina have mature features, including distinct retinal layers and defined anterior segments like the cornea and ciliary body (d). At P0, the lens is fully mature, with organized fiber cells and a defined capsule, while anterior structures like the cornea and iris continue developing (e). (f–j) In Foxe3rv/fs animals, ocular development appears largely normal, with minor lens fiber vacuolization (*) (i, j), which may contribute to the adult cataract phenotype observed in nonnull mice (Table 2). In some individuals, a small delay in lens detachment can be observed (*) (g). (k–t) In Foxe3fs/fs animals, initial development appears normal, but by E13.5, the anterior epithelial layer of the lens is disorganized (**) (l, q) and a delay in lens detachment is observed, manifesting by a persistent lenticulo-corneal connection (*) (l, q). Mild vacuolization and swelling of lens fibers become apparent by E14.5, worsening over time (* in m, r, s, and t). The lens remains unusually close to the presumptive cornea, giving the appearance of an open lens at the anterior pole (** in s and t), with possible protein release into the corneal mesenchyme. The anterior epithelial layer gradually disappears, resulting in a microphakic lens (n, o, s, and t). Additionally, inconsistent fibrosis is observed within the primary vitreous vascularization, extending from the posterior lens pole to the retina (**) (n, o, and t).

We assessed the statistical significance of ocular defect penetrance (including cataracts, anterior segment anomalies, and complex microphthalmia), based on observed frequencies of 0% in Foxe3+/+, 18% in Foxe3rv/+, 23% in Foxe3+/fs, 36% in Foxe3rv/rv, 53% in Foxe3rv/fs, and 100% in Foxe3fs/fs (Supplementary Fig. S6). Ocular penetrance in mice was significantly correlated with the number and the nature of the pathogenic variations (Supplementary Table S3). In the same manner, when classifying the phenotype in class of severity (Table 1), the severity of the ocular phenotype was significantly associated with the number and nature of the pathogenic variations (Table 3).

Analysis of Ocular Development in Foxe3 Mutant Mice Reveals Critical Role of FOXE3 in Lens Differentiation and Microphthalmia

To investigate the mechanisms behind Foxe3-related ocular anomalies, we compared ocular structures from compound heterozygous Foxe3rv/fs, homozygous Foxe3fs/fs, and wild-type Foxe3+/+ mice at key embryonic stages (E12.5, E13.5, E14.5, E18.5, P0). Histologic analyses (three animals, six eyes per genotype) showed nearly normal ocular development in Foxe3 rv/fs animals, with slight lens fiber vacuolization potentially linked to cataracts (Fig. 2B), as around 50% are expected to present this condition. In Foxe3fs/fs embryos, which all develop complex microphthalmia in adulthood, early lens structures were intact, but disorganization in the anterior lens epithelium and persistent lenticulo-corneal connections emerged by E13.5, leading to progressive lens abnormalities and degeneration (Fig. 2B). These findings suggest that while Foxe3 is expressed from E9.5 in the lens placode, it is not required for lens formation, as Foxe3fs/fs embryos had normal lens structure at E12.5, and Foxe3 expression is confined to the anterior lens vesicle. However, Foxe3 is critical for anterior epithelial cell elongation and differentiation into lens fibers starting at E12.5. Disruption of these processes leads to early and progressive lens abnormalities, contributing to the complex microphthalmia phenotype (Fig. 2B).

FOXE3 Levels in the Lens Exhibit a Genotype-Dependent Correlation in Mice

To explore the correlation between Foxe3 expression levels and genotypes, we measured Foxe3 mRNA using real-time quantitative PCR analysis from lens capsules of at least four adult animals per genotype. The capsules from both lenses of each animal were pooled and analyzed together. This study demonstrated a 50% reduction in Foxe3 abundance (P < 0.01) in the lens capsule and by inference in the anterior lens epithelium, the only tissue expressing Foxe3 in the adult lens in Foxe3rv/rv mice compared to Foxe3+/+ wild-type littermates. Correspondingly, the Foxe3 levels in heterozygous Foxe3+/rv mice were intermediate between those of Foxe3rv/rv and Foxe3+/+ capsules. The Foxe3fs/fs mice exhibited a near-complete loss of Foxe3 transcript expression, with mRNA levels reduced to approximately 1% of those measured in wild-type controls. Interestingly, contrary to the expected reduction in Foxe3 abundance in Foxe3+/fs and Foxe3rv/fs mice, we observed a trend toward increased Foxe3 levels in both genotypes compared to wild-type Foxe3+/+ controls (Fig. 3A). This was further supported by quantitative mRNA detection through RNAscope analysis of E12.5 embryos, which revealed significantly increased Foxe3 staining in the anterior lens epithelium of Foxe3fs/fs null mice and decreased staining in Foxe3rv/rv mice compared to wild-type controls. In Foxe3rv/fs mice, staining showed intermediate expression (Fig. 3B), reflecting reduced expression from the allele carrying the regulatory variant and increased expression from the null allele, suggesting Foxe3 upregulation associated with the p.Gly28Argfs*112 variation during development. FOXE3 protein levels were quantified through Western blot analysis, using total protein extracts from the lens capsules of each animal (three adult mice per genotype). The genotypes included mice with the regulatory variant in single heterozygosity, homozygosity, or compound heterozygosity with the null allele, as well as from the microphakic lenses of mice homozygous for the null allele. This analysis revealed a stepwise decrease in FOXE3 abundance across genotypes, from Foxe3+/+ to Foxe3+/fs to Foxe3rv/rv to Foxe3rv/fs to Foxe3fs/fs (Fig. 3C). Notably, despite increased mRNA levels, FOXE3 protein abundance was markedly reduced in lenses with the null allele, with a decrease of approximately 50% in Foxe3+/fs mice and around 75% in Foxe3rv/fs mice, and FOXE3 was virtually undetectable in Foxe3fs/fs lenses (Fig. 3C). It is important to note that the antibody used in our Western blot detects the C-terminal domain of FOXE3. As such, it would not recognize any truncated protein lacking this region. Given the position of the frameshift mutation, any resulting protein would include only the N-terminal 28 amino acids of FOXE3. While we cannot formally exclude the presence of a functional truncated protein, the severity of the phenotype supports a complete loss of function.

Figure 3.

Comparative analysis of Foxe3 mRNA and protein expression in lens epithelium of Foxe3 mutant and control mice. (A) Relative Foxe3 abundance in the anterior lens epithelium was measured by quantitative RT-PCR in Foxe3+/+, Foxe3rv/+, Foxe3+/fs, Foxe3rv/rv, Foxe3rv/fs, and Foxe3fs/fs mice. (B) RNAscope in situ hybridization provides quantitative mRNA detection in tissue sections; however, the probes cannot differentiate between wild-type Foxe3 mRNA and the frameshift variant produced in fs/fs mice. In situ hybridization was performed on lens sections from Foxe3+/+, Foxe3rv/rv, Foxe3rv/fs, and Foxe3fs/fs mice at E12.5. (C) FOXE3 protein levels in the lens epithelium, as determined by Western blot (WB). Quantitative analysis of WB signals shows a significant reduction in FOXE3 protein abundance with the rv variant. Only statistically significant differences are indicated.

Together, these results provide strong support for the pathogenicity of the rs745674596 noncoding G>A variant through decreasing expression of the gene and underscore a correlation between reduced FOXE3 protein levels and an increase in the penetrance and severity of ocular anomalies.

DNA Pull-Down and Mass Spectrometry Reveal Interactions Influencing FOXE3 Expression by the rs745674596 Noncoding G>A Variant

DNA pull-down and mass spectrometry analyses were performed to determine whether the rs745674596 noncoding G>A variant influences FOXE3 expression by affecting the binding of specific transcription factors. We utilized whole lysates from immortalized HLEpiC and nuclear extracts from mouse CCE-RAX and CCE cell lines, using oligonucleotides specific to either wild-type or mutant human noncoding sequences (Supplementary Table S1).

Our experiments identified 16, 16, and 17 transcription factors differentially bound to wild-type and mutant oligonucleotides in HLEpiC, CCE-RAX, and CCE cell lines, respectively (Fig. 4A). Notably, USF2, GABPA, and GABPB1 were common across all three cell lines, although GABPB1 was excluded from further analysis due to its role as a GABPA cofactor. CNBP, linked to microphthalmia in mice,³¹ was also considered despite being detected only in HLEpiC and CCE cells. GABPA demonstrated a preference for binding to the mutant variant over its wild-type counterpart. In contrast, USF2 and CNBP displayed stronger binding to the wild-type version, suggesting that their interactions were either partially or completely diminished in the mutant form.

Figure 4.

Identification and functional analysis of transcription factors binding to Foxe3 regulatory element in lens cells. (A) Combined DNA-PD and mass spectrometry analysis in murine CCE cells, CCE-Rx cells, and immortalized HLEpiC. Factors with a significant preference for binding the wild-type oligonucleotide are marked in red (P < 0.05), while those preferring the mutant oligonucleotide are marked in blue (P < 0.05). (B) RNAscope in situ hybridization shows spatiotemporal expression patterns of candidate genes. At E12.5, Foxe3 expression localizes within the anterior layer of the presumptive lens, while Cnbp, Gabpa, and Usf2 display ubiquitous expression. (C) Relative Foxe3 expression in lens epithelial cells from Foxe3+/+ and rv/rv genotypes after inhibition of Cnbp, Usf2, and Gabpa, as measured by quantitative RT-PCR (*P < 0.05).

RNAscope in situ hybridization of developing mouse eyes demonstrated colocalization of Usf2, Cnbp, and Gabpa with Foxe3 mRNA in the anterior lens epithelium, although their expression was broader compared to the lens-specific expression of Foxe3 (Fig. 4B).

siRNA-mediated knockdown analysis allowed a significant reduction in the expression of the candidate genes, confirming effective target depletion in lens epithelial cells (Supplementary Fig. S7). Knockdown experiments using siRNAs revealed a statistically significant reduction in Foxe3 mRNA in Foxe3+/+ cells (P = 0.0322) and a marked trend toward reduction in Foxe3rv/rv cells upon Usf2 inhibition. In contrast, Cnbp and Gabpa depletion did not affect Foxe3 expression (Fig. 4C).

These results suggest that the noncoding rs745674596G>A variant impacts FOXE3 expression by reducing Foxe3 levels through its effect on USF2 binding, thereby further supporting the pathogenicity of the variant and highlighting the critical role of USF2 in lens and eye development.

Discussion

The FOXE3 region, located within a 50-kb TAD on chromosome 1p33, contains several evolutionarily conserved sequences upstream of the gene. Among these, the chr1:47877964–47878774 region (GRCh37/hg19), located 3 kb upstream of FOXE3, has been linked to reduced Foxe3 expression and complex microphthalmia in the rct mutant SJL/J mouse carrying a 22-bp homozygous deletion.¹¹ Consistent with a regulatory role, this region displays open chromatin and DNase hypersensitive areas, despite lacking CpG islands. Studies have also shown binding of transcription factors such as SOX2, PAX6, PITX3, and TFAP2A, which are crucial for ocular development.⁸ Therefore, genetic alterations in this conserved region have the potential to impact FOXE3 expression and contribute to both recessive and dominant FOXE3-associated diseases reported in the literature.

Conforming with established genotype–phenotype correlations linking recessive FOXE3 mutations to the most severe FOXE3-related diseases,⁴ the ultra-rare noncoding rs745674596 G>A change at chr1:47878428 identified in trans of the FOXE3 c.720C>A (p.Cys240*) truncating mutation in a patient with complex microphthalmia was expected to result in a loss of function through the dampening of FOXE3 expression. Intriguingly, luciferase reporter assay analysis of the variant revealed increased transactivation activity compared to the wild-type. While supporting an impact of the change on FOXE3 expression, this observation challenged the recessive inheritance model and suggested the possibility of dominant inheritance with incomplete penetrance. However, in contrast to the in vitro results, in vivo studies revealed markedly reduced Foxe3 expression in the lens of mice homozygous for the noncoding rs745674596 G>A variant, confirming the expected loss-of-function mechanism. This discrepancy highlights the limitations of basic in vitro analyses in assessing pathogenic mechanisms of noncoding variants. Regulatory regions can contain both activating and repressing motifs, and gene expression regulation can vary over time and across different tissues. It is plausible that the full complement of factors required for optimal combinatorial transcriptional regulation is not fully expressed or functionally active within the in vitro system, potentially leading to discrepancies between in vivo and in vitro observations. Another concern regarding mRNA-level analysis arises from the unexpectedly elevated levels of Foxe3 mRNA with the p.Gly28Argfs*112 frameshift mutation observed in the developing lenses of Foxe3fs/fs embryos compared to wild-type littermates. This issue is further highlighted by comparable Foxe3 mRNA levels in wild-type and Foxe3rv/fs mice, where reduced expression from the noncoding rs745674596G>A allele is masked by increased expression of the frameshift mutation. The mechanisms behind elevated levels of mRNA encoding premature termination codons remain unclear. Given that Foxe3 is a single-exon gene and both null and wild-type mice share the same genetic background, splicing and increased transcription are unlikely to explain the higher levels of mRNA encoding the p.Gly28Argfs*112 frameshift variation. This accumulation is more likely driven by the absence of nonsense-mediated mRNA decay or the stabilization conferred by RNA-binding proteins or other factors that prevent degradation.³² Interestingly, in adulthood, the elevated levels of mRNA transcripts carrying premature termination codons appear to be observed only in Foxe3+/fs and Foxe3rv/fs lenses but not in Foxe3fs/fs. One possible explanation is that, in the heterozygous or compound heterozygous contexts, the lens epithelial cells largely retain their identity and transcriptional activity, allowing for the accumulation or stabilization of these mutant transcripts. In contrast, in the homozygous Foxe3fs/fs background, epithelial cell identity may be severely compromised, leading to a general transcriptional shutdown or altered RNA processing, which results in a drastic reduction of the fs-containing transcripts. Whether the c.720C>A nonsense mutation (p.Cys240*) identified in the individual has a similar impact on FOXE3 content in the lens remains an open question. Yet, the evidence that both the noncoding variant and the frameshift mutation had unexpected effects on expression levels underscores the pitfalls of relying solely on RNA levels as biomarkers for pathogenicity.

Interestingly, a pronounced reduction in Foxe3 expression in Foxe3fs/fs lenses compared to Foxe3+/+ was accompanied by a downregulation of Prox1 and Pitx3, two critical transcription factors governing lens fiber cell differentiation (Supplementary Fig. S8). This concomitant decrease in Prox1 and Pitx3 suggests a complex regulatory interplay rather than a simple linear relationship. One plausible mechanism is that loss of Foxe3 disrupts the transcriptional and epigenetic landscape required for the appropriate activation of Prox1 and Pitx3, potentially through perturbation of lens epithelial cell identity or induction of cellular stress pathways. These findings corroborate prior work by Wada et al.,³³ who demonstrated that expression of truncated Pitx3 results in the downregulation of both Foxe3 and Prox1. Collectively, these data support a model wherein FOXE3 functions not only as a direct transcriptional regulator but also as a central upstream integrator within a gene regulatory network essential for lens morphogenesis. The concomitant dysregulation of Prox1 and Pitx3 highlights the context-dependent and multifactorial role of FOXE3 in maintaining transcriptional homeostasis in lens epithelial cells.

In contrast to mRNA studies, which failed to link Foxe3 expression with the penetrance of ocular defects, analysis of FOXE3 protein content in the lens of mice revealed a correlation between FOXE3 levels and both the frequency and severity of ocular anomalies. The spectrum of ocular anomalies affecting eye growth, cornea, anterior segment, and lens, along with their graded frequency and severity in relation to Foxe3 expression in mice, aligns with human findings, where genotype–phenotype studies have linked the penetrance of these anomalies to the number and type of FOXE3 mutations.^4,9

More specifically, the phenotype of the Foxe3fs/fs null mice exhibited normal eye structure by E12.5. However, from E13.5 to E14.5, mild and discrete abnormalities emerged, including delayed lens vesicle detachment, disorganized anterior lens epithelium, mild vacuolization, and abnormal lens positioning, resulting in microphakia by E18.5. In adults, we observed a consistent phenotype of complex microphthalmia in all mice, accompanied by severe anterior chamber lesions and lens anomalies but no “true” anophthalmia. These findings align with the established role of the FOXE3 transcription factor in lens epithelial cell growth, where it regulates proliferation, apoptosis, and cell cycle.^34,35 Foxe3 expression begins in the lens placode at E9.5, facilitating lens vesicle closure and separation from the ectoderm. This expression continues through the lens vesicle stage at E12.5, playing a crucial role in balancing the proliferation and differentiation of lens fiber cells and the anterior lens epithelium into adulthood.³⁶

The observed findings are consistent with other Foxe3 mutant lines, including the homozygous dyl/dyl mutant^36,37 and engineered null mutants from 129 × C57Bl/6³⁸ and Balb/c mice,³⁶ which also display severe lens abnormalities and eye growth defects such as microphthalmia. A common phenotype among these mutants is a persistent connection between the lens and cornea due to delayed lens vesicle closure, resulting in a lens with reduced size, irregular shape, and structural disorganization with vacuoles. Although fiber elongation and crystallin expression initiate normally, the number of fibers drastically declines, leading to cataracts and, in some cases, fiber expulsion through the ectodermal connection. Both dyl/dyl and null mutants show impaired lens epithelium proliferation and failed lens fiber differentiation.^36,38 Notably, the E14.5 anterior epithelium in mutants displays many apoptotic cells, unlike wild-type lenses. Similar findings in rct mice,^11,39 which carry a 22-bp deletion in the Foxe3 regulatory region, demonstrate mild microphthalmia and significant lens degeneration.

The first phenotypic deviation in Foxe3 mutants is the delayed separation of the lens vesicle from the ectoderm and, secondarily, anterior lens epithelium disorganization, leading to lens degeneration. However, the mechanism linking this degeneration to microphthalmia remains unclear. Possible explanations include the larger size of the lens in mice compared to humans, suggesting that its absence could collapse the eye and cause microphthalmia; however, cases of aphakia without microphthalmia have been observed in humans.⁹ The persistent ectodermal connection may reduce intraocular pressure, limiting eye growth, yet this connection is also found in some compound heterozygous Foxe3rv/fs individuals without microphthalmia (Fig. 2B, g). The Foxe3 mutant model resembles cavefish (Astyanax mexicanus), where early lens apoptosis leads to complete eye degeneration.^40–42 Although alpha-A crystallin deregulation may play a role,⁴³ the genetic basis for lens apoptosis in that species remains unknown.

In all vertebrates, optic vesicle folding coincides with lens vesicle formation, and their interaction is essential for proper eye morphogenesis.¹ Lens–retina crosstalk, mediated by factors and pathways (e.g., BMP, FGF, Wnt), is critical for eye development.¹ Studies have explored interactions among the lens placode and optic vesicle, the lens vesicle and retina, and periocular mesenchyme and neural crest cells. PAX6 is essential for lens placode development but not for optic vesicle induction.^44–46 The “induction” hypothesis by Spemann⁴⁷ is being reevaluated, with evidence showing optic cup formation can occur without lens induction.² The presence of optic cups in aphakic patients further supports this.⁷ The lens, evolving secondarily, may primarily enhance optic cup invagination and formation.² Our study proposes a model of eye growth defects wherein initial developmental cues successfully establish ocular structure formation. However, as development advances, “inductive” processes are supplanted by “involutive” ones, culminating in lens degeneration and disrupted ocular growth. This raises the question of whether complete eye involution (as in cavefish) and incomplete involution with retinal hyperproliferation (other mouse models⁴⁶) involve the same or distinct mechanisms leading to eyeball degeneration.

Due to challenges in accessing and culturing embryonic lens epithelial cells, we selected the different cell types to investigate transcription factor binding to the noncoding region surrounding the noncoding rs745674596G>A allele variant. Additionally, we immortalized the human epithelial lens cell line to ensure our study encompassed relevant human lens biology. DNA-PD and mass spectrometry using these cell lines revealed a limited set of 34 candidate transcription factors, which displayed differential binding in the presence and absence of the variant. To enhance selectivity, we focused on transcription factors shared among the lines or known to play a role in eye development, resulting in a shortlist of three candidates. Notably, none of them were identified through in silico analysis using chromatin immunoprecipitation sequencing data from the ENCODE project or JASPAR predicted transcription factor binding sites. The siRNA knockdown analysis indicated that only USF2 influenced Foxe3 expression. The other two candidates, particularly CNBP—associated with microphthalmia in mice³¹—may still play a role in lens development in vivo, especially given their expression patterns during ocular development. Both colocalize with Foxe3 in the lens vesicle during early stages and later in the lens epithelium, suggesting a potential overlap in function. However, further investigations are needed.

Foxe3 expression decreased upon Usf2 inhibition in both wild-type and rv/rv cells, with statistical significance achieved in wild-type but not in rv/rv. This finding aligns with DNA pull-down assay results, which showed preferential binding of USF2 to the wild-type sequence compared to the rv variant. The USF2 gene encodes a transcription factor from the basic helix-loop-helix leucine zipper (bHLH-LZ) family.⁴⁸ This protein plays a role in facilitating transcription by interacting with pyrimidine-rich initiator (Inr) elements⁴⁹ and E-box motifs (CACGTC).^50,51 Interestingly, the rs745674596 G>A variant specifically alters the E-box motif within the FOXE3 regulatory region, potentially reducing the binding affinity of USF2 to the DNA.

USF2 plays a critical role in regulating cell proliferation, differentiation, and apoptosis, processes essential for tissue development and maintenance.⁵² The Usf2 knockout mouse model exhibits growth delay, apathy, and increased postnatal lethality,^53,54 although ocular malformations have not been specifically investigated. Interestingly, Fujimi and Aruga⁵⁵ demonstrated USF2 expression in neural tissues, including the eye, during key stages of embryonic development in Xenopus laevis. Our findings similarly show that Usf2 is expressed in ocular structures during mouse development, indicating a possible link between USF2 and ocular anomalies. Moreover, the expression of USF2 is enriched expression in the mouse lens at various stages as per the lens expression database iSyTE.⁵⁶ Given its role in the development of neural tissues, including the eye,⁵⁵ disruptions in USF2 expression or function may contribute to ocular developmental defects. Although no direct link between USF2 mutations and human disease has been confirmed, exploring USF2 variants in individuals with ocular anomalies could yield important insights. Further investigations may uncover critical associations between USF2 and eye development or related disorders. Several truncating mutations in USF2 have been reported in large-scale genome databases (GnomAD), but only once at the homozygous state, providing support for its possible implication in human autosomal recessive diseases.

In conclusion, this study highlights the significance of highly conserved noncoding regions in understanding genetic diseases while emphasizing the challenges in interpreting variants within these regions. By focusing on genotype–phenotype correlations, we demonstrate their critical role in elucidating the impact of noncoding variants, particularly in single heterozygous cases of potential recessive diseases where biallelic mutations are absent, thus hampering optimal genetic counseling. Fully characterizing such variants requires a comprehensive approach that integrates in-depth knowledge of disease expression and genetics, along with multiple lines of experimental evidence of pathogenicity, rather than relying solely on standard molecular diagnostics. Furthermore, this study not only underscores the medical relevance of searching for pathogenic regulatory variations but also highlights the impact of these analyses on advancing our understanding of gene regulation and identifying novel candidate genes for genetic diseases. Finally, we provide the first description of a potential novel mechanism causing microphthalmia via lens degeneration, which carries significant therapeutic implications. Previously, treatment options were limited due to early ocular developmental defects; however, the discovery of a degenerative mechanism opens new possibilities for intervention, especially if it occurs later in pregnancy.