EPHB4 kinase–inactivating mutations cause autosomal dominant lymphatic-related hydrops fetalis

Hydrops fetalis describes fluid accumulation in at least 2 fetal compartments, including abdominal cavities, pleura, and pericardium, or in body tissue. The majority of hydrops fetalis cases are nonimmune conditions that present with generalized edema of the fetus, and approximately 15% of these nonimmune cases result from a lymphatic abnormality. Here, we have identified an autosomal dominant, inherited form of lymphatic-related (nonimmune) hydrops fetalis (LRHF). Independent exome sequencing projects on 2 families with a history of in utero and neonatal deaths associated with nonimmune hydrops fetalis uncovered 2 heterozygous missense variants in the gene encoding Eph receptor B4 (EPHB4). Biochemical analysis determined that the mutant EPHB4 proteins are devoid of tyrosine kinase activity, indicating that loss of EPHB4 signaling contributes to LRHF pathogenesis. Further, inactivation of Ephb4 in lymphatic endothelial cells of developing mouse embryos led to defective lymphovenous valve formation and consequent subcutaneous edema. Together, these findings identify EPHB4 as a critical regulator of early lymphatic vascular development and demonstrate that mutations in the gene can cause an autosomal dominant form of LRHF that is associated with a high mortality rate.


Introduction
Hydrops fetalis decribesis defined as excessive fluid accumulation or edema in at least two fetal compartments. Non-immune hydrops fetalis is the cause in more than 85% of cases, of which 15% have been reported to have a lymphatic related abnormality (1). In 20% of nonimmune hydrops fetalis cases the cause is not known. Lymphatic-related (non-immune) hydrops fetalis (LRHF) has been included in a subgroup of primary lymphedemas under the umbrella term Generalised Lymphatic Dysplasia (GLD) by Connell et al. (2). In this classification, GLD was defined as lymphedema associated with systemic or visceral involvement (including hydrops fetalis), even if the lymphedema was not widespread. The GLD group includes patients with a widespread developmental abnormality of the lymphatic system, often presenting prenatally with hydrothoraces or non-immune hydrops fetalis.
Hennekam syndrome (OMIM#235510) is an example of a GLD which is inherited in an autosomal recessive manner. Mutations in collagen and calcium binding EGF domain 1 (CCBE1) and FAT atypical cadherin 4 (FAT4) have been identified as causal (3)(4)(5). Another recessively inherited form of GLD with a high incidence of LRHF has recently been reported as caused by mutations in piezo-type mechanosensitive ion channel component 1 (PIEZO1) (6), adding to the genetic heterogeneity of the GLD group.
We have recently ascertained two families with a history of non-immune hydrops and postnatal lymphatic dysfunction, but with a pattern suggestive of autosomal dominant inheritance, and with sporadic occurrence in one of the families. We have identified heterozygous inactivating mutations in the kinase domain of EPHB4 as causative for this condition. This suggests not only is LRHF/GLD genetically heterogeneous, but it should also be considered in both dominant and recessive forms. The importance of Ephb4 for lymphatic vascular development is further supported by analysis of a genetic mouse model with lymphatic endothelial specific deletion of Ephb4.

Results
Genetic analysis of LRHF identifies causative mutations in EPHB4. We report two multigenerational families (one from Norway, GLDNOR, and one from the UK, GLDUK) ( Figure   1). Clinical findings in these families include antenatal non-immune hydrops fetalis or bilateral hydrothoraces, and neonatal chylothoraces of variable severity (Table 1)  Sanger sequencing identified no pathogenic variants in the genes known to be associated with congenital primary lymphedema (i.e. CCBE1, VEGFR3 and VEGFC) in the UK proband (GLDUK:I.2). Whole-exome sequencing (WES) was performed on the family to identify pathogenic variants. When filtering the WES data of the UK family for variants in genes known to have relevance to the lymphatic system and lymphangiogenesis, an unreported variant (NM_004444.4: c.2216G>A, p.Arg739Glu) in EPHB4 was identified. Initially, it was thought that the variant did not fully co-segregate with the disorder status in the family ( Figure 1A), as two clinically unaffected family members were found to carry the variant (GLDUK:II.4 and GLDUK:III.2). Neither of these presented with hydrops fetalis but both were later diagnosed with an ASD. The next step of the analysis was to apply a specific autosomal dominant inheritance filter criterion and MIER2 was the only gene that fulfilled that ( Figure   1A). However, the MIER2 variant (NM_017550.1: c.865C>T, p.Arg289Trp) has been reported as a SNP (rs148482834) with a heterozygous genotype observed at a frequency of 0.001.
Meanwhile, an independent study of a Norwegian GLD family was undertaken. In this family, the condition initially appeared to be sporadic in monozygotic twins (GLDNOR:II.2 and II.3), who both had subcutaneous edema at birth that resolved in infancy (Table 1). GLDNOR:II.3 required ventilation and thoracentesis for bilateral chylothoraces. Both sisters had sons with non-immune hydrops, one died at 1.5 days of age, the other was moribund in the neonatal period but the edema eventually resolved. Both sons also had an ASD. Three genes (PTPN11, FOXC2 and VEGFR3) were ruled out in GLDNOR:II.2 by Sanger Sequencing, and a high resolution microarray CGH (Affymetrix 6.0) was normal in GLDNOR:II.3. When analysing the WES data, the only gene to fulfil an autosomal dominant model with de novo occurrence was EPHB4. Sanger sequencing of additional family members showed all affected family members to carry the variant (c.2345T>G, p.Ile782Ser) ( Figure 1B). GLDNOR:II.2 and II.3 were bout found to be mosaic for this variant and GLDNOR:II.2 had a lower mutation load than her twin sister, GLDNOR:II.3, consistent with her milder neonatal presentation (Supplemental Figure 1).
As MIER2 had shown perfect co-segregation in GLDUK, the WES data of GLDNOR:III.5 was scrutinised but no variants were found in MIER2 and the coverage was found to be of sufficient depth and quality. For good measure, GLDNOR:III.5 was also screened by Sanger sequencing for all exons of MIER2; still no variant was identified. Neither EPHB4 variants had been previously reported in public databases or in 900 in-house controls and it was therefore concluded that mutations in EPHB4 are the likely cause of the LRHF/GLD phenotype seen in these two families despite the variable expression observed.
LRHF associated mutations lead to inactive EPHB4 kinase. EPHB4 binds the transmembrane ephrinB2. Binding of ephrinB2 to EPHB4 stimulates phosphorylation and activates downstream signalling cascades (7,8). The two EPHB4 mutations (p.Arg739Glu and p.Ile782Ser) occur at highly conserved residues located in the tyrosine kinase domain of the EPHB4 protein (Supplemental Figure 2 and 3A). Moreover, p.Arg739Glu is located within the catalytic loop HRD (His-Arg-Asp) motif, also highly conserved in many tyrosine kinases (Supplemental Figure 3B Ephb4 deficiency in mice results in subcutaneous edema and abnormal lymphatic development. EphrinB2 -EphB4 signalling is critically required for the development of the cardiovascular system during early embryogenesis (9,10). EphrinB2 and EphB4 are also essential for lymphatic vessel remodelling and valve formation during late embryonic and early postnatal development (11,12), but whether they have an earlier role in lymphatic vessel morphogenesis that could explain Ephb4 loss of function induced LRHF/GLD in humans is not known. Whole mount immunofluorescence analysis confirmed the previously reported venous and lymphatic endothelial specific expression of EphB4 in embryonic skin and mesenteries (Supplemental Figure 4A and B). To assess the potential contribution of lymphatic endothelial loss of EphB4 function to LRHF/GLD, we deleted Ephb4 specifically in the lymphatic vasculature using tamoxifen inducible Prox1-CreER T2 mice crossed with a conditional Ephb4 flox line ( Figure 5A, Supplemental Figure 5). The mice were further crossed with the R26-mTmG double reporter to monitor Cre activity, and to label gene-deleted cells with green fluorescent protein (GFP). Ephb4 deletion was induced from the earliest stage of lymphatic development by administration of 4-hydroxytamoxifen (4-OHT) for five consecutive days starting at E10.5 ( Figure 5A). At E15.5 a high proportion of mutant embryos showed subcutaneous edema ( Figure 5B, and Figure 5C, left panel). In addition, a proportion of dermal lymphatic vessels contained blood in 71% of edematous mutant embryos (n=14), but not in non-edematous mutants (n=5) or in control embryos (n=20) ( Figure 5C and data not shown). Whole-mount immunofluorescence of the skin revealed tortuous and dilated dermal lymphatic vessels in the Ephb4 mutants ( Figure 5C, right panels). Notably, abnormal vessel morphology was also observed in vessels that showed a low contribution of GFP + (i.e. Ephb4 deficient cells) ( Figure 5C), suggesting that edema and/or blood filling of lymphatic vessels secondarily caused vessel dilation. In support of a non-cell-autonomous effect of early embryonic deletion of Ephb4 to dermal lymphatic vasculature, inactivation of Ephb4 from E12.5, when dermal lymphatic vessel formation begins (13) ( Figure 5A), resulted in normal vasculature despite efficient gene targeting ( Figure 5D). These results suggest E10-E12 as a critical time-window for EphB4 function during lymphatic development.

Ephb4 is required for the formation of lymphovenous and lymphatic valves. Previous studies
have shown that between E10.5-E13.5 formation of specialized lymphovenous valves (LVV) occurs at the connection sites between the primordial thoracic duct (pTD) and the cardinal vein (14-16) ( Figure 6A). It was therefore reasoned that edema in Ephb4 mutants might be due to defective LVVs leading to inefficient lymph drainage. To investigate this, we induced Cre recombination in the developing LVVs in Ephb4 flox ;R26-mTmG;Prox1-CreER T2 embryos by 4-OHT treatment between E10.5-E12.5. Analysis of immunostained transverse vibratome sections of E13.5 control embryos showed preferential and efficient targeting of the dual LVVs by the Prox1-CreER T2 transgene, while pTD endothelium exhibited mosaic labeling ( Figure 6B). Control LVVs (11 out of 11) consisted of two well-defined leaflets extending to the lumen of the cardinal vein ( Figure 6C and 6D, Supplemental Movie 1). In contrast, the majority of Ephb4 deficient LVVs (9 out of 13) did not show extended leaflets but instead consisted of abnormal clusters of GFP + cells ( Figure 6C and 6D, Supplemental Movie 2).
Interestingly, studies using a function blocking antibody and a chemical genetic approach showed that EphB4 kinase signalling regulates lymphatic valve formation (11), while genetic studies have demonstrated an important function for its ligand EphrinB2 in the formation of both lymphatic and venous valves (12,17). Using our genetic loss-of-function model we confirmed the essential role of EphB4 in lymphatic valve morphogenesis. Deletion of Ephb4 during embryonic valve formation led to a complete absence of valves that form in control mesenteries by E18.5 by lymphatic endothelial cells (LECs) expressing high levels of Prox1 (Supplemental Figure 4C). In addition, early postnatal deletion of Ephb4 led to a complete loss of lymphatic valves (Supplemental Figure 4D). These results demonstrate a critical role of Ephb4 in the formation and early postnatal maintenance of lymphatic valves, and highlight conserved mechanisms regulating the formation of valves at different anatomical sites.

Discussion
This study identifies the EPHB4 receptor tyrosine kinase as a critical regulator of early lymphatic vessel development and a novel causative gene for LRHF and primary lymphedema. We have shown here that kinase inactivating mutations in EPHB4 can produce a lymphatic phenotype in humans that can be classified as LRHF/GLD. However, this phenotype shows highly variable expression. Some individuals present with severe in utero swelling, which may cause perinatal demise (or fully resolve to become completely asymptomatic), others with no edema but only an atrial septal defect. It can be distinguished from the majority of Hennekam Syndrome cases, where the swelling presents in the antenatal period but persists throughout life (3)(4)(5). The large number of miscarriages in GLDNOR may well be related to this disorder. In this regard, it is of interest that EphB4 and ephrin-B2 have been shown to be instrumental in the human placental development (18).
Invasive cytotrophoblasts uses the EPHB4 expression on veins to ensure that migration of these cells into EPHB4 expressing uterine veins is limited and instead biased toward the arterial side of the circulation (19). Expression of EPHB4 at half the levels normally encountered may disturb the complex migration patterns seen in the process of placentation. A failure of the invasive cytotrophoblasts to take on an arterial phenotype is suspected to lead to the loss of pregnancy during the late first or early second trimester (19). Perinatal deaths were also of a higher frequency in the autosomal recessive form of LRHF/GLD caused by PIEZO1 mutations but, in this condition, were probably related to the hydrops fetalis (6). imaging is suggestive of rerouting through skin and superficial tissues rather than a main lymphatic tract as seen in the control ( Figure 2B). She has a small ASD and interestingly her son had large, multiple ASDs requiring surgical closure. Variable expression has been observed in other primary lymphedemas e.g. PIEZO1-related LRHF/GLD (6).
Like other forms of GLD (4,6), this novel condition presents antenatally with non-immune hydrops or pleural effusions. The swelling may completely resolve with no residual lymphatic phenotype, similar to observations in the recently identified PIEZO1 related GLD (6). However, the report of one affected individual with bilateral lower limb edema (GLDUK:I.2) with abnormal lymph scans suggests that there may be residual, lymphatic weakness in the survivors. Further studies will be needed to investigate the specific nature and extent of the lymphatic dysfunction in these patients.
LEC specific deletion of Ephb4 in mouse embryos led to subcutaneous edema and abnormal lymphatic vessel morphology, thus recapitulating aspects of the human LRHF phenotype.
Temporal analysis of Ephb4 function demonstrated a critical requirement of Ephb4 during early stages of lymphatic development. Specifically, we found that Ephb4 regulates the formation of lymphovenous valves that are critical for efficient lymphatic function by maintaining unidirectional flow of lymph into blood (14,20,21). We additionally confirmed the previously reported critical role of Ephb4 in both formation and maintenance of lymphatic valves (11). Lymphatic valve defects are, however, an unlikely cause of in utero swelling due to their late embryonic development (22,23). We postulate, therefore, that defective lymphovenous valve formation, caused by the lack of EPHB4 could contribute to the LRHF seen in the GLDNOR and GLDUK patients. In agreement with this, defective LVVs were recently demonstrated in mouse models of primary lymphedemas caused by loss of function of Foxc2, Connexin37 and Gata2 (16,24).
Edema in the mouse embryos lacking Ephb4 specifically in the lymphatic endothelia appears to be milder than that observed in the patients, suggesting that defective lymphovenous valves may only partially explain human LRHF/GLD. It is well known that EphB4 is also expressed in venous and capillary endothelium (9,10), and therefore the impact of the mutations on the venous system needs to be considered. In accordance with this hypothesis Kinase activity is critical for EphB4 forward signalling in lymphatic endothelia (11). Our in vitro data show that EPHB4 mutants carrying the LRHF associated mutations p.Arg739Glu and p.Ile782Ser are kinase-dead but do not have a dominant negative effect on WT protein.
In contrast, VEGFR3 mutants in Milroy Disease have a slower turnover (27), which may affect the signalling capacity of the WT tyrosine kinase receptor due to accumulation of mutant receptors on the cell surface. Unlike typical receptor tyrosine kinases that dimerise upon ligand stimulation, Ephrin receptors form higher order clusters, with cluster size being an important determinant of the quality and strength of cellular response (28,29). Inclusion of a kinase-dead receptor may thus significantly weaken the signalling strength of higher order clusters and thereby alter cellular responses. EphB2 receptor-mediated endocytosis requires the kinase activity of the receptor (30), so that kinase-dead EphB4 could also show defective endocytosis, influencing clustering dynamics and cellular responses. Further studies will aim to investigate those and other functional consequences of the LRHF/GLD associated mutations.
In conclusion, we report on kinase inactivating EPHB4 mutations in eleven individuals from two extended family pedigrees presenting with a phenotypic spectrum from severe, lethal non-immune hydrops to ASD only. The inheritance pattern is typical of autosomal dominant inheritance with variable expression. Using a genetic mouse model, we have further shown that Ephb4 deficiency in lymphatic endothelium leads to defective lymphovenous valves, which may critically contribute to edema formation in LRHF/GLD patients. This is the first report in the literature of a human phenotype associated with EPHB4 mutations and also the first report of an autosomal dominant form of LRHF.

Methods
Exome sequencing. For GLDUK, sequencing libraries were made following the protocol from Roche/Nimblegen's SeqCap EZ Exome Library v2.0 kit. The libraries were then sequenced on HiSeq2000 (Illumina) machines. Sequence reads were aligned to the reference genome (hg19) using Novoalign (Novocraft Technologies). Duplicate reads, resulting from PCR clonality or optical duplicates, and reads mapping to multiple locations were excluded from downstream analysis. Depth and breadth of sequence coverage were calculated with custom scripts and the BedTools package (31).
All variants were annotated using a custom annotation pipeline. Single-nucleotide substitutions and small indel variants were identified and quality filtered within the SamTools software package (32) and in-house software tools (33). Variants were annotated with respect to genes and transcripts with the Annovar tool (34). Variants were filtered for novelty by comparing them to dbSNP135 and 1000 Genomes SNP calls and to variants identified in 900 control exomes (primarily of European origin), which were sequenced and analyzed by the same method. Summary statistics for the exome sequencing is given in Supplemental Tables 1 and 2 For GLDNOR the sequencing analysis was performed using the SOLID 5500xl platform (Life Technologies). Exon sequences were enriched by SureSelect Human All Exon v5 (Agilent Technologies), which targets ~21,500 human genes, covering a total of 50Mb of genomic sequence. Read alignment and variant calling were performed with Lifescope v1.3 software.
All variants were annotated using a custom annotation pipeline. Variants from the exome were filtered for known variants in dbSNP, intronic and UTR variants, synonymous variants and variants in our in-house database. Variants with less than 5 variation reads were also omitted. Summary statistics for the exome sequencing is given in Supplemental Table 3.
Confirmation Sequencing. Samples of available family members were analyzed by Sanger Sequencing. Primers were designed for the coding regions and associated splice sites for exons 13 and 14 of EPHB4 and all exons of MIER2 using Primer3 software (35) or ExonPrimer Mouse lines. R26-mTmG mice were acquired from the Jackson Laboratory (36). Prox1-CreER T2 mice were previously described (17). For the generation of the Ephb4 flox line, a conditional knock-out strategy, flanking Ephb4 exons 2 and 3 with LoxP sites, was used to target the Ephb4 locus. The targeting vector was built using homologous recombination in bacteria (37). A C57BL/6 mouse BAC, served as template for the extraction of the homology arms of the targeting vector. The targeting vector contained a frt flanked neomycin phosphotransferase, Neo, selectable marker cassette. After linearization, the targeting construct was electroporated into AZX1, a C57Bl/6JOlaHsd derived embryonic stem cell line.
PCR screens and Southern blot analyses revealed clones that had undergone the desired homologous recombination event. Several of these clones were expanded and injected into Balb/cOlaHsd blastocysts to generate chimeric males which were then bred to C57Bl/6JOlaHsd females and black-coated offspring were genotyped on both sides of the homology arms for correct integration into the EphB4 locus. The neomycin phosphotransferase selectable marker cassette, which was flanked by frt sites, was deleted after subsequent breeding to mice expressing flp recombinase under the CAG promoter. Lymphoscintigraphy. Lymphoscintigraphy is the imaging of the lymphatic system by injecting radioactive isotope (technetium-99m) into the web spaces between the toes and/or fingers and quantification of uptake into the inguinal lymph nodes for foot injections and axillary nodes for hand injection after 2 h with a gamma camera.
Statistical analysis. P values representing difference in proportion of edematous and nonedematous mutant versus control littermates ( Figure 5B) and proportion of normal and abnormal LVVs ( Figure 6D) were calculated using Chi square/Fisher's Exact (2-tailed). A P value less than 0.05 was considered significant.