Bladder exstrophy (BE), a severe lower midline ventral defect, includes a spectrum of urological abnormalities in which all or part of the abdominal wall fails to close, resulting in the bladder and part of the urethra being exposed. Additional features of BE include pelvic bone malformations, widened pubic bones, cleft clitoris and shortened vagina in female infants, and epispadias in male infants (Ebert, Reutter, Ludwig, & Rosch, 2009; Reinfeldt Engberg, Mantel, Fossum, & Nordenskjold, 2016). The estimated prevalence of BE varies from 0.5–4.6 per 100,000 live births (Siffel et al., 2011). In most (Higgins, 1962; ICBDMS, 1987; Lattimer & Smith, 1966; Nelson, Dunn, & Wei, 2005; Reinfeldt Engberg et al., 2016; Reutter et al., 2011; Shapiro, Lepor, & Jeffs, 1984; Yang et al., 1994), but not all studies (Nelson et al., 2005; Yang et al., 1994), male predominance has been reported, ranging from 1.5–6.0:1.

The etiology of BE remains largely unknown; however, recent findings for familial cases (Froster, Heinritz, Bennek, Horn, & Faber, 2004; Kajbafzadeh, Tajik, Payabvash, Farzan, & Solhpour, 2006; Reutter, Hoischen, et al., 2007; Reutter, Shapiro, & Gruen, 2003), from analysis (Messelink, Aronson, Knuist, Heij, & Vos, 1994; Reutter et al., 2003; Shapiro et al., 1984) of twin studies (Reutter, Qi, et al., 2007), and of reports of BE in individuals with trisomy 21 continue to support an underlying genetic susceptibility (Ebert et al., 2009; Reutter, Betz, Ludwig, & Boemers, 2006; Reutter et al., 2009). More recently, this susceptibility has been supported by findings from array-based approaches, including common variants identified through genome-wide association studies (Draaken et al., 2015; Reutter et al., 2014; Zhang, Knapp, Kause, Reutter, & Ludwig, 2017) and rare variants identified through copy number analyses (Beaman et al., 2019; Draaken et al., 2015; Draaken et al., 2013; Draaken et al., 2010; Jorgez et al., 2014; Lundin et al., 2010; Soderhall et al., 2014; von Lowtzow et al., 2016). Building on investigations for rare variants for BE, we conducted exome sequencing on a modestly sized, but well-characterized, population-based sample of children with BE and their parents. Our approach provides additional data on rare variants that may contribute to the etiology of BE and provides sequencing data for use in future studies.

Among the 26 live born children with BE, 18 (69.2%) were males, 23 (88.4%) were full term, and all were singleton births. Among case mothers, age at delivery was nearly equally divided between 20–29 and 30–39 years (50.0% and 46.2%, respectively), and most were non-Hispanic white (73.1%) (Table 1a). Several children had additional malformations (Table 1b).

Using our primary criteria for variant prioritization, two genes emerged. Five children inherited heterozygous variants, two in the TUBE1 gene and three in the TSPAN4 gene (Table 2a). The variants identified in TUBE1 (c.1093A>T, p.R365X; c.673dupG, p.V225fs) were loss-of-function variants, with each predicted to lead to absence of the protein product due to nonsense mediated decay. Of the three children with putatively functional variants in TSPAN4, one had a loss-of-function variant (c.411_432+3del, p.A137fs), whereas the other two had missense variants (c.491G>A, p.R164Q; c.698C>G, p.A233G) (Table 2a). One of these missense variants (c.491G>A, p.R164Q) lies within the tetraspanin extracellular EC2 domain, as annotated in the InterPro database (Mitchell et al., 2019). Each of the five variants detected in TUBE1 and TSPAN4 have mappability scores of 1 (Derrien et al., 2012). All children with TUBE1 and TSPAN4 variants were of non-Hispanic white race/ethnicity. The GDI percentile of TUBE1 is 49.1% suggesting that nearly half of all human genes are less tolerant to damaging variation than TUBE1. Other bioinformatic algorithms reveal conflicting evidence regarding the relevance of loss-of-function variants in TUBE1. Two algorithms suggest that heterozygous loss-of-function variants in this gene could lead to haploinsufficiency. The gnomAD loss-of-function observed/expected (o/e) constraint score was below 1 (0.31 [90% CI: 0.17 – 0.58]) (K. J. Karczewski et al., 2020; Lek et al., 2016) indicating that fewer than expected loss-of-function TUBE1 variants were found in gnomAD (Konrad J. Karczewski et al., 2019; Lek et al., 2016). In addition, the percentile of the predicted haploinsufficiency score (Huang, Lee, Marcotte, & Hurles, 2010) of TUBE1 was low (14.86%, range: 0 – 100; with 0 being more likely to cause haploinsufficiency) indicating that heterozygous loss-of-function variants are more likely to lead to a severe phenotype. However, gnomAD’s probability of loss of function intolerance (pLI) was only 0.03 (ranges from 0 to 1; 0 being more likely to be tolerant to loss-of-function variation), which contradicts the two scores above. TSPAN4 has a GDI percentile of 40.3%, a gnomAD loss-of-function o/e constraint score of 0.87 (0.56 – 1.41), a gnomAD missense o/e constraint score of 1.09 (0.96 – 1.24), haploinsufficiency score of 71.26%, and a gnomad pLI of 0. Therefore, loss-of-function and potentially damaging missense variants in TSPAN4 are predicted to be well-tolerated.

Table 2b lists inherited variants near or in genes previously reported to be associated with BE in the literature, WNT3, CRKL, LZTR1, MYH9, ISL1, TP63, and SLC20A1. We identified four variants – one per child and one each in four of these genes (WNT3, CRKL, LZTR1, MYH9). Missense variants were identified in three children: c.286G>C, p.D96H in WNT3, c.76A>G, p.T26A in CRKL and c.4958A>G, p.E1653G in MYH9 (Table 2b); one child had a nonsense variant (c.628C>T, p.R210X) in the LZTR1 gene.

Our investigation of variants using autosomal recessive or X-linked recessive models (Tables 3 and 4 respectively) identified 3 missense and 13 loss-of-function variants. We also identified 17 de novo missense and 3 de novo loss-of-function variants (Table 5). Two of the de novo variants (in AKR1C2 and USP18) were flagged as being potentially false positive by GATK due to mapping differences between reads with and without alternate alleles; however, both were confirmed by Sanger sequencing. All 20 de novo variants passed manual review in IGV, including these two variants. Screenshots of all variants reviewed in IGV can be found in the Supplemental Table 1. The loss-of-function variants were found one each in AKR1C2, PRRX1, and PPM1D; this last variant occurred in a child classified clinically as ‘multiple’ (congenital anomalies of the skeletal system in addition to bladder exstrophy).

We detected inherited compound heterozygous loss-of-function variants in PLCH2 and CLEC4M that met our primary screening criteria (Supplemental Table 2). Other compound heterozygous variants of interest identified are listed in Supplemental Table 2. Variants in additional genes that met our secondary inclusion criteria are listed in Supplemental Table 3. Of note, we identified one missense variant in CD151 in a consanguineous family (Supplemental Table 4). CD151 is a member of the tetraspanin family, which creates complexes with integrins and other molecules and is involved in cell adhesion. Many of the children in our study had more than one variant (see case child numbers in tables), which may suggest an oligogenic model of inheritance for BE.

We identified a total of seven rare autosomal CNVs with length greater than 100kb in six children (three deletions and four duplications). All seven were predicted to be pathogenic by AnnotSV. None of the regions with CNVs were identified in more than one child. Six of the CNVs were inherited and one was a de novo deletion in child 10 (child “10” had an additional paternally inherited duplication). A paternally inherited duplication in child “26” was found within 22q11.21 (chr22:17400001–21700000 on the hg38 reference genome) that has previously been associated with bladder exstrophy (among the genes overlapped by the duplication in child “26” are CRKL, THAP7, and LZTR1) (Figure 1). Additional information for all seven CNVs (e.g. the breakpoints from ExomeDepth and annotations from AnnotSV), the images used to manually review call quality and screenshots from the UCSC Genome Browser (Kent et al., 2002) (http://genome.ucsc.edu) can be found in supplementary file 2.

A subset of 13 variants was selected for confirmation via Sanger sequencing. This subset included 1) the five variants in TUBE1 and TSPAN4 (Table 2a) selected using our primary criteria for candidate variants; 2) the five variants in WNT3, MYH9, LZTR1, THAP7, and CRKL (Table 2b) selected using our criteria for known genes; 3) an additional variant in MYH9 (rs768181629) that we suspected to be an artifact and was not described in the manuscript but was still nominated for validation because it is in a gene with prior evidence of association; and 4) two de novo variants in AKR1C2 and USP18 that were flagged by GATK. All five variants in TUBE1 and TSPAN4, plus variants in MYH9 (rs780486984), WNT3 and LZTR1 were validated in probands and the parent from which the variant was inherited. The CRKL variant was validated in the proband, but parent DNA was not available for validation studies. THAP7 and MYH9 rs768181629 variants were the only two that did not validate; both were lower quality de novo variants with low alternate allele counts. The de novo variants in AKR1C2 and USP18 were confirmed in probands and were absent from both parents.

There have been numerous postulated mechanisms for the pathogenesis of BE; cell migration and cell adhesion defects are two of the theories that have been proposed (Martinez-Frias, Bermejo, Rodriguez-Pinilla, & Frias, 2001; Sadler, 2010; Sadler & Feldkamp, 2008). We identified variants in TUBE1, which is implicated in cell migration (Chang, Giddings, Winey, & Stearns, 2003; Garcin & Straube, 2019), and TSPAN4, which is involved in both cell migration (Berditchevski, 2001) and cell adhesion (Jiang et al., 2019; Levy, Todd, & Maecker, 1998). In two children, we identified heterozygous predicted loss-of-function variants in TUBE1, including a stop-gain mutation and an insertion leading to a frameshift mutation. The protein epsilon tubulin 1, encoded by TUBE1, is localized to the centrosome (Chang & Stearns, 2000). Directed cell migration requires polarization of cells via the microtubule-organizing center (MTOC) (centrosome) and its disruption results in loss of cell polarity (Cheng et al., 2019). TUBE1 is a gene crucial for the organization of the microtubules during centriole duplication. Members of this superfamily are essential in maintaining the microtubule cytoskeleton (Chang et al., 2003). One theory is that BE is caused by failure of one or both lateral body wall folds to move ventrally to meet in the midline resulting in internal organs, such as the bladder, protruding through the defective region (Duhamel, 1963; Feldkamp, Carey, & Sadler, 2007; Sadler & Feldkamp, 2008). It is known that dorsal movement of the neural folds to form the neural tube is driven in part by alterations in the shaping and position of the cells. (Colas & Schoenwolf, 2001; Lee & Nagele, 1985; Sadler, 2005); we hypothesize that the same may potentially be true for the ventral abdominal wall (Lee & Nagele, 1985; Sadler & Feldkamp, 2008). The hypothesis that TUBE1 variants could be possibly be associated with defective ventral body wall closure is also supported by the finding that the cadherin receptor gene CELSR3, a planar cell polarity core gene which, like TUBE1 is related to centriole positioning (Carvajal-Gonzalez, Mulero-Navarro, & Mlodzik, 2016; Su et al., 2016), has been suggested as a potential BE-associated gene (Reutter et al., 2016).

Disrupted fusion of the lateral body folds, once they meet in the midline, is another theory to explain failure of ventral body wall closure. Functional variants in the TSPAN4 gene might contribute to BE because of disrupted fusion. Two out of the three TSPAN4 variants were missense variants, and the other a loss-of-function variant. TSPAN4 belongs to the tetraspanin superfamily. Tetraspanins are known to play a role in cell adhesion and motility (Jiang et al., 2019; Levy et al., 1998) and form complexes with adhesion receptors from the integrin family, another mediator of cell adhesion (Berditchevski & Odintsova, 1999; Maecker, Todd, & Levy, 1997). In attached cells, tetraspanins remain concentrated within adhesion structures that morphologically are similar to Rac-dependent peripheral focal complexes, which stabilize adhesive interactions of the cell front (Berditchevski & Odintsova, 1999; Nakamura, Iwamoto, & Mekada, 1995; Nobes & Hall, 1995). Tetraspanins may also be involved in either post-adhesion signaling or affect integrin-mediated cell attachment directly (Shaw et al., 1995; Yanez-Mo et al., 1998). In other tissues that fuse during embryogenesis, such as closure of the lip and palate and the neural tube, adhesion is initiated by cell surface molecules, such as glycoproteins, and more permanent adhesion between the two folds takes place through new cell-to-cell contact (Burk, Sadler, & Langman, 1979; Colas & Schoenwolf, 2001; Greene & Pratt, 1976; Sadler, 2005). Although the exact process by which the ventral wall fuses is unknown, our data suggest that the function of TSPAN4 in cell adhesion could potentially be involved in fusion of the lateral folds.

Tetraspanins are also involved in cell migration. The human TSPAN4-like membrane protein (TM4SF1), is crucial for nanopodia formation. Nanopodia have been shown to play an important role in human endothelial cell polarization and migration, and their formation is dependent on TM4SF1 expression, as demonstrated by their dose-dependent increase or reduction in various cell lines (e.g. fibroblasts, melanocytes) (Zukauskas et al., 2011). In the Xenopus model, Tspan4 has been shown to be expressed during gastrulation in the dorsal and ventral mesoderm (Kashef, Diana, Oelgeschlager, & Nazarenko, 2013); in humans, lateral body folds are formed from the parietal layer of lateral plate mesoderm and overlying ectoderm (Sadler & Feldkamp, 2008). We hypothesize that TSPAN4 may play a role in both cell adhesion and cell migration.

The gnomAD loss-of-function o/e constraint score suggests that healthy individuals do have loss-of-function variants in TUBE1 but that they are observed less often than expected in healthy individuals. In addition, a haploinsufficiency score (Huang et al., 2010) for this gene was low, suggesting that heterozygous loss-of-function variants in TUBE1 are more likely to lead to a severe phenotype. Variants in TSPAN4 are predicted to be better tolerated. It is possible that the inherited variants in these two genes in children in our cohort may have occurred by chance. Therefore, functional follow-up would be needed to establish a causal relationship with bladder exstrophy for either gene. In addition, no pathogenic deletions or duplications of either gene for BE were noted in ClinVar, ClinGen, OMIM, dbVar, or DECIPHER.

We identified and confirmed variants in WNT3, CRKL, LZTR1, and MYH9 – four of the eight genes previously reported to be associated with BE – in four patients. Each gene was reported in association with BE as part of a deletion or insertion in the gene region, or as a point mutation. The association of BE with a variant near WNT3 was suggested by previous copy number variant (CNV) analysis and supported by mouse expression data and genome-wide association studies (Reutter et al., 2014). WNT3 variants have been associated with other defects (Niemann et al., 2004), whereas WNT is related to persistent cloaca and p63 that is a WNT suppressor has been associated with human bladder exstrophy-epispadias complex (Ching et al., 2010; Katoh et al., 2016; Mahfuz, Darling, Wilkins, White, & Cheng, 2013; Qi et al., 2011).

Duplications of the 22q11.21 region have been associated with classic BE (Beaman et al., 2019; Draaken et al., 2014; Draaken et al., 2010; Lundin et al., 2010). Genes included in the duplication region – Crkl, and Lztr1 – have been suggested as candidate genes for BE in mouse studies (Draaken et al., 2014). We identified rare sequence variants in each of these two genes, including a rare nonsense variant in LZTR1 also seen in the autosomal recessive form of Noonan syndrome (Johnston et al., 2018). If a duplication of a large region like this increases the risk of disease, it likely has a gain-of-function effect. The missense variant we have identified in CRKL could also have a gain-of-function effect, but the loss-of-function allele (stop gain mutation in the middle of the gene) in LZTR1 would be expected to have the opposite effect.

We confirmed a MYH9 missense variant in one child and father. MYH9 encodes a non-muscle myosin heavy chain IIA essential for cell adhesion in cell migration during embryonic development. Interestingly, another de novo missense variant in this gene was previously reported as the causative gene in an individual with BE and congenital macrothrombocytopenia. (Utsch et al., 2006).

Lastly, we identified a de novo stop-gain mutation in exon 6 of PPM1D. Stop-gain variants in the last exons of this gene (exons 5 or 6) have been reported to cause an intellectual disability syndrome along with facial dysmorphology, short stature, small hands and/or feet and hyperlordosis (Jansen et al., 2017). The child sequenced in our study reportedly had craniofacial dysmorphology and skeletal anomalies that were not reported by Jansen et al. To the best of our knowledge, this is the first case with a PPM1D mutation reported to have BE.

A primary strength of our analysis is the use of a trio-based design to discern between inherited versus de novo variants. Other strengths include that children were systematically ascertained for the NBDPS from population-based surveillance programs, minimizing missed cases within each site’s catchment area, and medical record information for each child was reviewed by clinical geneticists, producing a well-characterized study sample. In addition, we expanded investigation of rare variants by applying exome sequencing. Limitations of our study are the modest sample size with only a subset of eligible children providing DNA and limited racial/ethnic diversity among families that provided DNA. Nonetheless, these families were from a population-based sample and well-characterized. Another limitation was the use of exome sequencing rather than whole genome sequencing, which prevented us from studying non-coding regions; however, these exome data identified genes that could potentially be candidates for disease association and contribute to understanding the genetic etiology of BE. Lastly, we did not screen additional case populations for the variants detected because they were all rare, and therefore, even if they are causal, the likelihood of detecting them in additional populations was low. Functional studies were not conducted because none of the variants were recurrent. Despite these limitations and the potential for chance findings, our study suggests new hypotheses and our data can contribute to more large-scale efforts to characterize genetic susceptibility of BE.

Additional considerations regarding our findings included that the heterozygous variants we found in TUBE1 and TSPAN4 were each inherited from one of the unaffected parents. Some of the alleles in these two genes are present in public databases in a higher frequency than the disease prevalence. As such, we hypothesize that incomplete penetrance or other genes may be involved in BE etiology. Also, some children carried multiple variants. For example, five variants were seen in child 1 and seven variants were detected in child 8; therefore, although they were potential candidates based on initial screening, a number of the candidate variants are likely to be rare variants not associated with BE.

In conclusion, our study identified variants in several genes that may potentially contribute to BE. Our results underscore the potential role of genetic factors in the pathogenesis of BE and provide important clues for future investigations.