Genome-wide analysis of copy-number variation in humans with cleft lip and/or cleft palate identifies COBLL1, RIC1, and ARHGEF38 as clefting genes

Summary

Cleft lip with or without cleft palate (CL/P) is a common birth defect with a complex, heterogeneous etiology. It is well established that common and rare sequence variants contribute to the formation of CL/P, but the contribution of copy-number variants (CNVs) to cleft formation remains relatively understudied. To fill this knowledge gap, we conducted a large-scale comparative analysis of genome-wide CNV profiles of 869 individuals from the Philippines and 233 individuals of European ancestry with CL/P with three primary goals: first, to evaluate whether differences in CNV number, amount of genomic content, or amount of coding genomic content existed within clefting subtypes; second, to assess whether CNVs in our cohort overlapped with known Mendelian clefting loci; and third, to identify unestablished Mendelian clefting genes. Significant differences in CNVs across cleft types or in individuals with non-syndromic versus syndromic clefts were not observed; however, several CNVs in our cohort overlapped with known syndromic and non-syndromic Mendelian clefting loci. Moreover, employing a filtering strategy relying on population genetics data that rare variants are on the whole more deleterious than common variants, we identify several CNV-associated gene losses likely driving non-syndromic clefting phenotypes. By prioritizing genes deleted at a rare frequency across multiple individuals with clefts yet enriched in our cohort of individuals with clefts compared to control subjects, we identify COBLL1, RIC1, and ARHGEF38 as clefting genes. CRISPR-Cas9 mutagenesis of these genes in Xenopus laevis and Danio rerio yielded craniofacial dysmorphologies, including clefts analogous to those seen in human clefting disorders.

The contribution of copy-number variants to cleft lip with or without cleft palate (CL/P) has been relatively understudied. Using a strategy to identify likely higher effect size microdeletions, we identify COBLL1, RIC1, and ARHGEF38 as genes associated with CL/P that play important roles in vertebrate craniofacial development.

Introduction

Cleft lip and/or cleft palate (CL/P) is a common birth defect occurring on average in one in every 1,000 live births.² Approximately 70% of all clefts are isolated occurrences (non-syndromic [NSCL/P]), with the remaining individuals presenting with additional clinical phenotypes (syndromic [SCL/P]).³ Studies sub-stratifying individuals with clefts by cleft type (cleft lip and palate [CLP], cleft lip only [CL], cleft palate only [CPO]), cleft laterality [unilateral, bilateral], or sidedness [left, right]) have demonstrated their overlapping and unique epidemiology.⁴ CPO occurs more frequently in females than males while CL/P has increased prevalence in males,⁵ and left-sided, unilateral clefts are the most common cleft type while bilateral clefts occur least frequently.⁶ Although NSCL/P and NSCL have historically been grouped as etiologically similar entities,^2,7 recent studies have identified genetic factors which contribute uniquely to each cleft subtype.^4,8,9,10,11

Copy-number variants (CNVs), defined as abnormal gains or losses of portions of chromosomal DNA greater than 1 kilobase (kb) in size, are common causes of disease,^{12,13,14,15,16,17,18} and CNVs carried by individuals with CL/P have begun to explain a portion of the missing heritability of clefting.^{19,20,21,22,23,24,25,26,27} Published estimates for the detection of pathogenic CNVs in individuals with SCL/P range from 21.4% (31/145)²⁸ to 60% (3/5),²⁷ while a detection rate of 7.2% (9/125)²⁸ has been cited for individuals with NSCL/P. However, due to the inherent heterogeneity of syndromes associated with CL/P, studying the CNV landscape of individuals with NSCL/P compared with that of SCL/P has been challenging, and to our knowledge CNV data from large NSCL/P cohorts have not been utilized to conduct comparative analyses within NSCL/P subtypes. One systematic review published in 2012 reported that clinically significant chromosomal defects detected using routine karyotyping or microarray were observed nearly exclusively in individuals with SCL/P in postnatal cohorts.²⁹ Several important limitations of the study noted by the authors include the inconsistent use of karyotyping versus microarray across clefting subgroups and studies, as well as considerable variability in how syndromic versus non-syndromic individuals were defined.²⁹ Furthermore, a landmark paper which defined CNV profiles of individuals with developmental delay found that individuals with neurodevelopmental phenotypes and craniofacial findings, including CL/P, harbored larger CNVs than individuals with autism spectrum disorder or epilepsy alone.¹⁷ Conversely, a research study employing early microarray technology identified sub-karyotypic deletions overlapping known cleft loci in individuals with SCL/P and NSCL/P at nearly equivalent frequencies and sizes.²⁶ These included a classic ∼2.7 megabase (Mb) deletion of 22q11.21 overlapping the DiGeorge syndrome (MIM: 188400) locus, deletions of IRF6 (MIM: 607199) ranging from 100 kilobases (kb) to 1 Mb in size (in individuals with Van der Woude syndrome [MIM: 119300] within their SCL/P cohort),²⁶ and large deletions in two individuals with NSCL/P (one with a 3.2 Mb deletion at chromosome 6q25.1q25.2 and one with a 2.2 Mb deletion at 10q26.11q26.13 overlapping FGFR2 [MIM: 176943]).²⁶ Other studies of individuals affected with clefting phenotypes have used microarrays to identify rare, likely etiologic deletions overlapping known Mendelian clefting loci,²⁵ as well as putative clefting regions,^{20,23,25,27,28} and have begun to explore the role of higher-frequency variants (known as copy-number polymorphisms, or CNPs) in clefting.^19,24 However, these studies were conducted on relatively small collections of individuals with CL/P, precluding the discovery of additional robust genetic associations. Here we present the largest study to date of CNVs in individuals with SCL/P and NSCL/P, allowing us to comprehensively assess the respective CNV profiles of individuals between and within these subgroups, compare these CNVs to known causes of Mendelian clefting, and also identify clefting loci within individuals with NSCL/P.

In this study, we generated high-resolution comparative genomic hybridization array data (aCGH) using a cohort of 869 individuals of Filipino ancestry (792 of whom had NSCL/P and 77 of whom had SCL/P). We then assessed the total number of CNVs, CNV load, and CNV burden between cleft type and within NSCL/P subgroups and observed no significant differences between any groups. In the second part of our study, we identified 28 individuals with CNVs overlapping known syndromic loci in addition to 160 clinically relevant clefting genes overlapped by at least one CNV in our cohort. Finally, using a proprietary gene discovery pipeline, we identified genes deleted at a rare frequency in individuals with clefts whose loss was enriched in our affected cohort compared to control subjects, leading to prioritization of three putative NSCL/P clefting loci: RIC1 homolog, RAB6A GEF complex partner (RIC1 [MIM: 610354]), Rho guanine nucleotide exchange factor 38 (ARHGEF38 [MIM: 619919]), and Cordon-bleu WH2 repeat protein like 1 (COBLL1 [MIM 610318]). F0 CRISPR-Cas9 deletion of these candidate genes in Xenopus laevis (African clawed frog) and Danio rerio (zebrafish) resulted in craniofacial dysmorphologies, including medial clefts in frogs that are analogous to human clefts.

Material and methods

Sample collection

Samples were gathered from individuals seen during surgical screening as part of Operation Smile medical missions in the Philippines (910)³⁰ or evaluated at the University of Iowa (308 individuals of European ancestry and 101 individuals of unknown ancestry) as part of contact during clinical care or epidemiologic surveys. All individuals were recruited following signed informed consent obtained in compliance with Institutional Review Board (IRB No. 199804081 [Philippines] and IRB No. 199804080 [Iowa]). After bioinformatic quality controls of the array-based comparative genomic hybridization (aCGH) data,²⁰ 1,108 probands were analyzed (1,025 NSCL/P; 83 SCL/P).

Array-based comparative genomic hybridization

Comparative genomic hybridization was performed as recommended by the manufacturers (Roche NimbleGen cgh_cnv_userguide_v7p0; Agilent G4410-90020v3_1_CGH_ULS_Protocol). Briefly, 1 μg (Agilent: 0.5 μg) of DNA from an individual with a cleft was labeled with Cy3-coupled nonamers and 1 μg (0.5 μg) of control DNA (from an unaffected male from the Philippines for male and female samples gathered during Operation Smile medical missions in the Philippines, and an unaffected male of European ancestry for samples gathered at the University of Iowa) was labeled with Cy5-coupled nonamers. Each labeled DNA sample was co-hybridized to a Roche NimbleGen (Human CGH 2.1M Whole-Genome Tiling v2.0D Array) or Agilent (SurePrint G3 Human CGH Microarray Kit 1 × 1M) human whole-genome tiling array, and the array was processed, scanned, and analyzed as previously described.^15,31

Copy-number-variant calling and quality control

BioDiscovery’s Nexus Copy Number FASST2 Segmentation Algorithm, a Hidden Markov Model (HMM)-based approach, was used to make initial copy number calls as previously described,²⁰ and quality control (QC) steps were performed as described below. Microarray data were QC’d based on several data metrics generated by NimbleGen’s DEVA, Agilent’s Feature Extraction and Nexus software.³¹ Arrays were retained if less than eight or seven metrics (for NimbleGen or Agilent, respectively) fell outside of two standard deviations of the mean.

For CNV call QC, duplicate arrays per individual and sex discordant arrays were removed from further analysis. X-shift values of females were adjusted to control for hybridization against a male control using the median X-shift across all females by array platform. Due to the increased fragmentation of calls made by Nexus software, all calls of the same type (gains versus losses) were merged using BEDTOOLS when breakpoints were within one base pair of each other. Nexus CNV calls were verified using DEVA’s segMNT algorithm (NimbleGen) or CytoGenomic’s ADM-2 algorithm (Agilent). Parameters were adjusted from default settings to more similarly call CNVs when compared to Nexus by setting the minimum segment difference to 0.3 and requiring a minimum of 3 probes per segment in DEVA. Calls between Nexus and either calling algorithm were compared using BEDTOOLS intersect, requiring 70% overlap in either direction to be retained for further analysis. To reduce false positive calls, we required a minimum number of probes encompassed by the CNV (NimbleGen: 10, Agilent: 4), a shift value of ≥0.42 or ≤−0.7 for gains or losses, respectively, and that less than 70% of the CNV was overlapped by a segmental duplication, centromere, telomere, or pseudoautosomal region. All calls occurring in ≥70% of individuals run on each platform or ≥70% of the population were also removed, as these were likely the result of array-specific or control-specific artifacts, respectively. Finally, any arrays with the total number of calls falling two standard deviations from the mean or containing less than 20 CNV calls were considered outliers and removed from further analysis. This resulted in 869 high-quality NimbleGen arrays and 239 high-quality Agilent arrays for analysis, of which all CNV calls were visually inspected (Figure 1).

Figure 1.

Bioinformatics copy-number-variant prioritization pipeline

Of the original 1,218 arrays, 1,102 passed quality controls and were used for downstream analyses. Copy-number variants (CNVs) which were overlapping an exon of a gene which passed minimum quality-control metrics (probe coverage and shift value) but occurring in less than 70% of the cohort or sharing less than 70% overlap with common CNVs or segmental duplications, were visually inspected. Genes that were recurrently deleted but at a frequency of less than 1% of the cohort with CL/P and at a higher frequency in individuals with clefts versus control subjects were prioritized for functional analysis.

Due to the fact that no large-scale studies of CNVs in control populations from the Philippines have been published, CNV calling was performed on 1,783 control samples obtained from females from the Philippines enrolled in the Cebu Longitudinal Health and Nutrition survey.^32,33 Samples were genotyped on Affymetrix 5.0 arrays, analyzed using PennCNV,³⁴ and compared to calls within the individuals with CL/P for functional validation prioritization. Copy-number variants (CNVs) from select trios (when available) were validated in Illumina Omni 5 Exome BeadChip microarrays with PennCNV v.1.0.5.³⁴ In brief, PennCNV uses Log R Ratio (LRR), a measure of signal intensity, and B Allelic Frequency (BAF), a measure of allelic ratio, to infer the presence of duplications and deletions in microarrays. The CNV calls generated by PennCNV were annotated with RefSeq³⁵ hg19 (GRCh37) gene coordinates using a custom script to confirm the presence of deletions of interest, and the presence of deletions in COBLL1 were confirmed by visual inspection of LRR and BAF plots.

Summary statistic generation

From our list of high-confidence CNV calls (see “copy-number-variant calling and quality control”), we employed the BEDTOOLS intersect function to calculate frequency of each CNV within the cohort by population requiring a 70% reciprocal overlap. Calls were separated into three lists for analysis: (1) all calls (Table S1), (2) semi-rare calls (occurring at a 1%–5% frequency), and (3) rare calls (occurring at a less than 1% frequency). Due to the fact that CNV number was largely influenced by hybridization platform and population, only the cohort for which we had sufficient power (samples hybridized using the NimbleGen platform) was used for the generation of the summary statistics. All statistical comparisons were conducted using VassarStats: http://vassarstats.net/.

Comparison with genomic disorder loci

All identified variants in samples gathered from individuals from the Philippines or at the University of Iowa with CL/P were compared to an in-house curated list of known genomic disorders (utilized by The University of Iowa’s clinical cytogenetics laboratory; Table S2) using the BEDTOOLS intersect function and requiring a minimum of 70% reciprocal overlap (see Tables S2 and S3 for a summary and complete list of CNVs identified in our affected cohort, respectively).

Comparison with known clefting loci

All identified variants in our cohort of individuals with CL/P were compared to two lists of genes which contribute to Mendelian CL/P disorders: a list of 358 genes which are putatively involved in CL/P formation³⁶ and 336 clinically relevant genes involved in Mendelian clefting³⁷ (see Table S4 for a list of Mendelian clefting genes overlapped by CNVs in our samples).

Rare variant analysis

High-confidence CNVs identified in samples obtained from individuals in the Philippines hybridized on the NimbleGen platform were filtered to identify calls overlapping coding regions of the genome. We compared the frequency of these calls within our cohort to a control dataset of individuals from the Philippines (see “copy-number-variant calling and quality control”), as well as an in-house curated list of CNVs from the Database of Genomic Variants²⁰ using the BEDTOOLS 70% intersect function and custom Python scripts. For our functional analysis, we focused on replicated deletions passing visual inspection which overlapped protein coding genes and occurred at a frequency of less than 1% but were deleted at a higher frequency in the individuals with NSCL/P from the Philippines versus control subjects. Three hundred twenty genes fit these criteria (Table S5) which were further prioritized for functional validation based on of several annotations, including haploinsufficiency score, presence of a deletion in our European cleft cohort, expression patterns in mouse (MGI: www.informatics.jax.org),³⁸ fish (ZFIN: zfin.org),³⁹ and frog (Xenbase: www.xenbase.org),⁴⁰ known human disease association (OMIM: omim.org), constraint (gnomAD: gnomad.broadinstitute.org),⁴¹ gene function (NCBI: www.ncbi.nlm.nih.gov), and presence of craniofacial anomalies in individuals harboring deletions of these genes (DECIPHER: https://decipher.sanger.ac.uk/).⁴²

Pathway analysis

Chromosomal position over-representative analysis was performed using GeneTrail 3.2⁴³ for genes deleted in one or more individuals with NSCL/P from the Philippines that were deleted at a higher frequency in affected individuals versus control subjects, yet at a frequency of less than 1% of the cohort overall (calls that occur at a less than 1% frequency are listed in Table S5), in addition to genes deleted in only one individual within the NSCL/P cohort from the Philippines yet at a higher frequency in affected individuals versus control subjects (singleton deletions fulfilling these criteria can be found in Table S6). Default values were used including Benjamini-Yekutieli to control for false discovery rate in multiple testing.⁴⁴ Pathway analysis was performed using Ingenuity Pathway Analysis (IPA) Core Analysis with default settings including consideration of direct and indirect relationships, experimentally observed and high (predicted) confidence levels, and all available species, as well as an “Enriched diseases and functions” output (Qiagen). Also included in the analysis pipeline was a Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis. Canonical pathways that achieved statistical significance after Benjamini-Hochberg correction were included in the results (Tables S7A–S7H).

Xenopus laevis

Staging

Xenopus laevis embryos were obtained using standard procedures⁴⁵ approved by the VCU Institutional Animal Care and Use Committee (IACUC protocol number AD20261) and University of Iowa IACUC (protocol number 2021664). Embryos were staged according to Nieuwkoop and Faber.⁴⁶ Stages are also reported as hours postfertilization at 23°C for better comparisons across vertebrates.

In situ hybridization

Since X. laevis is allotetraploid and thus has two homeologs of a large number of genes (denoted S and L), we designed our in situ probes to maximize the likelihood that both homeologs would be detected. To design in situ probes for arhgef38, cobll1, and ric1 that are most likely to detect both homeologs of each gene, we aligned the L and S sequences using MultAlin and identified the most similar 800 base pair (bp) region from highly conserved exons (exons 10–13 for arhgef38, exon 13 for cobll1, and exons 19–22 for ric1). The sequences were synthesized by IDT and inserted into the pUCIDT vector at the EcoRV site. Antisense and sense probes were synthesized using T3 and T7 polymerases, respectively. Probes were diluted to 1 μg/mL in hybridization buffer prior to use. Whole-mount in situ hybridization was performed as previously described.⁴⁷ A minimum of six embryos were used for each probe at each stage.

Morpholino knockdown

Splice-blocking antisense morpholinos (MOs) were designed and purchased from GeneTools (sequences in Table S8). A standard control MO was obtained from GeneTools. All MOs were labeled with fluorescein which allowed separation of un-injected individuals from injected, fluorescent animals by 24 h of development. Microinjections were performed using an Eppendorf Femtojet micro-injector and a Zeiss Discovery V8 stereoscope. Embryos were placed in a dish lined with nylon Spectra mesh (1,000 μm opening and 1,350 μm thickness) at the bottom to hold embryos in place and filled with 3% Ficoll 400 (Fisher, cat # BP52 5) dissolved in 0.1X MBS. MOs were diluted to 34 ng/μL and 17 ng/μL, and 1–2 nL were injected into each embryo (effective concentrations reported in Table S8). To assess whether the MOs resulted in changes in mRNA structure, PCR was performed with Apex Hotstart Taq master mix (Bioline, cat # 42-144) on a BioRad MJ Mini Personal Thermocycler. The PCR products were analyzed on a 2% agarose gel prepared with molecular grade agarose (Bioline, cat # BIO-41025) in TAE buffer.

CRISPR-Cas9 mutagenesis

sgRNA was designed using the ChopChop software.⁴⁸ A sequence was chosen that best targeted the desired gene with no off-targets and high efficiency (see sequences in Table S9). The sgRNAs were purchased from Synthego and diluted as recommended in low EDTA TE buffer. Then, 200 pg of sgRNA was incubated with 2 pg of Cas9 protein (PNA Bio Inc., cat # CP01) for 10 min and 1–2 μL was injected into each embryo. A negative control consisted of the same concentrations of Cas9 protein. Since there was no way to visually determine whether an embryo was mutant, all embryos injected were counted but only those with some phenotype were used to calculate percentages with craniofacial defects. The DNA was extracted from 10 randomly selected embryos with a craniofacial malformation using the HotShot protocol.⁴⁹ Each embryo was immersed in 40 μL of an alkaline lysis buffer (25 mM NaOH, 0.2 mM Na-EDTA) and heated for 40 min at 95°C. The solution was then cooled, and an equal volume of neutralization buffer (40 mM Tris-HCL) was added. One mL of this solution was used in a standard PCR reaction with Hotstart Taq master mix and primers that flanked the predicted mutation site (primer sequences can be found in Table S9). The product was then sent for purification and sequencing at Genewiz (Azenta Life Sciences) using the same forward primers.

Imaging facial features of X. laevis

At stage 42–43, tadpoles were anesthetized in 1% tricaine for 10 min and then fixed in 4% paraformaldehyde overnight at 4°C. A No. 15 scalpel (VWR, cat # 82029-856) and Dumont No. 5 forceps (Fisher, cat # NC9404145) were used to make two cuts to isolate the head: first at the posterior end of the gut and then second caudal to the gills. Isolated heads were mounted in small holes or depressions in either agarose or clay-lined dishes containing PBS with 0.1% Tween (PBT). The faces were imaged using a Discovery V8 stereoscope fitted with an Axiovision digital camera (Zeiss).

Alcian blue staining

At stage 45, tadpoles were anesthetized in 1% tricaine for 10 min and then fixed in Bouin’s solution overnight at room temperature. After washing out the fixative in 70% EtOH, tadpoles were soaked for 3–4 days in an Alcian blue solution (20% acetic acid, 80% EtOH, 0.1 mg/mL Alcian blue). Tadpoles were washed with acid alcohol (AA; 1% HCl, 70% EtOH) and rehydrated in PBT, and pigment was removed by soaking in 3% H₂O₂ for 45–60 min. Tadpoles were then washed in 1% KOH and mounted in 75% glycerol for imaging.

Danio rerio

D. rerio maintenance

Danio rerio embryos, larvae, and adults were reared as described previously⁵⁰ in the University of Iowa Zebrafish Facility, and approved by the University of Iowa IACUC (ACURF protocol number 1003051). Animals were staged by hours or days postfertilization at 28.5°C (hpf or dpf, respectively).⁵¹

CRISPR-Cas9 mutagenesis

Fertilized Danio rerio embryos were co-injected at the one- to two-cell stage with sgRNA (200–400 pg per embryo) and/or Cas9 protein (IDT) at 2 ng per embryo. The efficacy of each sgRNA within Danio rerio embryos was vetted by high-resolution melt analysis on eight individual injected embryos (sgRNA sequences are listed in Table S10).

Alcian blue staining

Danio rerio larvae were euthanized then fixed overnight in 4% paraformaldehyde (PFA). After a wash in phosphate-buffered solution (0.8% NaCl, 0.02% KCl, 0.02 M PO4 [pH 7]) with 0.25% Tween 20 (PBST), pigment was removed by soaking in a 3% H₂O₂ and 0.5% KOH medium for 20–30 min. Larvae were then washed in PBST and soaked overnight in an acid alcohol (AA) solution containing Alcian blue (0.37% HCl, 70% EtOH, and 0.1% Alcian blue). The larvae were then washed extensively in AA, rehydrated in PBST, and mounted in 4% methyl cellulose for imaging.

Results

Comparative analysis of CNV profiles between clefting subtypes

We analyzed aCGH data from 869 individuals from the Philippines, 792 with NSCL/P (206 CL, 531 CLP, 54 CPO, 1 unknown) and 77 with SCL/P (8 CL, 56 CLP, 9 CPO, 4 unknown), hybridized on NimbleGen 2.1M feature whole-genome tiling arrays that passed quality controls (see material and methods). Data analysis was performed using Nexus Copy Number (version 7.5; BioDiscovery), DEVA (version 1.2; Roche NimbleGen), BEDTOOLS, and several in-house custom Python scripts (see material and methods). We conducted a comparative analysis of individuals with NSCL/P and SCL/P considering total number of CNVs detected (Figures S1A–S1D), the amount of genomic content overlapped by CNV events (“CNV load;” Figures S1E–S1H), and the number of protein-coding genes overlapped by CNVs (“CNV burden;” Figures S1I–S1L). Overall, we observed no significant differences between groups (2-sided Mann-Whitney; Bonferroni corrected p value required for significance: <0.00185), even when stratifying by CNV type (loss or gain) or CNV frequency (all CNVs, Figure S1; or CNVs occurring at a 1%–5% frequency or <1% frequency, data not shown). We also assessed the 792 individuals from the Philippines with NSCL/P for differences between CNV number, load, and burden after stratifying by genotypic sex (500 male, 292 female; Figures S2A, S2E, and S2I, respectively), cleft type (206 CL, 531 CLP, 54 CPO; Figures S2B, S2F, and S2J, respectively), unilateral cleft sidedness (346 left, 170 right; Figures S2C, S2G, and S2K, respectively), and cleft laterality (206 bilateral, 516 unilateral; Figures S2D, S2H, and S2L, respectively), and observed no significant differences in CNV profiles between these NSCL/P subgroups (2-sided Mann-Whitney and Kruskal-Wallis; Bonferroni corrected p value required for significance: <0.00139). Finally, a qualitative investigation into the largest CNV occurring within each individual by cohort (total, NSCL/P, or SCL/P) showed a moderate increase in gains sized 300–400 kb within the SCL/P cohort, but otherwise no readily apparent differences in the largest gains and losses were observed (Figure S3).

Detection of CNVs overlapping known genomic disorders

To identify known pathogenic CNVs within our cohort of individuals with CL/P, we compared all CNVs passing quality controls to a list of loci previously implicated in genomic disorders (Table S2). Twenty-eight individuals (25 from the Philippines; 3 of European ancestry) were found to have CNVs which shared 70% reciprocal overlap with a known syndromic disorder locus (Tables 2, S2, and S3). The most common finding was an ∼80 kb duplication of the HOXD cluster at 2q31.1 in 11 individuals (4 SCLP; 4 NSCL; 3 NSCLP), followed by four ∼1.6 Mb type II deletions (1 SCPO, 3 NSCLP) and two ∼1.2 Mb type II duplications (2 NSCPO) of 22q11.2. Two individuals had CNVs overlapping the thrombocytopenia-absent radius (TAR) susceptibility region on 1q21.1 (one gain, one loss; both with NSCLP [MIM: 274000]), two had ∼1.3 Mb deletions of 3q29 (2 NSCLP), and two individuals had ∼540 kb deletions of 16p11.2 (1 NSCLP; 1 NSCPO). In addition, one individual each was found to have an ∼874 kb duplication of the Williams-Beuren syndrome locus on 7q11.23 (NSCLP [MIM: 194050]), an ∼744 kb duplication of 15q11.2 (NSCL), an ∼786 duplication of 16p13.11 (NSCLP), an ∼280 kb duplication of Xq28 (NSCL), and confirmation of trisomy 21 in an individual with Down syndrome (MIM: 190685) and SCLP. Finally, we note that two individuals diagnosed with Van der Woude syndrome (VWS) who had been previously reported as negative for sequence variants within IRF6 and GRHL3 ([MIM: 608317]; known Mendelian causes of VWS^52,53) were assessed for CNVs overlapping these genes or nearby non-coding regions, and no variants were detected.

Table 2.

Summary of copy-number variants overlapping genomic disorder loci

Genomic disorder locus	CNV type	Chr	Start coordinate (hg18)	Stop coordinate (hg18)	CNV size range (kb)	Number of individuals	Ancestral country of origin	Cleft type	Cleft lip laterality	Genotypic sex	Diagnosis (syndromic)
1q21.1 TAR susceptibility	del	1	144,112,609	144,525,602	413	1	P	NSCLP	B	F	N/A
1q21.1 TAR susceptibility	dup	1	144,112,609	144,494,342	382	1	E	NSCLP	Un	M	N/A
HOXD cluster	dup	2	176,665,222	176,750,540	73–85	11	P	NSCLP:3; NSCL:4; SCLP:4	L:2; R:3; B: 6	M:6; F:5	C&L:1; Ank:1; Un:2
3q29	del	3	197,513,580	198,827,680	1314	2	P	NSCLP	L:1; R:1	M:1; F:1	N/A
WBS distal	dup	7	74,907,157	75,754,545	847	1	P	NSCLP	L	F	N/A
15q11.2 BP1-2	dup	15	20,428,073	21,172,461	744	1	P	NSCL	L	F	N/A
16p13.11	dup	16	15,413,831	16,199,769	786	1	P	NSCLP	L	M	N/A
16 p11.2 (593kb)	del	16	29,557,497	30,107,356	540–549	2	P:1; E:1	NSCLP:1; NSCPO:1	L	M	N/A
Trisomy 21	dup	21	26,384,126	46,944,323	20,560	1	P	SCLP	L	M	T21
DGS/VCFS Type II	dup	22	17,271,718	18,691,318	1,226–1,290	2	P:1; E:1	NSCPO	N/A	M	N/A
DGS/VCFS Type II	del	22	17,402,877	18,691,904	1,289–1,650	4	P	NSCLP:3; SCPO:1	L:2; B:1	M:2; F:2	PR
Xq28	dup	X	153,260,249	153,540,650	280	1	P	NSCL	L	F	N/A

Twenty-eight individuals were found to have copy-number variants overlapping known genomic disorder loci (see Table S2 for a complete list of loci and Table S3 for the coordinates of the call identified in each proband). hg18, human genome build coordinates GRCh18; kb, kilobase; TAR, thrombocytopenia absent radius; WBS, Williams-Buren syndrome; BP1-2, breakpoints 1 and 2; DGS/VCFS, DiGeorge/velocardiofacial syndrome; del, deletion; dup, duplication; chr, chromosome; P, Philippines; E, Europe; NSCLP, non-syndromic cleft lip and cleft palate; NSCL, non-syndromic cleft lip; SCLP, syndromic cleft lip and cleft palate; NSCPO, non-syndromic cleft palate only; SCPO, syndromic cleft palate only; B, bilateral; Un, unknown; L, left; R, right; N/A, not applicable; F, female; M, male; C&L, cleft and limb defects; Ank, ankyloblepharon; T21, Trisomy 21; PR, Pierre Robin.

Detection of CNVs overlapping known clefting loci

In order to determine whether any of the CNVs passing our quality control filters overlapped with genes associated with Mendelian clefting disorders, we compared any genes within the deleted or duplicated interval with a list of candidate genes from 2017,³⁶ as well as a list of clinically relevant genes from 2020,³⁷ associated with clefting phenotypes. This yielded a list of 123 candidate genes and 160 clinically relevant genes implicated in clefting that were overlapped by at least one CNV in our cohort (56 of which overlapped between both lists; Table S4). Further restriction of this list by deletions occurring in ≤1% of our cohort and with a population frequency cutoff of 0.1% in our Filipino and DGV curated controls (see material and methods) resulted in a list of 51 genes (16 candidate: ANK1 [MIM: 612641], AUTS2 [MIM: 607270], COMT [MIM: 116790], CRLF1 [MIM: 604237], GMPPB [MIM: 615320], LMNA [MIM: 150330], OTX2 [MIM: 600037], RAI1 [MIM: 607642], RBM8A [MIM: 605313], SCLT1 [MIM; 611399], SMOC1 [MIM; 608488], SPEG [MIM: 615950], TAC3 [MIM: 162330], TACR3 [MIM: 162332], TBX4 [MIM: 601719], and TPM2 [MIM: 190990]; 24 clinically relevant: ARID3B [MIM: 612457], BMPR1A [MIM: 601299], CHD7 [MIM: 608892], DHCR7 [MIM: 602858], FZD2 [MIM: 600667], GDF11 [MIM: 603936], GREM1 [MIM: 603054], IQGAP2 [MIM: 605401], ISM1 [MIM: 615793], KDM6A [MIM: 300128], KMT2A [MIM: 159555], MMP3 [MIM: 185250], MSX1 [MIM: 142983], NBAS [MIM: 608025], NECTIN1 [MIM: 600644], NEDD4L [MIM: 606384], PORCN [MIM: 300651], RYR1 [MIM: 180901], SCAMP1 [MIM: 606911], SIX1 [MIM: 601205], SNAP29 [MIM: 604202], STK11 [MIM: 602216], TBX1 [MIM: 602054], and UFD1 [MIM: 601754]; and 11 appearing on both lists: ASXL1 [MIM: 612990], B3GLCT [MIM: 610308], BCOR [MIM: 300485], DYNC2H1 [MIM: 603297], FGFR2 [MIM: 176943], HDAC8 [MIM: 300269], IFT140 [MIM: 614620], MSX2 [MIM: 123101], SKI [MIM: 164780], WDR11 [MIM: 606417], and YAP1 [MIM: 606608]). The majority of the aforementioned genes were overlapped by one or two deletions in the cohort (43/51; 84%), and TACR3 (candidate gene) was the gene most frequently overlapped by rare losses, with eight individuals with clefts harboring a deletion. TBX1 (clinically relevant) was deleted in five individuals, and COMT (candidate), NEDD4L (clinically relevant), SNAP29 (clinically relevant), STK11 (clinically relevant), and UFD1 (clinically relevant) were each overlapped by four rare deletions. Of note, COMT, SNAP29, UFD1L, and TBX1 were overlapped by the same deletion in four or five different individuals, respectively, within the 22q11.2 locus, and likely indicate the presence of DiGeorge syndrome in these individuals.

Pathway analysis

Over-representation analysis using chromosomal position was performed for genes deleted at a rare frequency within the NSCL/P cohort from the Philippines (arbitrarily defined as deleted in less than 1% of the affected cohort) that were also deleted at a higher frequency in individuals with NSCL/P versus control subjects. The most statistically significant results for genes deleted in more than one individual with NSCL/P were chromosomal loci at which known genomic disorders have been identified including 3q29, 4p16.1, and 22q11.21 (Table S7A). These disorders all contain CL/P within their associated spectrum of disease.^{54,55,56,57,58} This analysis was repeated for genes deleted in only one individual in our NSCL/P cohort from the Philippines, and several emerging and potential disease loci were identified. These include 7q35, 11q22.1q11.2, 14q32.32q32.33, and 19p13.3 (Table S7B). In addition, we observed that these singleton deletions overlapped well-known genomic disorder loci, including 1q21.1 and 16p11.2.

To further explore the genes which were recurrently deleted or only deleted in one individual in our NSCL/P cohort from the Philippines, pathway analysis was performed using the QIAGEN Ingenuity Pathway Analysis (IPA) platform (QIAGEN IPA: https://digitalinsights.qiagen.com/IPA).⁵⁹ Although none of the pathways identified for recurrently deleted genes reached significance, singleton deleted genes showed statistically significant enrichment for 15 canonical pathways including inhibition of matrix metalloproteases, synaptogenesis signaling pathway, role of osteoblasts, osteoclasts, and chondrocytes in rheumatoid arthritis, and axonal guidance signaling (Table S7C). Next, we preselected genes based on DECIPHER (DECIPHER: https://decipher.sanger.ac.uk/) haploinsufficiency scores of <10 for our analysis pipeline, since scores in this range denote likely haploinsufficiency loci (genes for which loss of one copy is deleterious and likely to cause a disease phenotype). For singleton deleted genes with haploinsufficiency scores <10, numerous other pathways emerged, including epithelial adherens junction signaling, RANK signaling in osteoclasts, regulation of the epithelial-mesenchymal transition pathway, gap junction signaling, apoptosis signaling, WNT/beta-catenin signaling, and integrin signaling, among others (Table S7D). Additionally, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis of these genes revealed 18 enriched cellular pathways including adherens junction, apoptosis, axon guidance, and focal adhesion (Table S7E).^60,61,62

In order to further mine our dataset, we used IPA/KEGG to look for canonical pathways, KEGG, and functions pathways using a combined list of all genes deleted in one or more individuals at a less than 1% cohort frequency with HI scores of 10 or less. We chose to focus on this list given the strong haploinsufficiency scores which suggest that they would be the most likely genes to show disease phenotypes upon monoallelic loss. Among the statistically significant categories known to be important for cranial neural crest cell (CNCC) function were the function annotations “migration of cells” (adjusted p = 6.92 × 10⁻⁴) which includes 27 genes, “formation of cellular protrusions” (adjusted p = 3.86 × 10⁻⁵) which includes 19 genes, and “organization of cytoskeleton” (adjusted p = 1.09 × 10⁻³) which includes 20 genes (Tables S7F–S7H). These analyses also led to several additional enriched pathways, including ephrin receptor signaling.

Identification of putative clefting loci

Due to our previous identification of a clefting locus, Isthmin 1 (ISM1 [MIM: 615793]), in a cohort of 140 individuals from the Philippines with NSCL/P,²⁰ we employed a similar filtering strategy to identify putative clefting candidates within the 792 probands from the Philippines with NSCL/P. Since our strategy relies upon population genetics data in which high effect size variants are generally rare in frequency in the population (due to selection against such variants), we thus focused on protein coding genes that were overlapped by a deletion (which are generally more deleterious than gains) at a rare frequency in individuals with NSCL/P. We stringently defined this threshold as deletions occurring at a less than 1% frequency yet occurring in two or more affected individuals that were also found at a higher frequency in individuals with NSCL/P than control subjects. These filtering criteria resulted in 320 genes for further prioritization (a list of these genes can be found in Table S5). We then assessed each call for replication within the cohort of European ancestry with NSCL/P (229) and used haploinsufficiency score (<10 predicted as likely haploinsufficient in DECIPHER; DECIPHER: https://decipher.sanger.ac.uk/); gnomAD constraint score (gnomAD: https://gnomad.broadinstitute.org); overlap by CNVs in DECIPHER within individuals with craniofacial phenotypes⁴²; craniofacial expression pattern (if known) in Danio rerio (ZFIN: https://zfin.org), Xenopus (Xenbase: www.xenbase.org), and mouse (MGI: www.informatics.jax.org/); ortholog presence in Danio rerio and Xenopus laevis; known biological function (National Center for Biotechnology Information [NCBI]; NCBI: https://www.ncbi.nlm.nih.gov/gene); and previous implication in clefting^36,37 to select the most promising Mendelian clefting candidate genes (all annotations can be found in Table S5). Deletions overlapping a maximum of three genes were preferentially selected for ease of functional interpretation. Given the available data for each gene and their fulfillment of the above criteria, we identified seven genes that had a strong likelihood of being associated with craniofacial development (Table 1): R. guanine nucleotide exchange factor 38 (ARHGEF38); Cordon-bleu WH2 repeat protein-like 1 (COBLL1); Exocyst complex component 4 (EXOC4 [MIM: 608185]); Lipoprotein lipase (LPL [MIM: 246650]); Plakophilin 2 (PKP2 [MIM: 602861]); RAB6A GEF complex (RIC1); and von Willebrand factor D and EGF domains (VWDE) (Log2 plots of these deletions generated by Nexus software for EXOC4, LPL, PKP2, and VWDE can be found in Figure S4; for ARHGEF38, COBLL1 and RIC1, see Figure 2).

Table 1.

Top seven candidate genes with strong likelihood of involvement in craniofacial development

Gene	Number probands with gene deleted	Gene function
Rho guanine nucleotide exchange factor 38 (ARHGEF38)	4	unknown; Rho signaling
Cordon-bleu WH2 repeat protein-like 1 (COBLL1)	3	promotes actin filament formation and dendritic branching via WH2 domain
Exocyst complex component 4 (EXOC4)	3	part of the exocyst complex which targets exocytic vesicles for docking on the plasma membrane; essential for epithelial polarity and interacts with actin cytoskeleton
Lipoprotein lipase (LPL)	4	triglyceride hydrolase and ligand factor for receptor-mediated lipoprotein uptake
Plakophilin 2 (PKP2)	4	Armadillo repeats allow localization to cell desmosomes and nuclei, linking cadherins to intermediate filaments in the cytoskeleton; may regulate β-catenin
RAB6A GEF complex partner 1 (RIC1)	3	necessary for nucleotide exchange on Rab6A; Rab6A functions in the exocytic pathway and interacts with ARHGEF10
von Willebrand factor D and EGF domains (VWDE)	7	enables cell adhesion in the blood stream

Top candidate genes deleted in our cohort are indicated in the first column, with the number of probands carrying deletions of each gene indicated in the second column. The third column describes any known functions of each gene.

Figure 2.

Log2 plots of top candidate clefting genes

Log2 plots of (A) ARHGEF38, (B) COBLL1, and (C) RIC1 were generated using BioDiscovery’s Nexus 7 software and the UCSC genome browser. NSCLP, non-syndromic cleft lip and palate; NSCL, non-syndromic cleft lip only; NSC, non-syndromic cleft; SCLP, syndromic cleft lip and palate.

For the individuals with CL/P harboring deletions overlapping these candidate Mendelian NSCL/P genes, we assessed whether the CNV event was de novo or inherited by hybridizing parent samples on Agilent 1M feature whole-genome tiling arrays or Illumina HumanOmni5Exome SNP arrays, when available. Of note, due to the low quality of some of the parental samples, parentage could not be confirmed in all individuals. In one tested trio per gene, we detected a de novo deletion overlapping EXOC4, LPL, and RIC1, whereas a second deletion of RIC1 was inherited from an affected mother, and one deletion overlapping COBLL1 was inherited from an unaffected father. Although not selected for functional analysis, we also identified three heterozygous, putative de novo deletions overlapping HNRNPL (MIM: 603083), EBI3 (MIM: 605816), and CHP2, heterozygous deletions of FAM149A, PCDH9 (MIM: 603581), and INPP5F (MIM: 609389) (each of which were inherited from an affected mother), and two putative de novo, heterozygous 3q29 deletions (MIM: 609425).

Functional validation of putative clefting genes

Rho and Rab GTPase signaling components are known to play a role in a variety of processes associated with cellular dynamics including the formation of adhesion junctions, actin organization, cell division, cell migration, and membrane trafficking),^{63,64,65,66,67} all of which are linked to craniofacial development. Additionally, genes involved in both Rho and Rab signaling are associated with craniofacial anomalies.^{68,69,70,71,72,73} Thus, we chose the three genes in our list of top candidates that play roles in Rho/Rab GTPase signaling for functional validation in Xenopus laevis (X. laevis): ARHGEF38, COBLL1, and RIC1. In situ hybridization using probes designed to detect both homeologs of each gene (see material and methods) revealed expression of all three genes in the head region of X. laevis embryos during early events in craniofacial development such as CNCC migration and branchial arch specification (Figure S5). At stage 25, arhgef38 is expressed ubiquitously in the head, including the presumptive branchial arches, with concentrated expression in the brain and spinal cord as well as the cement gland. Stages 27 and 29 show similar expression patterns to stage 25, although strong expression in the cement gland is no longer seen at stage 27 and is instead more pronounced in the otic vesicle and developing eye at stage 29. At stage 25, cobll1 is strongly expressed in the otic vesicle, hatching gland, and cement gland as well as scattered epidermal cells. At stages 27 and 29, cobll1 is ubiquitously expressed throughout the embryo including the presumptive branchial arches, with the strongest expression concentrated in the otic vesicle, cement gland, brain, and spinal cord. At stage 29, expression is further pronounced in the region of (and anterior to) the first branchial arch that gives rise to the midface. Finally, at stage 25, ric1 shows very low levels of expression, while at stage 27 its expression is observed throughout the head, including the presumptive branchial arches and eye. Stage 29 expression is similar to that of stage 27, with refined expression in the posterior-most branchial arches, otic vesicle, brain, and eye. In summary, all three genes show expression in the region of the branchial arches (the equivalent of the mammalian pharyngeal arches) which give rise to many of the structures forming the jaws and palate.

To determine whether cobll1, ric1, and arhgef38 are required for craniofacial development, we used both antisense oligos (morpholinos; MOs) and CRISPR-Cas9 to target individual homeologs of each gene. MOs or short guide RNAs (sgRNAs)/Cas9 targeting the S or L homeolog of each gene (Tables S8 and S9) were injected into the 1-cell stage of X. laevis embryos and their effectiveness was determined. To confirm that the MOs caused splicing defects, we used RT-PCR and primers flanking the exon targeted for deletion. Results indicated that indeed alternative gene products were produced in morphant tadpoles with craniofacial abnormalities (Figures S6–S8). The size of the alternative gene products was consistent with the predicted exon deletions. The effectiveness of CRISPR techniques were assessed by sequencing randomly selected mutants (with a craniofacial malformation) using primers that flanked the sgRNA target. Results indicated that all 10 CRISPR mutants for each gene tested had alternative sequences near the sgRNA target sites as predicted (Figures S6–S8C and S8D).

We next assessed the shape of the mouth for evidence of dysmorphologies by imaging and observing the faces of morphants and mutants at stages 42–43 (80–87 hours postfertilization; hpf), by which time the tissues that form the roof of the oral cavity have migrated to the region and have begun to expand and differentiate. A small percentage (ranging from 2.6% to 8.2%) of the control tadpoles injected with either control MOs or sgRNAs/Cas9 were smaller but despite this did not display obvious changes in craniofacial morphology including the shape of the mouth (Figure 3).

Figure 3.

Knockdown of Ric1, Cobll1, and Arhgef38 cause craniofacial malformations in Xenopus laevis

(A) Schematic showing injection of reagents at the one-cell stage followed by imaging at stage 42–43 (80–87 hpf). Xenopus illustrations © Natalya Zahn (2022).⁷⁴

(B and C) Frontal view of the face of representative tadpoles injected with control morpholinos (MOs) or Cas9.

(D) Stacked bar graphs showing that 100% of the tadpoles were normal with respect to their craniofacial morphology.

(E–H) Frontal views of the faces of representative tadpoles injected with splice-blocking MOs or short guide RNA (sgRNA)/Cas9 targeting ric1.L and ric1.S, respectively (three biological replicates for each).

(I) Stacked bar graphs showing the percentage of the tadpoles that had normal faces, had craniofacial malformations, or triangular mouths that appeared cleft-like.

(J–M) Frontal views of the faces of representative tadpoles injected with MOs or sgRNA/Cas9 targeting cobll1.L and cobll1.S (two biological replicates for each).

(N) Stacked bar graphs showing the percentage of the tadpoles that had normal faces, had craniofacial malformations, or triangular mouths that appeared cleft-like.

(O–R) Frontal views of the faces of representative tadpoles injected with MOs or sgRNA/Cas9 targeting arhgef38.L and arhgef38.S (two biological replicates for each).

(S) Stacked bar graphs showing the percentage of the tadpoles that had normal faces, had craniofacial malformations, or triangular mouths that appeared cleft-like. The tadpole mouth opening is outlined in pink dots. Numbers of tadpoles are reported in the bottom right corner. CMO, control morpholino; CR, CRISPR-Cas9; st, stage; cg, cement gland; L, Xenopus laevis L homeolog; S, Xenopus laevis S homeolog.

The ric1.L and ric1.S morphants (86.5% and 69.5%, respectively) and mutants (54.4% and 63.3%, respectively) had craniofacial defects which included narrower faces and eyes that were closer set, so much so that sometimes they appeared fused (Figures 3E–3I). Of the craniofacial malformations observed in these tadpoles, a portion also had triangular-shaped mouths that appeared cleft-like, indicative of primary palate malformation (ric1.L MO = 23.1%, ric1.S MO = 20.9%, ric1.L CR = 21.4%, ric1.S CR = 16.7%, Figures 3E–3I). In addition, for many of the ric1 morphants, the buccopharyngeal membrane (a layer of cells that covers the mouth opening) failed to break down (Figure S8H).

The cobll1.L and cobll1.S morphants (86.0% and 92.0%, respectively) and mutants (61.5% and 54.8%, respectively) also had craniofacial defects which included narrower faces and smaller eyes (Figures 3J–3N). Of the craniofacial malformations observed in these tadpoles, a portion had cleft-like triangular-like shaped mouths (cobll1.L MO = 27.2%, cobll1.S MO = 45.1%, cobll1.L CR = 5.5%, cobll1.S CR = 3.5%, Figures 3J–3N). Notable was the lower penetrance of mouth shape defects in the cobll1 mutants compared with their morphant counterparts. Such lower penetrance has been observed in mosaic F0 X. laevis mutants when a large field of mutant cells is required to observe a phenotype.⁷⁵ Further mechanistic studies would be necessary to uncover the reason for lower penetrance of the craniofacial phenotype in cobll1 mutants.

The arhgef38.L and arhgef38.S morphants (82.24% and 82.57%, respectively) and mutants (56.36% and 59.05%, respectively) had craniofacial defects which again included narrower faces and smaller eyes (Figures 3O–3S). Of the craniofacial malformations observed in these tadpoles, a subset also had cleft-like triangular shaped mouths (arhgef38.L MO = 13.08%, arhgef38.S MO = 14.68%, arhgef38.L CR = 24.55%, arhgef38.S CR = 25.71%, Figures 3O–3S). It is important to note that in these knockdown experiments, the range of defects was similar across both homeologs of each gene suggesting that they did not have distinctive roles in the embryo, and that both homeologs contributed to craniofacial development (Figures S6–S8). Further, we noted striking similarities as well as a similar spectrum of craniofacial morphology between morphants and mutants, suggesting that the malformations were not likely caused by off-target effects (Figures S6–S8). Intriguingly, this spectrum of moderate to severe craniofacial abnormalities recapitulates the phenotypes observed in our previous modeling of a clefting gene, ism1.²⁰

Since the shape and structure of the face is determined in part by the cranial cartilages, these structures were examined in arhgef38, ric1, and cobll1 morphants. Collagen labeling revealed a reduction in cartilages of the face and this reduction was more profound in tadpoles with severe craniofacial defects (Figure S9). In particular, we observed defects in the ceratobranchial, ceratohyal, Meckel’s, and trabecular cartilages in each of our morphant groups, in addition to other less common cartilage defects. Importantly, a reduction in these elements in the morphants is consistent with the expression of arhgef38, ric1, and cobll1 in the branchial arches, precursors of the cranial skeleton.

As an additional test of whether the three candidate genes contribute to craniofacial development, we examined phenotypes in Danio rerio (D. rerio) larvae injected with sgRNAs targeting each gene (along with Cas9) at the single-cell stage (Table S10). Since cobll1 has two paralogs (cobll1a, cobll1b) in D. rerio, we chose cobll1b for functional analysis given its reported expression in head regions during development,⁷⁶ in addition to its higher conservation to human COBLL1. We confirmed the efficacy of all sgRNAs using high-resolution melt analysis on eight individually injected embryos (see material and methods). The resulting embryos are expected to be genetically mosaic with a subset of cells exhibiting bi-allelic mutation of the targeted gene.⁷⁷ We examined the sgRNA plus Cas9 protein injected (mutant) embryos at 48 h postfertilization (hpf) for gross morphological defects of the head, including edema, ectopic blood in the fourth ventricle, and abnormal head shape (Figure S10A). More than 10% of mutant embryos had gross head deformities at 48 hpf. Surviving larvae were fixed at 4 days postfertilization (dpf) and processed to reveal the cartilage, and then examined for phenotypic abnormalities (Figure S10A–S10F). At 4 dpf, larvae sorted earlier based on abnormal head shape revealed the characteristic phenotypes described in Figure S10A (arhgef38, disorganized arches, collapsed ceratohyal; cobll1b, small bent Meckel’s and ceratohyal arches). F0 larvae injected with sgRNAs targeting ric1 recapitulated the hypoplastic Meckel’s cartilage and abnormal ceratohyal cartilages phenotypes reported in ric1 mutants,⁷³ and larvae injected with sgRNAs targeting radil1 (which was deleted in only one individual in our cohort and used as a positive control) demonstrated the expected head deformities, edema, and cartilage defects.⁷⁸ Collectively, our knockdown studies in X. laevis and D. rerio support ric1, cobll1, and arhgef38 as having important roles in midface development and being required for proper craniofacial development.

Discussion

Copy-number variants (CNVs) have been shown to contribute to orofacial clefting,^{19,20,21,23,24,25,26,27} and prior studies have largely focused on either the detection of CNVs overlapping known Mendelian clefting loci, the utilization of segregation in large pedigrees, or common CNVs that contribute to clefting phenotypes. To our knowledge, no comparative analyses of CNV profiles between individuals with varying cleft subtypes have been performed to date, and a limited number of studies have pursued functional validations of CNVs in individuals with CL/P for confirmation of a gene’s role in clefting. Our study furthers the investigation of the contribution of CNVs to CL/P by assessing CNVs occurring in individuals with NSCL/P versus SCL/P, as well as within NSCL/P subgroups. In addition, we report CNVs within these cohorts which overlap with known genomic disorder or Mendelian clefting loci. Finally, using an in-house analysis strategy to increase the likelihood of identifying clefting driver genes, we report the identification and validation in two vertebrate model organisms of NSCL/P candidates, COBLL1, RIC1, and ARHGEF38, which are overlapped by rare, recurrent deletions within our cohort of affected individuals.

By utilizing whole-genome tiling arrays and employing a series of stringent filters to identify high-confidence CNV calls in a cohort of 869 individuals with clefts from the Philippines (792 NSCL/P, 77 SCL/P), we assessed the total number of genomic CNVs, CNV load, and CNV burden within individuals with NSCL/P and SCL/P. No significant differences were observed between the two cohorts, suggesting that the overall number of CNVs and CNV content may not vary by cleft type. However, it is important to note that the inherent heterogeneity within the SCL/P cohort may be a confounding factor in drawing any strong conclusions about similarities or differences in global CNV profiles between individuals with NSCL/P versus those with SCL/P. In accordance with previously published work,^79,80,81,82 we observed that a higher proportion of the detected CNVs regardless of cleft type were gains rather than losses. This can likely be attributed to the fact that gains are generally considered to be better tolerated in the population than deletions.⁸³ We also note that the largest CNVs occurring within individuals are more often gains, and that the largest deletions occurring within individuals with CL/P (regardless of cleft type) are more frequently <500 kb (Figure S3). Of note, we observed a slight, qualitative increase in gains of 2–3 Mb in individuals with NSCL/P (18%) versus individuals with SCL/P (14%), and an elevated percent of individuals with SCL/P (22%) harboring gains 300–400 kb compared to those with NSCL/P (10.9%). Additional investigations which consider the genetic content of these regions and frequency of each CNV event are needed to determine whether there is clinical relevance to these findings.

Individuals with NSCL/P may be subcategorized by cleft type (cleft lip only [CL], cleft palate only [CPO], and cleft lip and palate [CLP]; listed from least to most prevalent), sex (occurring twice as frequently in males as females), cleft laterality (unilateral clefts versus the less common bilateral clefting), and cleft sidedness (with left-sided clefts occurring at a higher frequency than right-sided clefts).⁹ Recently, genetic modifiers significantly associated with these subgroups have been discovered and suggest a genetic basis for this phenotypic heterogeneity.^8,84 Stratification of our NSCL/P cohort by these subgroups demonstrated no difference in number of CNVs, CNV load, or CNV burden, suggesting that there is no direct correlation between an individual’s genomic CNV profile and clefting epidemiology.

The assessment of regions previously implicated in genomic disorders resulted in the identification of known pathogenic CNVs in 28 of the individuals with CL/P in our study (Tables 2, S2, and S3). Six individuals had alterations of the 22q11.2 locus which has been associated with 22q11.2 deletion syndrome (also known as velocardiofacial or DiGeorge syndrome) and 22q11.2 duplication syndrome (MIM: 608363).⁵⁴ Recurrent CNV gains and losses of this region are mediated by flanking low-copy repeats (LCRs) resulting in either a common 2.54 Mb CNV or smaller atypical or nested copy-number changes. All four individuals from our cohort carrying deletions of this region had smaller ∼1.3 to ∼1.6 Mb deletions, and the two duplications were ∼1.2 and ∼1.3 Mb in size. The reported phenotypic features of individuals with these syndromes most commonly include congenital heart defects, palatal anomalies (including CP), immune deficiency, hearing loss, characteristic facial features, and learning difficulties.⁵⁴ The majority of our individuals with CN alterations of this region (5/6) reportedly had NSCPO, three of whom also had CL (two unilateral; one bilateral). However, it is important to note that the phenotyping of the individuals from the Philippines was time limited, and investigators were unable to assess cardiac or other internal abnormalities. In addition, the parents of these individuals were not evaluated, so it is possible that features consistent with 22q11.2 deletion syndrome were missed. These data could also suggest that non-syndromic clefting may also be associated with smaller CNVs of this region or that individuals who present with clefting as the sole observable phenotype may, in fact, have later-onset syndromic findings, or subtle features requiring additional clinical testing (such as echocardiogram or behavioral assessment) for detection.

Two individuals each had ∼1.3 Mb deletions of 3q29, ∼545 kb deletions of 16p11.2 (MIM: 611913) and alterations of the 1q21.1 thrombocytopenia-absent radius (TAR) susceptibility locus (one individual with an ∼413 kb deletion and one individual with an ∼382 kb duplication). Recurrent, ∼1.6 Mb deletions of 3q29 are associated with neurodevelopmental findings including intellectual disability, autism spectrum disorder (ASD), and attention-deficit/hyperactivity disorder, as well as failure to thrive, patent ductus arteriosus, gastrointestinal concerns, and others.⁸⁵ Rarely, CL/P^55,56 or submucous CPO⁸⁶ have been reported in individuals with this syndrome. Notably, both of our individuals with 3q29 deletions had non-syndromic, unilateral clefts, suggesting that smaller deletions of this region may be a rare cause of NSCLP, or that additional syndromic features of individuals with deletions within the 3q29 microdeletion syndrome region may be subtle and easily overlooked during routine clinical phenotyping.

The recurrent ∼593 kb microdeletions overlapping 16p11.2 are associated with variable delayed language development, learning difficulties, and/or ASD.⁸⁷ Although CL/P is not considered part of the primary clinical phenotype, CPO has been reported in individuals with microdeletions or microduplications of 16p11.2, and CLP has also been observed in individuals with recurrent 16p11.2 microdeletions.⁸⁸ Microdeletions of this locus were observed in our cohort in one individual with NSCPO and one with NSCLP. Due to the variable penetrance of this microdeletion syndrome⁸⁷ as well as the rarity of CL/P reported in affected individuals, it is unclear whether this microdeletion is contributing to the observed clefting phenotypes or if they are of a separate etiology.

Finally, ∼200 kb recurrent deletions of the TAR locus at 1q21.1 have been reported in individuals with bilateral absent radii and thrombocytopenia in addition to other variable clinical features.⁸⁹ Proximal microdeletions and microduplications, as well as larger microdeletions involving this locus (including a 1.7 Mb interstitial deletion), have been reported in individuals with SCLP,^90,91,92 but to our knowledge NSCLP has not been associated with a microduplication or microdeletion of this region. Of note, we also identified several singleton gains and losses overlapping known genomic disorder loci, as well as a 14.4 Mb gain of 13q25 in one individual diagnosed with NSCLP (Tables S2 and S3).

Intriguingly, 11 individuals (4 SCLP with bilateral clefts; 4 NSCL; 3 NSCLP) had ∼80 kb duplications of the HOXD cluster. Although variants within this region have been previously implicated in limb defects, to our knowledge one person has been reported with a cleft, resulting from a translocation disrupting this cluster.⁹³ Although additional work within a larger, high-powered study is needed to better understand the clinical relevance of CNVs within this region, connections between Hox loci and craniofacial development have been well documented during mouse embryogenesis. For example, in both the hindbrain and in CNCCs, Hox genes are expressed in a nested fashion to combinatorially specify regional properties of the head,^94,95,96 and perturbation studies have shown that Hox genes are required for neural crest cell specification, migration, and differentiation.^95,97,98 However, the first branchial arch (which patterns the maxilla) appears to be devoid of Hox expression,⁹⁹ complicating interpretation of the results. One potential explanation is that gains of HOXD result in ectopic expression of the genes within the first branchial arch, altering their developmental trajectory. More work is clearly needed to mechanistically connect HOXD cluster duplications to orofacial clefting.

We next conducted an analysis of CNVs in our cohort overlapping putative or known Mendelian clefting loci and identified deletions overlapping 227 clefting genes (123 candidate³⁶ and 160 clinically relevant,³⁷ with 56 overlapping between both lists; Table S4). After restricting by cohort and population frequency, we observed TACR3 was overlapped with the highest number of deletions (eight individuals), followed by TBX1 (five individuals), and COMT, NEDD4L, SNAP29, STK11, and UFD1 (four individuals each). TACR3 has been associated with autosomal-recessive congenital hypogonadotrophic hypogonadism with or without anosmia. To our knowledge, deletions of TACR3 have not been observed in individuals with clefts outside of the present study. However, individuals with variants in other genes associated with hypogonadotrophic hypogonadism (such as FGFR1 and CHD7) have been reported to have CL/P,^{100,101,102,103} suggesting that alterations of genes involved in this disease pathway may also contribute to clefting phenotypes. TBX1 falls within the 22q11.2 microdeletion syndrome locus.⁵⁴ Murine studies assessing the impact of Tbx1 deletion in the palate demonstrated that knockout mice have a small mandible and tongue compared to wild-type controls, and that this loss resulted in the dysregulation of several genes previously implicated in cleft palate in humans, including MYH3 and NEB.¹⁰⁴ Additional work has further established TBX1 as part of the gene regulatory network required for palatal formation,^105,106,107 strongly supporting a role for dysregulation of this gene in cleft formation.

Of the genes that were deleted in four individuals in our cohort (COMT, NEDD4L, SNAP29, STK11, and UFD1), four of them (NEDD4L, SNAP29, STK11, and UFD1) have been observed in clinical syndromes. Variants in NEDD4L are associated with autosomal-dominant periventricular nodular heterotopia syndrome (MIM: 617201), which includes cleft palate, syndactyly, and neurodevelopmental delay.^108,109 Variants in SNAP29 cause autosomal-recessive cerebral dysgenesis, neuropathy, ichthyosis, and keratoderma (CEDNIK [MIM: 609528]) syndrome,^110,111 have been more recently reported in association with Pelizaeus-Merzbacher-like disorder,¹¹⁰ and result in variable yet complex phenotypic features (including CL/P) when observed in combination with 22q11.2 deletion.¹¹² Loss-of-function variants in STK11 are associated with Peutz-Jeghers syndrome (PJS [MIM: 175200]), an autosomal-dominant cancer predisposition syndrome. To our knowledge, focal deletions encompassing STK11 have not been reported in individuals with CL/P; however, a deletion of 19p13.3 encompassing STK11 and neighboring genes was reported in an individual with submucosal cleft palate, mild developmental delay, and seizures.¹¹³ As a result, upper and lower gastrointestinal screening was performed which identified polyps consistent with PJS. These data suggest that deletions within 19p13.3 may result in clefting phenotypes, and further support the recommendation that individuals with deletions overlapping STK11 identified as an incidental finding would benefit from additional clinical screening. Finally, UFD1 (also known as UFD1L) falls within the 22q11.2 recurrent deletion locus. The identification of four deletions in our cohort overlapping UFD1 suggests that this gene may be an additional contributor to clefting phenotypes within this genomic disorder region.

For the next phase of our study, we turned to enrichment strategies to explore whether particular genomic regions, or gene pathways, showed statistically significant enrichments connected to clefting or craniofacial development. In order to identify genomic locations that might be enriched in our disease cohort, we performed an over-representation analysis of genes deleted in one or more individuals and at a greater frequency in individuals with clefts versus controls. This analysis revealed several chromosomal regions associated with syndromic CNV disorders that include clefting or other craniofacial anomalies (Tables S7A and S7B), including 1q21.1 and 22q11.21, and provide additional evidence that microdeletions of these regions are likely associated with clefting. We next sought to determine whether this list of deleted genes was enriched for pathways associated with craniofacial development using the IPA platform, including KEGG analysis. Although we were unable to find enriched pathways for recurrently deleted genes, numerous pathways were significantly enriched for singleton deleted genes or singleton deleted genes with haploinsufficiency scores of 10 or less (Tables S7C–S7H; corrected p values <0.05). Several of these pathways play key roles in CNCC function (WNT/beta-catenin signaling; EMT transition; adherens junction pathway; focal adhesion pathway; migration of cells; formation of cellular protrusions; organization of cytoskeleton; Ephrin B/Ephrin receptor signaling; integrin signaling; HIF1alpha signaling; apoptosis signaling; inhibition of matrix metalloproteases; role of osteoblasts, osteoclasts, and chondrocytes in rheumatoid arthritis), whose alteration is associated with neurocristopathies which includes clefting.^{114,115,116,117,118} Collectively, these pathways are represented by a total of 54 genes identified as being deleted in our cohort. It is important to point out that several members in these pathways are shared between the other enriched pathways, suggesting that they likely play key roles in different aspects of craniofacial development and that their loss would be particularly deleterious. Finally, while not found in any significantly enriched IPA pathways, the Rho and Rab GTPase components that we have functionally validated as bona fide craniofacial patterning genes (RIC1, ARHGEF38, COBLL1; see below) all play roles in cellular processes that occur during CNCC development, as does the clefting-associated ISM1 gene that we previously identified in an earlier subset of the larger clefting cohort reported here.²⁰

While statistical methodologies are important tools for associating genes and genomic regions with a particular disease process, functional validation studies in a suitable model organism are critical for proving a gene’s involvement. The substantial amount of CNV data generated from our large cohort of individuals with CL/P allowed us to utilize an unconventional search strategy for identifying putative clefting loci. We made use of population genetics data supporting that rare variants are expected to drive disease phenotypes^119,120 and hypothesized that any genes which were rarely deleted in individuals with clefts yet were deleted with greater frequency in affected individuals versus control subjects might have a higher effect size (i.e., haploinsufficiency loci) and allow for functional validation. Indeed, we previously reported a proof-of-principle study on a subset of the current cohort (∼140 individuals) which resulted in the identification of a Mendelian clefting locus, ISM1.²⁰

Due to their connections to related biological pathways associated with CNC cellular dynamics (Rho, Rab GTPase signaling)^65,66,67 and because Rho and Rab signaling components have been associated with craniofacial anomalies,^{68,69,70,71,72,73} we selected RIC1, ARHGEF38, and COBLL1 for functional follow-up in X. laevis and D. rerio. Although these models do not form a secondary palate, conserved genetic networks regulate the formation of the face across vertebrates,^{121,122,123,124,125,126,127,128,129,130} and the primary palate and the roof of the mouth in X. laevis are thought to be analogous to the primary and secondary palate in mammals.^123,131,132 Therefore, our experiments tested whether the genes are involved in craniofacial morphogenesis rather than specifically in palate formation.

First, we performed in situ hybridization of the three genes at three developmental stages in X. laevis and observed expression of each gene in the cranial region that will later give rise to structures forming the face (Figure S5), thus suggesting a potential role for ric1, arhgef38, and cobll1 during craniofacial development. To further test this hypothesis, we targeted these genes for knockdown/mutation using morpholino (MO) and F0 CRISPR-Cas9 ablation, respectively, in X. laevis. Prior studies have demonstrated that malformation of the structures important for primary palate development results in a change in the shape of the mouth in X. laevis tadpoles. Specifically, the dorsal aspect of the mouth becomes narrower than the ventral aspect, creating a more triangular shape which can appear as a median oral cleft.¹²³ Observation and imaging of X. laevis tadpoles injected with MOs and sgRNAs targeting each candidate gene resulted in the majority of injected tadpoles developing craniofacial defects (Figures 3I, 3N, and 3S). Such defects included midface hypoplasia and a reduction in cranial cartilages. Further, a portion of tadpoles with malformations affecting the craniofacial region also had malformed mouths resembling a median cleft in the primary palate, similar to what has been observed in knockdowns in X. laevis of other genes and pathways associated with orofacial clefting.^20,123

To determine whether the role of these genes in craniofacial morphogenesis was evolutionarily conserved, we performed additional loss-of-function studies using CRISPR-Cas9 ablation in D. rerio. Although an abnormal craniofacial skeleton is a relatively frequent finding in D. rerio models,¹³⁰ we did observe characteristic changes in the craniofacial skeleton for arhgef38, cobll1b, and ric1 mutants, and we replicated the previously reported findings of reduced Meckel’s cartilage with CRISPR-Cas9 mutagenesis of radil1 (which was deleted in one individual with CLP in our cohort) as an internal control (Figure S10).⁷⁸ Collectively, these results indicate that ric1, arhgef38, and cobll1 are required for craniofacial morphogenesis, and our data in developmental models suggest that decreased function of these three genes in humans could result in craniofacial birth defects such as cleft lip and palate.

Intriguingly, our top three candidate genes are involved in either the Rho or Rab GTPase signaling pathways, which have been implicated in a wide array of cellular dynamics including the formation of adhesion junctions, actin organization, cell division, cell migration, and membrane trafficking.^65,66,67 Notably, a recent paper demonstrated the importance of actomyosin dynamics in secondary palate formation whereby midline epithelial seam cells are removed through actomyosin-dependent streaming migration of epithelial trails and islands in order to allow confluence of mesenchymal secondary palate cells.¹³³ In addition, several pathway components of Rho or Rab signaling have been associated with orofacial clefting and other craniofacial anomalies.^{68,69,70,71,72,73} COBLL1 is a WH2 domain containing protein which is involved in F-actin binding and filament formation,¹³⁴ and genes involved in actin cytoskeletal formation and organization have been previously associated with NSCL/P^135,136 and some forms of SCL/P.¹³⁷ Carroll and colleagues identified Cobll1 in mice due to its sequence similarity to Cordon-bleu (Cobl), yet described nonoverlapping expression patterns between the two genes.¹³⁸ While mouse Cobl shows strong expression in the neural tube, Cobll1 is distinctly expressed in the first branchial arch (its first embryonic region of expression, which gives rise to the maxilla and mandible), branchial clefts, and nasal placodes.¹³⁹ These structures are well conserved among vertebrates as regions critical for craniofacial development. Cobll1 encodes a Rho GTPase signaling effector that is expressed in Xenopus CNCCs¹⁴⁰ and in the region of the branchial arches (this report), with Rho GTPase signaling being required for both cell division and migration.⁶⁴ Collectively, these data implicate Cobll1 in developmental events associated with craniofacial development.¹⁴¹

The Drosophila ortholog of RIC1 was originally identified as a gene important for expression of N-Cadherin (CDH2 [MIM: 114020]) within photoreceptor cell synapses in Drosophila¹⁴² and is a binding partner of Rab6 whose function is tightly regulated by guanine nucleotide exchange factors (GEFs) and GTPase activating proteins (GAPs) within the Rho signaling pathway.^{142,143,144,145,146,147} While our work implicates RIC1 in NSCL/P, RIC1 variants were recently shown to be responsible for a Mendelian syndrome that includes cataract, tooth abnormality, intellectual disability, facial dysmorphism, attention-deficit hyperactivity disorder, and clefting (CATIFA [MIM: 618761]),⁷³ in addition to a related syndrome including brain atrophy, microcephaly, and CL/P.¹⁴⁸ Work in D. rerio has demonstrated a role for ric1 in normal skeletogenesis, where it is required for procollagen secretion from craniofacial chondrocytes, with mutant animals presenting with craniofacial anomalies including flattened heads and reduced jaw size.⁷³ Similar to zebrafish, Xenopus with decreased ric1 also had craniofacial defects and reduced jaw cartilages. In Xenopus, the RIC1 paralog Ric-8A is required for CNCC migration, failure of which leads to craniofacial anomalies including clefting,^70,141 and Ric1 itself is expressed in CNCCs¹⁴⁰ and in the region of the branchial arches (this report). Collectively, these data provide evidence for a role of RIC1 in craniofacial patterning.

Although to our knowledge the function of ARHGEF38 has never been investigated, as a member of the GEF family of GTPase regulatory proteins it is likely involved in the regulation of cellular dynamics in an interplay with GAPs, and both GEFs and GAPs are known to control cell migration.^63,64 Consistent with this hypothesis, a recent report correlated high levels of ARHGEF38 with aggressive prostate cancer, where the authors proposed a role for ARHGEF38 in promoting prostate cancer cell migration.¹⁴⁹ Moreover, similar to RIC1 and COBLL1, ARHGEF38 is expressed in CNCCs in Xenopus¹⁴⁰ as well as other head structures, while pathogenic variants in GEF-associated GAPs have been previously identified in humans with NSCL/P,⁷¹ thus suggesting ARHGEF38 may be an additional player in this signaling pathway.

We conclude that although this study supports a role for deletions overlapping COBLL1, RIC1, and ARHGEF38 as causal variants for CL/P formation, additional studies using larger cohorts are needed to more fully define the contribution of rare and common CNVs to clefting. Future investigations within our cohort and others could include utilizing sequencing data to determine whether sequence-level variants are present within the non-deleted allele, and additional functional studies should consider whether any CNVs detected within our cohort disrupt regulatory regions of true causative loci which themselves are not encompassed by the CNVs. These investigations, in combination with familial studies to assess segregation and penetrance, will help to further our understanding of the contribution of CNVs in CL/P formation.

Published: December 8, 2022

Web resources

• dbGaP, https://www.ncbi.nlm.nih.gov/projects/gap/

• DECIPHER, https://decipher.sanger.ac.uk/

• Gene Expression Omnibus, https://www.ncbi.nlm.nih.gov/geo/

• gnomAD, gnomad.broadinstitute.org

• Mouse Genome Database (MGI), www.informatics.jax.org

• National Center for Biotechnology Information, https://www.ncbi.nlm.nih.gov/

• OMIM, https://www.omim.org/

• QIAGEN IPA, https://digitalinsights.qiagen.com/IPA

• VassarStats, http://vassarstats.net/

• Xenbase, www.xenbase.org

• The Zebrafish Information Network (ZFIN), zfin.org

Data and code availability

Plasmids are available upon request. The data from the individuals with CL/P discussed in this publication have been deposited in NCBI’s Gene Expression Omnibus.¹ The accession number for the data reported in this paper is GEO: GSE212296 (GEO: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE212296). The control data discussed in this publication are based on data available within the dbGaP website under phs000523.v1.p1 (dbGaP: https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000523.v1.p1).