A total of 59 patients with CS from 57 families (one family comprises the mother and her two sons with similar phenotype) was included in the study. The following inclusion criteria were applied: patients with mostly coronal or multiple suture synostosis with either syndromic or nonsyndromic presentation, and without a detected pathogenic or likely pathogenic variant at previous targeted testing. Thirty-seven patients up to 2016 were retrieved from the Gothenburg Craniofacial Registry at the Sahlgrenska University Hospital, and the remaining 22 patients were retrieved from the clinical laboratory records for patients with negative outcomes following mutation screening between 2016 and 2020. Patient phenotypes were assessed retroactively by corroborating the medical records with registered photos and three-dimensional computed tomography reconstructions. A patient was considered to have a syndromic form of CS if one or several of the following features were present: 1) craniofacial changes involving the eyes (e.g., proptosis, hypertelorism), maxilla (e.g., midface retrusion with relative prognathism, micrognathia), nasal pyramid (e.g., short, small, beaked), and ears (e.g., low-set, dysplastic); 2) neurodevelopmental abnormalities (e.g., motor and/or speech delay, intellectual disability, seizures); and 3) other associated malformations (e.g., cleft palate, heart defect).

The gender proportion was 1:1.2 (27 males and 32 females). A syndromic form of CS was recognized in 51% of families. The distribution of the sutural pattern in the SCS and NCS groups is depicted in Supplementary Figure S1. In the multiple synostosis group (20 cases, including 16 SCS), the involvement of different sutures was as follows: coronal, 16 cases; sagittal, 13 cases; lambdoid, seven cases (5 combined with coronal); metopic, four cases (3 combined with coronal and 1 with sagittal); frontosphenoidal, two cases (combined with coronal); and pansynostosis, one case. Combined coronal and sagittal involvement were observed in 10 of 20 cases and predominantly in SCS. Mercedes synostosis (sagittal and bilateral lambdoid) was noted in two SCS cases. The phenotypic details, including the sutural pattern for each patient, are shown in Supplementary Table S1.

Molecular outcomes according to phenotypic presentation and suture pattern are depicted in Supplementary Figures S2, S3. We detected causal (pathogenic or likely pathogenic according to the ACMG criteria) variants in 19% of the families and 38% of unrelated patients with SCS (Table 1). Seven causal and novel variants were detected in FGFR2, TWIST1, TCF12, KIAA0586, HDAC9, FOXP1, and NSD2, and a de novo deletion in HDAC9 (P2605_132) detected using the in silico panel on WGS data was confirmed by multiplex ligation-dependent probe amplification. In this case, the breakpoints were located upstream of TWIST1, and the deletion included HDAC9, PRPS1L1, and SNX13 (Supplementary Figure S4), none of which had been clearly associated with disease in humans. The typical Apert phenotype in one patient (P_3) prompted a detailed examination of FGFR2, which revealed insertion of an Alu element 18-bp upstream of the boundary between intron 7 and exon 8 (NM_000141.5). This was subsequently confirmed by targeted NGS analysis at an external lab. Additionally, we identified the novel splice variant in TCF12 in our clinical screening and classified it as a variant of uncertain significance. The variant was subsequently upgraded to likely pathogenic after complementary parental segregation and mRNA analyses. cDNA sequencing revealed introduction of a cryptic acceptor splice site (TTTCTAG) at position −8 of intron 18, leading to a frameshift and an early stop codon. Furthermore, we detected larger de novo structural genomic rearrangements in two patients (P_2 and P2605_175) (Table 1). For patient P2605_175, we were unable to confirm the breakpoints in the complex rearrangement involving chromosome 2 by SNP array. Supplementary Table S2 outlines additional findings in CS-relevant genes with possible modulator effects, detected in patients with causal variants.

Novel and possibly relevant variants were detected in 87% of families in which no causal variants were observed (Supplementary Tables S3, S4). When parental samples were available, segregation analysis was performed in cases harboring very low frequency variants in control populations and predicting damaging effects. Thirty-four novel and possibly relevant variants were detected in ADAMTSL4, ASH1L, ATRX, C2CD3, CHD5(2), DMP1, ERF, H4C5, IFT122 (2), IFT140, KDM6B, KMT2D, LTBP1(2), MAP3K7, NOTCH2, NSD1(2), SOS1, SPRY1, POLR2A, PRRX1, RECQL4, TAB2, TAOK1(3), TET3(2), TGFBR1, TCF20, and ZBTB20. New causal and possibly relevant variants were detected in 23 of 37 patients previously screened using the targeted 63-gene NGS panel. Eight variants that had been observed but not reported at previous targeted NGS analysis were confirmed with WGS and reassessed as possibly relevant (Supplementary Tables S3, S4). Additional variants, including novel ones considered unlikely to contribute to the phenotype and/or discarded after segregation analysis, are listed in Supplementary Table S5.

A total of 115 genetic variants were detected using a combination of the in silico panel along with HPO-driven and CNV analyses. The majority of variants (89%) were detected by HPO-driven filtration, of which 50% were identified exclusively by one software program (Moon). The in silico panel detected 41% of the variants, of which 8% were identified exclusively by this filtration method. Six of the causal variants (5.2%) were detected by CNV analysis combined with the in silico panel and HPO filtration in four cases (Table 1). Overall, causal and potentially significant variants were identified in 89% of the unrelated patients. The contribution of WGS/WES to the diagnostic yield (causal variants) was 16% for patients initially analyzed using the targeted 63-gene NGS panel and 25% for patients initially analyzed using the in silico clinical panel.

We report causal and likely causal variants (whereof eight novel) in the following genes (Table 1): 1) CS core genes FGFR2, TWIST1, and TCF12; 2) HDAC9, a gene situated in a candidate region with regulatory elements for TWIST1; 3) genes occasionally associated with CS (e.g., FOXP1, NSD2, and YY1); and 4) KIAA0586, which was not previously associated with CS. Furthermore, we report two de novo chromosomal aberrations (a microduplication of 22q13.1-q13.2 and a complex rearrangement of chromosome 2) and an additional case of a recurrent pathogenic variant in KAT6A.

The previously reported pathogenic variant in KIAA0586 (centrosomal protein homolog to chicken TALPID3) is the most frequently identified variant in Joubert syndrome type 23. This variant has been observed in a homozygous state in two individuals in the control population, suggesting a hypomorphic allele with pathogenic effect only in a compound heterozygous state with more severe variants (Pauli et al., 2019). The phenotype of the CS patient harboring this variant was not suggestive of Joubert syndrome, although some of the features, such as gross motor delay with impaired balance and coordination, were indicative of cerebellar abnormalities. To the best of our knowledge, this is the first case of CS among KIAA0586-related disorders, which are hybrid ciliopathies with a broad phenotypic spectrum ranging from severe skeletal abnormalities to Joubert syndrome. As such, patients may present a clinical picture that overlaps with these entities (Alby et al., 2015; Malicdan et al., 2015).

Recurrent de novo occurrence of the same pathogenic variant in KAT6A was reported by Kennedy et al. (2019) in two patients, one of which presented with sagittal CS. Microcephaly and CS are recurrent features in Arboleda−Tham syndrome, with several cases of midline synostosis and pansynostosis described in association with truncating variants in KAT6A (Tham et al., 2015; Kennedy et al., 2019; Timberlake et al., 2019; Marji et al., 2021; Korakavi et al., 2022). Interestingly, Clarke et al. (2018) reported two rare missense variants in KAT6A in three cases of isolated NCS (two metopic and one coronal).

There is only one report of CS in Gabriele-de-Vries syndrome (Gabriele et al., 2017), with that describing a likely pathogenic de novo missense variant in YY1 encoding a transcription factor [c.1015A>C, p. (Lys339Gln)]. However, facial asymmetry and a broad head have been described in other patients, suggesting that mild CS may be underdiagnosed. Furthermore, the variant in our patient [c.1057T>C, p. (Phe353Leu)] is expected to have occurred sporadically and is located in the same zinc-finger protein domain as the variant reported by Gabriele et al. (2017), thus supporting its potential pathogenic role.

To the best of our knowledge, this study includes the second reported case of CS associated with a de novo truncating variant in FOXP1. Urreizti et al. (2018) reported a patient with metopic synostosis and an Opitz C-trigonocephaly-like phenotype. The sagittal synostosis in our patient may have occurred independent of the FOXP1 variant, explaining only the neurodevelopmental symptoms. However, recent transcriptome studies show that neural crest-derived cells from the sagittal suture of human embryonic calvaria are highly enriched for the FOXP1/2 transcriptional network (He et al., 2021). Moreover, a truncating variant in FOXP2 inherited from an affected parent was recently reported in a patient with sagittal SCS (Tønne et al., 2022). This further supports the involvement of FOXP1/2 in midline synostosis.

The deletion identified in NSD2 is to the best of our knowledge the first limited to exons 1 through 8. Larger structural variants (SVs) involving NSD2 are associated with Wolf–Hirschhorn 4p16.3 deletion syndrome, and loss-of-function (LoF) and missense variants in NSD2 were recently associated with the developmental disorder Rauch–Steindl syndrome (Zanoni et al., 2021). In that study, CS was diagnosed in one of the patients with a de novo missense mutation in NSD2, but the authors attributed this to the co-occurrence of a second de novo missense mutation in AGO2, which is associated with skull deformities, such as scaphocephaly. However, the microcephaly, dolichocephaly, and craniofacial asymmetry described in Rauch–Steindl syndrome suggest that CS is underdiagnosed. The present findings support that CS may occasionally be part of Rauch–Steindl syndrome.

We report the fourth case of CS and deletion involving HDAC9 without disrupting the TWIST1-coding region. Recently, Hirsch et al. (2022) showed that SVs involving HDAC9 disrupt TWIST1-regulatory elements within HDAC9 in patients with CS. Our patient had coronal synostosis that was also described in two of the previously reported patients with deletion and one with a translocation breakpoint disrupting the HDAC9–TWIST1 locus (Hirsch et al., 2022). Interestingly, the craniofacial features of our patient during infancy were not suggestive of Sathre–Chotzen syndrome. No information about possible syndactyly was available, and the presence of dysplastic helices, hearing loss, and a branchial cyst suggested branchiootic syndrome.

CS has not been documented in previous cases of 22q13.1-q13.2 duplications, although prominent forehead with brachycephaly are recurrent features (Samanich et al., 2012; Rahikkala et al., 2013). Notably, de novo mutations in EP300 (included in the 22q13 duplication) are suggested to result in gain-of-function and are associated with a phenotype distinct from Rubinstein–Taybi syndrome caused by LoF variants (Menke et al., 2018). The facial appearance with prominent forehead and short up-slanting palpebral fissures are reminiscent of features observed in association with 22q13.1-q13.2 duplications, suggesting EP300 as a candidate gene for the phenotype. However, the sagittal synostosis in our patient could be related to the paternally inherited MSX2 variant with incomplete penetrance.

Another interesting finding was the de novo complex genomic rearrangement, including a 2p25 duplication and 2q22 deletion in a patient with a Saethre–Chotzen-like phenotype. The duplication involves SOX11, in which a de novo missense variant was reported in a patient with lambdoid synostosis and Coffin–Siris features (Timberlake et al., 2019). We could not determine a clear correlation between the phenotype and the genes included in the 2q22 deletion (containing the known protein-coding genes HNMT, LRP1B, and NXPH2). Only HNMT is disease-associated, specifically with a recessive type of intellectual disability (OMIM # 616739). Both LRP1B and NXPH2 are dosage-sensitive and likely intolerant to haploinsufficiency.

Alu insertions represent a very rare mutational mechanism in Apert syndrome. To the best of our knowledge, there are only three previously reported cases caused by similar Alu insertions (Oldridge et al., 1999; Bochukova et al., 2009). We recommend undertaking specific searches for this type of mutation in patients with typical Apert syndrome and without coding variants in FGFR2.

Overall, we observed enrichment of causal variants in developmental genes, including those encoding transcription factors and chromatin modifiers (e.g., FOXP1, KAT6A, NSD2, YY1) associated with several disorders with pleiotropic phenotypes. CS is a variable feature in these pathologies, which may depend on the widespread downstream transcriptional and epigenetic effects of these genes (Zollino et al., 2017).