Coffin–Siris Syndrome and the BAF Complex: Genotype–Phenotype Study in 63 Patients

ARID1A

In ARID1A four variants were identified which were truncating and therefore classified as pathogenic. The two nonsense substitutions were mosaic, and although it was less obvious for the frameshift variants, the variant peaks were consistently lower than the wild-type peaks, which suggest mosaicism (Supp. Fig. S1). In addition to DNA isolated from peripheral blood, we sequenced DNA isolated from cultured fibroblasts in one patient and found the variant in a much higher percentage (Supp. Fig. S1). We did not identify deletions in ARID1A by MLPA.

ARID1B

In ARID1B, 28 pathogenic truncating variants were identified (Table 1). In all cases where parental samples were available (n = 21), these variants arose de novo. We also identified a frameshift variant (c.6700_6701del;p.Leu2234Glyfs*7) in patient 36 near the 3′ end of the coding region (16 amino acids from the coding terminus). The status of this variant remains unclear because parental DNA was not available. We therefore decided not to include this patient in the phenotypic analysis. We also identified several missense variants and in-frame deletions and duplications in the repeat region in exon 1 (not reported), but in all cases, these were inherited from one of the healthy parents and therefore classified as nonpathogenic.

SMARCA2/SMARCA4

In the SMARCA2 and SMARCA4 genes, eight pathogenic variants were identified. All were missense variants, located at conserved positions in different functional domains. The variants arose de novo in all cases where parental DNA was available (n = 5). SMARCA2 and SMARCA4 encode the ATP-hydroxylase containing subunits of the BAF complex, and the missense variants are expected to influence the ATP-hydrolysis and therefore the function of this complex.

SMARCB1

All four pathogenic variants in SMARCB1 were identified in two adjacent amino acids. The variants arose de novo in all cases in which parental DNA was available (n = 3). One variant was found three times (c.1091_1093del; p.Lys364del) and the c.1089G>T;p.Lys363Asn variant was found in one patient.

SMARCE1

In SMARCE1, only a single possibly pathogenic missense variant was identified (c.314G>A;p.Arg105Gln). The variant is located in the high mobility group (HMG). The variant was inherited from the mother who has phenotypic similarities to her son. Grandparents or sibs of mother were not available for testing. Although the base and the amino acid are highly conserved (Table 1), the physicochemical distance between arginine and glutamine is small. Only MutationTaster predicts the change to be pathogenic.

ARID1A/ARID1B

As the pathogenic variants in both ARID1A and ARID1B lead to haploinsufficiency, and size of the coding region is comparable (6,855 bp vs. 6,747 bp), we expected a priori to find a similar number of pathogenic variants in ARID1A and ARID1B. Clearly, this has now been proven wrong as we found four pathogenic variants in ARID1A compared with 28 in ARID1B. Although selection bias cannot be excluded, an appealing explanation is that truncating germline variants in ARID1A are embryonically lethal, especially since all pathogenic variants we identified in ARID1A appeared mosaic in lymphocytes (Supp. Fig. S1). We have meanwhile identified another patient with a pathogenic variant in ARID1A which appears mosaic (data not shown). No ARID1B-knockout mice have been described in the literature, but heterozygous truncating variants in ARID1A haven been shown to be embryonically lethal in mice [Gao et al., 2008]. Tsurusaki et al. (2012) showed the Sanger sequencing results of three variants in ARID1A, in which the mutant alleles of two variants had lower peaks than the reference allele. The c.31_56del; p.Ser11Alafs*91 variant at the start of exon 1 has equal peak heights, which may be explained by (a) the variant is nonmosaic in blood, but mosaic in other tissue, (b) the alternative transcript (transcript ID uc001bmw.1, Ensembl transcript ID ENST0000037415, which lacks exon (1) is more important than the canonical transcript, or (c) there is an alternative translation start site, and the encoded protein has sufficient functionality to be nonlethal. As the c.31_56del variant was not tested in parental DNA, we cannot exclude that the variant is in fact benign.

We identified two deletions in ARID1B and none in ARID1A. This low number of larger deletions may be because most referring clinicians had sought copy number variation using array techniques, although these might not detect small deletions. The limited sensitivity of MLPA in detecting mosaicism may also play a role for ARID1A.

Although in the present study no RNA analyses were performed, Tsurusaki et al. have shown that the c.5632del; p.Asp1878Metfs*96 variant in the last exon of ARID1B seems to escape nonsense mediated decay. This may suggest that pathogenic variants in exon 20 result in a truncated protein and may point toward the presence of protein domains in the 3′ coding end essential for ARID1B function, as for example the BC-box identified in c.6244_6273 [Li et al., 2010]. However, without protein analysis one cannot exclude that the truncated protein or RNA is unstable leading to pure haploinsufficiency.

We found no evidence that missense variants in ARID1B might lead to haploinsufficiency in a recessive manner in our current study or in literature. Only variants leading to haploinsufficiency were considered as pathogenic when we reclassified the EVS data set. No such variants were present in ARID1A. In ARID1B, several frameshift variants in the EVS data might be considered pathogenic (Supp. Tables S1 and S3). The majority of these variants are located in the first exon. As the coverage reported in the EVS database is low in exon 1 (due to the high GC content of this exon) these frameshift variants may be artifacts. Similar to others [Hoyer et al., 2012], we have identified several in-frame duplications and deletions in our patients in exon 1, all of which were shown to be inherited from healthy parents if parental DNA was available. Such variants might contribute to artificial frameshift variants in the context of next generation sequencing variant calling.

An alternative explanation might have been that these variants do not lead to haploinsufficiency because transcription is rescued by an alternative starting codon. As we identified several de novo variants in exon 1, we rejected this explanation.

The c.6008del variant in the EVS is located in exon 20. Both we and Hoyer et al. (2012) (c.6463_6473del) have identified several de novo variants distal of this position, and as the c.6008del variant was reported only once in the EVS, it may be a false-positive. Alternatively, the individual with this variant might represent the mildest end of the phenotypic spectrum of ARID1B patients. As one of our patients has low-normal cognitive ability this possibility cannot be ruled out.

SMARCA2/SMARCA4

SMARCA2 and SMARCA4 are mutually exclusive in the BAF complex [Hargreaves and Crabtree, 2011]. They both have an important function in the hydrolysis of ATP to generate energy to remodel the chromatin. All variants linked to intellectual disability are missense variants in highly conserved functional domains of the protein. Haploinsufficiency of SMARCA4 causes an increased susceptibility to rhabdoid tumors (Rhabdoid Tumor Predisposition Syndrome) [Schneppenheim et al., 2010]. Haploinsufficiency of SMARCA2 has not been linked to human disease. We have hypothesized [Santen et al., 2012b] that the reason for such differing phenotypes between missense and nonsense variants is that missense variants may lead to properly constituted but nonfunctional BAF complex interfering with wild-type BAF complex, leading to a dominant negative effect by competitive inhibition.

The EVS contained 36 and 44 missense variants in SMARCA2 and SMARCA4, respectively. Three missense variants in SMARCA2 are located in the exons 15–25 (Supp. Table S3), which contain the ATP-hydroxylase domains. Each of these variants was found in a single individual, and all of them are highly conserved and predicted to be pathogenic by at least two protein prediction programs. In SMARCA4, 9/44 missense variants were located in annotated protein domains, and seven of these are found in only a single patient (Supp. Table S3). Therefore, these variants may represent false-positive findings or provide evidence that it is difficult to separate nonpathogenic from pathogenic missense variants in SMARCA2 and SMARCA4.

Canonical splice-site variants (one in SMARCA2 and five in SMARCA4) were also found in the EVS (Supp. Table S1). The c.1347+1G>A variant in SMARCA2 is predicted to cause an in-frame skip of exon 7. Exon 7 does not contain an annotated domain and we concluded that this variant is not pathogenic. The same holds for c.1118+1_1118+2insC in SMARCA4, which might also cause a skip of exon 7. The c.2506–2dup variant is predicted to have only a minor effect on the splice acceptor site of exon 19, and is therefore probably nonpathogenic. The other splice-site variants are all predicted to lead to haploinsufficiency, and thus to rhabdoid tumor predisposition syndrome rather than intellectual disability.

SMARCB1

Although SMARCB1 is an established core component of the BAF complex [Wilson and Roberts, 2011], its precise function is not clear. It does not appear to have a DNA-binding domain, and probably acts as some sort of regulator or stabilizer of the complex. Interestingly, pathogenic variants in very few positions of SMARCB1 have been identified in CSS [Tsurusaki et al., 2012], although one de novo variant was described in exon 2 in a patient with a clinical diagnosis of Kleefstra syndrome [Kleefstra et al., 2012]. Amino acids 363, 364, and 377 seem to be vital in the proper function of the BAF complex. As amino acids 352–378 are highly conserved new pathogenic variants may be expected there. Uniprot does not annotate a functional domain for these amino acids, indicating that variants in the other exons might yet be pathogenic. Like SMARCA4, haploinsufficiency of SMARCB1 causes rhabdoid tumor predisposition syndrome [Sevenet et al., 1999]. In addition, haploinsufficiency of SMARCB1 causes schwannomatosis in some families [Swensen et al., 2009].

All SMARCB1 variants in the EVS are classified as probably nonpathogenic in the context of CSS as they are not close to the thus far implicated amino acids. It is worth noting that only four missense variants are reported in the EVS in the other exons of SMARCB1, each of them present in a single patient.

SMARCE1

The c.314G>A variant we identified in the HMG domain of SMARCE1 in a patient and his mother with a similar phenotype remains of doubtful pathogenicity. The earlier reported SMARCE1 variant [Tsurusaki et al., 2012] is more conserved and the difference between the amino acids is larger (tyrosine to cysteine). All prediction programs predict this variant to be pathogenic.

The difficulty in classifying SMARCE1 variants is further illustrated by examination of the EVS test data set. Of the 15 missense variants in SMARCE1, 11 were observed in only a single patient in the EVS, and one was observed in ≥3 patients. This may signify an enrichment of false-positive variants. Only one of these variants was located in the HMG domain (c.330A>C), and is in many respects similar to the possibly pathogenic variant we identified. It has the same predictions by the protein prediction programs, but is slightly less conserved (phyloP 2.5). This may be due to the base being the third base in the codon. However, as very few pathogenic variants in SMARCE1 are known, it remains possible that variants outside the HMG domain are pathogenic.

General Conclusions About the CSS Phenotype

We have identified manifestations that are present in most individuals with CSS and others that are specific to pathogenic variants in one particular gene (Supp. Table S2). The most consistent manifestations, present in at least half of all patients with a pathogenic variant in any of the genes (excluding SMARCA2, as discussed below), are intellectual disability (98%), hypotonia (83%), feeding problems (76%), hypertrichosis (93%), sparse scalp hair (61%), thick eyebrows (92%), long eyelashes (82%), visual problems (62%), thick alae nasi (74%), large mouth (72%), thick lower vermillion (83%), malformed ears (63%), lax joints (76%), short fifth finger (65%), and one or more underdeveloped nails (76%). We consider these the core manifestations of CSS. Some of the findings, especially sparse hair, are more frequently seen in younger patients, with many having sparse scalp hair in infancy which resolves during childhood. Microcephaly has been reported as a sign of CSS [Schrier et al., 2012], but was uncommon in the present series. Underdevelopment or absence of the corpus callosum has been previously linked to CCS [Halgren et al., 2012; Santen et al., 2012a] but was observed here in only a minority of the patients.

Few organ malformations were found in the present series of patients. The finding of dextrocardia in two of the SMARCB1 patients and a horseshoe kidney in another SMARCB1 patient are the exceptions. This contrasts with earlier reports [Schrier et al., 2012].

Visual problems in the present series consisted mainly of severe myopia, up to −18D. Hearing problems were much less common and mainly explainable by recurrent upper airway infections.

Somatic and germline variants in BAF subunits have been linked to cancer [Wilson and Roberts, 2011]. There are no reports of individuals with CSS and a malignancy in the literature. In the present series, none of the molecularly confirmed patients had a malignancy; one of the patients without a detectable pathogenic variant had a clear cell carcinoma of the cervix uteri at age 20 years. Overall this would seem to indicate a low risk for developing a malignancy, although the short median follow-up in the present series (10 years) prevents a definitive conclusion to be drawn.

Manifestations per gene involved

ARID1B patients (n = 28) usually do not have a significantly impaired prenatal or postnatal growth, and have normal brain growth, inferred by normal skull circumferences, moderate feeding problems in two-thirds, seizures in one-thirds, and also an underdeveloped corpus callosum in one-third of cases. In six patients, a scoliosis developed. We found hypertelorism and anteverted nares more commonly than in the other groups. Marked fetal pads were common, but thickened distal phalanges did not occur. Several patients did have long fingers, and a few patients showed an ulnar deviation of the fifth finger. If nails were small this was almost always confined to the fifth rays, and nails on other fingers or toes were of normal size, although these could be dystrophic.

ARID1A patients (n = 4) tended to have moderately to severely impaired development and feeding problems but height and head circumference were usually normal. Seizures were uncommon. They did not show significantly sparse hair but did have hypertelorism, anteverted nares, and a thin upper vermillion. Nails of all fingers and toes were often small but those on the fifth rays were very small. Unusual findings in just one patient were Hirschsprung disease, narrow choanae, and thoracic vertebral anomalies.

SMARCA4 patients (n = 4) had less marked feeding problems and their intellectual disability was considered moderate to severe. Head circumference was around the −2SD as was growth in height. They showed no seizures and had normal vision. Sparse hair was present but only mildly so, and their nares were not anteverted. The distal phalanges of the toes and fingers were not thickened, nails of the fifth rays were small but less compared with ARID1A patients and the other nails were only mildly decreased in size.

Individuals with pathogenic variants in SMARCB1 (n = 4) all showed reduced growth in height and half of them had reduced weight, and often had a microcephaly of around −3SD. Developmental delay was severe, feeding problems were more marked than in ARID1B-patients and had a more prolonged course. Multifocal seizures were common and severe, and have been fatal in one patient. Unusual manifestations were gynecomastia and alacrima, each in one patient. The phenotype is characterized by often sparse scalp hair, very thick eyebrows, and common changes in the nose, philtrum, and mouth. Fetal pads were less marked in this group, and fingers could be long. Decrease in nail size was confined to the fifth rays.

The emerging phenotype–genotype correlation suggests that SMARCB1 patients have the most severe delay, growth retardation, and most distinctive phenotype, whereas SMARCA4 and ARID1B patients are less severely affected. The variability in phenotype seems most marked in ARID1A and ARID1B patients, which may be explained at least in part in ARID1A by their mosaicism. The distal limb anomalies of especially the fifth rays are most pronounced in individuals with a pathogenic variant in ARID1A and least expressed in SMARCB1 patients. Numbers of individuals with pathogenic variants in each of the genes studied here are small and although indicative, no definitive conclusions on the phenotype can be drawn.

Patients without a detectable pathogenic variant presented similarly to those with a pathogenic variant, which is to be expected as these individuals were all clinically diagnosed as having CSS. Differences may be the less frequent myopia and less common hypertrichosis and joint laxity in the group without a pathogenic variant and more frequent general brachydactyly in the latter. The presence of core manifestations associated with CSS was similar in both groups, although feeding problems were usually less marked in those without pathogenic variants. We hypothesize that either mosaicism for the genes studied in the present series, as recently shown for Cornelia de Lange syndrome [Huisman et al., 2013], or variations in other genes in the BAF complex might be identified to be the cause of the phenotype in this group of patients.

In the present study, we enrolled only patients with a CSS clinical phenotype. Pathogenic variants in ARID1B may also, however, lead to intellectual disability without a recognizable phenotype [Halgren et al., 2012; Hoyer et al., 2012; Santen et al., 2012a] and such individuals would not have been ascertained in our study. It remains to be seen how many ARID1B patients actually have CSS.

In the current series of patients referred with a clinical diagnosis of CSS, we have identified four pathogenic SMARCA2 variants. Pathogenic missense variants in the ATP-helicase domains in SMARCA2 (exons 15–25) have been, until now, identified only in NCBRS patients [Van Houdt et al., 2012; Wolff et al., 2012]. Tsurusaki et al. reported a pathogenic SMARCA2 variant in their CSS cohort, but retrospectively reclassified this patient as having NCBRS (N. Matsumoto, personal communication).

The phenotype in NCBRS is considered to be specific and separate from the phenotype of CSS [Sousa et al., 2009]. Main points of difference are the markedly sparse hair that becomes sparser with time, the progressive coarseness of the face, thinning of subcutaneous tissues in the face, markedly broadened and thickened distal phalanges and normal nails. Unfortunately, we had too limited photographic material to reevaluate fully the phenotype of our four patients, but available pictures were compatible with NCBRS. The literature data and the larger series of SMARCA2 patients known to one of us (R.C. Hennekam, unpublished) emphasize the clear phenotypic distinction between SMARCA2 (NCBRS) patients and patients with pathogenic variants in the other five BAF subunits who have CSS. Consequently, it is more likely that the four patients have in fact NCBRS, but our findings underline that distinguishing between the two conditions may be difficult. We recommend screening of SMARCA2 if the clinical diagnosis of CSS is uncertain and the manifestations show overlap with NCBRS.