Skewed X-chromosome inactivation in unsolved neurodevelopmental disease cases can guide re-evaluation For X-linked genes

Despite major advances in genome technology and analysis, >50% of patients with a neurodevelopmental disorder (NDD) remain undiagnosed after extensive evaluation. A point in case is our clinically heterogeneous cohort of NDD patients that remained undiagnosed after FRAXA testing, chromosomal microarray analysis and trio exome sequencing (ES). In this study, we explored the frequency of non-random X chromosome inactivation (XCI) in the mothers of male patients and affected females, the rationale being that skewed XCI might be masking previously discarded genetic variants found on the X chromosome. A multiplex fluorescent PCR-based assay was used to analyse the pattern of XCI after digestion with HhaI methylation-sensitive restriction enzyme. In families with skewed XCI, we re-evaluated trio-based ES and identified pathogenic variants and a deletion on the X chromosome. Linkage analysis and RT-PCR were used to further study the inactive X chromosome allele, and Xdrop long-DNA technology was used to define chromosome deletion boundaries. We found skewed XCI (>90%) in 16/186 (8.6%) mothers of NDD males and in 12/90 (13.3%) NDD females, far beyond the expected rate of XCI in the normal population (3.6%, OR = 4.10; OR = 2.51). By re-analyzing ES and clinical data, we solved 7/28 cases (25%) with skewed XCI, identifying variants in KDM5C, PDZD4, PHF6, TAF1, OTUD5 and ZMYM3, and a deletion in ATRX. We conclude that XCI profiling is a simple assay that targets a subgroup of patients that can benefit from re-evaluation of X-linked variants, thus improving the diagnostic yield in NDD patients and identifying new X-linked disorders.


INTRODUCTION
Major advances in exome sequencing (ES) technologies and data analysis, along with the continuing identification of new disease genes, have greatly contributed to increasing the diagnostic rate of neurodevelopmental disorders (NDDs).However, despite these advances, from 50-70% of NDD cases remain unsolved [1][2][3].Among the reasons that make molecular diagnostics challenging are: (i) variants in genes that create unspecific phenotypes; (ii) difficulties in interpreting variants of uncertain significance (VUS) [4]; (iii) newly-defined diseases that describe few patients, making it difficult to draw conclusions about phenotypic expansion [4]; and (iv) technical limitations of the diagnostic tools used [5].Several complementary approaches can be attempted to increase the diagnostic yield of unresolved NDDs, such as transcriptome analysis and whole genome sequencing.Re-analysis of ES data has proven to be the most effective, increasing the diagnostic yield by 10-15% [6].
Approximately 6% of NDDs (6% in males; 6.9% females) are estimated to be X-linked [7]where the pathogenic genetic lesions are can often lead to non-random X-chromosome inactivation (XCI) or skewing.Although this process was identified many decades ago, the actual mechanics and fine details of XCI have not yet been completely characterized [8].Physiologically, XCI is random and results in an approximate equal ratio of cells expressing either maternal or paternal X chromosome genes [9].XCI skewing is defined as "preferential" (80:20%) or "extreme" (90:10%) and is a frequent indication of the presence of an X-linked pathogenetic variant, be it in affected females [10] or in the mother of male patients [11].
Females heterozygous for an X-linked pathogenetic variant are usually healthy as skewed XCI favors expression of the wild-type allele, thus protecting females from the deleterious effects of the variant [11,12].More recently, skewed XCI has also been observed in affected females [10], likely caused again by skewing, but this time favouring expression of a deleterious allele that reaches a pathogenicity threshold.Such female patients are susceptible to X-linked recessive conditions with a phenotype similar to that observed in male patients [10].For X-linked dominant conditions, which may be lethal in males, skewing that selects against the deleterious allele has also been observed, thus decreasing lethality [13].
Of the 281 undiagnosed NNDs, 276 were informative by XCI skewing (90 females + 186 mothers of males).We show that XCI skewing analysis combined with targeted re-evaluation of ES data and functional analyses can increase the diagnostic yield and identify novel X-linked disease genes.

MATERIAL AND METHODS Study cohort
From the patient cohort that is part of a large international collaborative study aimed at identifying the genetic bases of NDDs, we selected 92 affected females, and 189 mothers of affected males with negative results after trio-ES, CMA (50 K Agilent) and FRAXA (see Supplementary materials and methods).

X-chromosome inactivation analysis
XCI was tested using DNA extracted from whole blood using an in-house developed protocol.The XCI pattern was calculated using three independent microsatellite polymorphic markers on the X chromosome: (i) the CA repeat in the promoter region of the SLIT and NTRK Like Family Member 4 (SLITRK4) gene; (ii) the CAG repeat located in exon 1 of the androgen receptor (AR) gene [14]; (iii) the CA and AG tandem repeats in the first intron of Proprotein Convertase Subtilisin/Kexin Type 1 Inhibitor (PCSK1N) gene (Supplementary Fig. 1; Supplementary materials and methods).

Characterization of ATRX breakpoints by indirect sequence capture coupled with Illumina sequencing
Xdrop-based enrichment and subsequent amplification of enriched DNA was conducted at the Samplix facility as previously described [15] and subsequently sequenced in 150PE on a NovaSeq6000 (Illumina) (Supplementary materials and methods).

RNA extraction and RT-PCR
To determine which X-chromosome-derived allele (wt or variant) was expressed, we generated and amplified cDNA from total RNA extracted from patients' fresh blood (Supplementary materials and methods).

X-chromosome inactivation phasing by linkage analysis
Families with variants inherited from the mother and segregated in different subjects were analyzed by a set of markers to phase the identified variant with the active or inactive X chromosome and exclude recombination events.The following genetic markers: DXS993, DXS991, DXS986, DXS1068, DXS990, were amplified using AmpliTaq Gold DNA Polymerase (Thermo Fisher Scientific) (see Supplementary Material and Methods for details), separated by capillary electrophoresis on an ABI 3130xl DNA analyzer (Thermo Fisher Scientific) with the GeneScan 500 LYZ size standard (Thermo Fisher Scientific) and analyzed using the Gene-Mapper software v.4 (Thermo Fisher Scientific).

X chromosome inactivation assay
The HUMARA test, based on the analysis of a CAG repeat in the AR locus, is currently the gold standard method for XCI evaluation [14].As HUMARA is not informative in ~21% of females (due to homozygosity or alleles of difficult interpretation) [16], we set up a fluorescent multiplex methylation-sensitive PCR assay that simultaneously amplifies the AR and two additional independent polymorphic microsatellites within SLITRK4 and PCSK1N (Supplementary Fig. 1).Firstly, we evaluated our assay in a female patient with a balanced Xq25;8q24 translocation and complete XCI skewing (100:0), previously assessed by HUMARA [17].Complete XCI skewing was confirmed using the two additional informative loci (Supplementary Table 2), demonstrating the validity of the test.
To further test the assay, we evaluated the XCI pattern in four females with an NDD and four healthy mothers of NDD males: all had received a previous clinical and molecular diagnosis of an X-linked condition with potential skewed XCI (Supplementary Table 3).XCI skewing (>80%) was documented in three affected females and three healthy mothers.Our findings were consistent with the literature which describes the occurrence of XCI skewing in patients with pathogenetic variants in NAA10, PQBP1, MECP2, and ACSL4 [18].Similarly, we found random XCI in a healthy mother heterozygous for a pathogenic IDS variant, which is in line with previous observations indicating occurrence of skewing in one affected female only [18].Finally, our female patient with DDX3X had a random XCI as reported for half of the patients affected by MRXSSB (MIM #300958) [18].No or limited information was available in the literature for HNRNPH2, RBM10.

Genetic analyses in XCI-skewed cases
To exclude the possibility that there were genetic causes underlying the observed skewing of XCI, we sequenced the XIST minimal promoter in the above described 28 cases.The rationale of this method was to uncover rare variants which might cause epigenetic and functional differences between X chromosomes in females as described by Plenge et al. [19].No such variants were found.Thus, we decided to re-evaluate the ES data, focusing on X-chromosome variants, and assessing their relevance also in the light of newly available clinical data.We identified an X-linked variant consistent with the phenotype in three females and four males (7/28; 25%) (Table 2), detailed below.
A novel TAF1 variant in multiple affected members of family #113 The proband was a 2-year-old girl with global developmental delay and delayed psychomotor development, and almost completely skewed XCI (95:5; Fig. 1A, Supplementary Table 2).She had a 14-year-old brother with ID, delayed speech and language development, feeding difficulties, and behavioral abnormalities (II.1); a second 11-year-old brother was healthy (II.2).Their mother (I.2) reported she had had teaching support at school.We found a c.745 G > A p.(Gly249Arg) variant in Transcription initiation factor TFIID subunit 1 (TAF1), a gene associated with X-linked syndromic intellectual developmental disorder-33 (MRX33; MIM# 300966).MRX33 is characterized by delayed psychomotor development, ID and typical facial dysmorphisms (Supplementary Table 8) [20].The variant was inherited from the mother, who also showed skewed XCI (90:10).The variant segregated with the affected brother but not with the healthy brother (Fig. 1A, B).The p.(Gly249Arg) variant was absent in the GnomAD database (ver.2.1.1),and changes a highly conserved nucleotide (PhyloP = 9.37; PhastCons = 1) and amino acid residue, which is maintained from vertebrates to Drosophila melanogaster (Fig. 1C).The change was predicted to be intolerant by MetaDome [21] (Fig. 1D) and deleterious by CADD (Phred: 24.7) [22].Most of the reported likely pathogenic/pathogenic TAF1 variants are missense substitutions that cluster between exons 16-30, whereas p.(Gly249Arg) is located in exon 6 (Fig. 1E).However, using MutScore (which takes into consideration positional clustering of variants already detected in disease-associated genes and variants found in the population), we noted that the variant reached a predicted pathogenicity score of 0.96 (maximum 1) [23].The pathogenicity of p.(Gly249Arg) was also supported by the predicted structural damage triggered by disallowed phi/psi alert in Missense 3D [24] (Supplementary Material and Methods, Supplementary Fig. 4A).
Using a series of microsatellite markers on the X chromosome (DXS993, DXS991, DXS986), we analyzed the segregation of the haplotype containing the c.745 G > A TAF1 variant in the family.Because the haplotype also spanned the AR microsatellite, we could determine the c.745 G > A was located on the inactive X chromosome (X i ) in both the mother (I.2) and her daughter (II.3) (Fig. 1A).
A rare de novo PHF6 variant affecting the female proband of family NWM24 The proband of family NWM24 was a 7-year-old girl, the second child of healthy parents.At birth, she was small for gestational age (SGA), and presented global developmental delay, autistic behavior, several dysmorphic features, divergent strabismus, and brachy/syndactyly.XCI was completely skewed (100:0) (Fig. 1F, Supplementary Table 2, Supplementary Fig. 6A).We identified a de novo c.890 G > T p.(Cys297Phe) variant in PHD finger protein 6 (PHF6; Fig. 1G), a gene associated with X-linked recessive Borjeson-Forssman-Lehmann syndrome (BFLS; MIM# 301900).We reconsidered this previously missed variant because, in the meantime, Table 2. Variants found in the eight families with X-skewed females.de novo heterozygous variants have been described in affected females with an overlapping but distinct phenotype including characteristic facial dysmorphism, dental, finger and toe abnormalities, and linear skin pigmentation (Supplementary Table 9) [25,26].These features are present in our patient.The variant is absent in GnomAD (ver 2.1.1)and changes a very conserved nucleotide (PhyloP = 9.36; PhastCons = 1) and amino acid (Fig. 1H).Cys297 is located within the PHD-like zinc-binding domain where most PHF6 pathogenic/likely pathogenic variants reported in ClinVar (MutScore = 0.949) map to.Cys297 is considered intolerant to change by MetaDome (Fig. 1I; PF13771; a.a.239-330; UniProt: Q8IWS0).Bioinformatic analyses predict the change to be deleterious (CADD Phred = 29.5;REVEL = 0.97; Table 2).Pathogenicity of p.(Cys297Phe) was also supported by the predicted structural damage, the amino acid substitution triggering a clash alert in Missense 3D [24] (local clash score: wild Type =10.47; mutant = 35.67;Supplementary Fig. 4B).

A KDM5C variant with variable expressivity in family #237
In Family 237, we found a 10-year-old girl with moderate ID and skewed XCI (Fig. 1J).She was the second of four siblings that included one affected brother (III.1) and two healthy sisters (III.3 and III.4).The parents were healthy, but several male maternal relatives were reported to have ID.We found a maternally inherited c.1204 G > A p.(Asp402Asn) missense variant in Lysinespecific demethylase 5 C (KDM5C), a gene associated with intellectual developmental disorder, X-linked, syndromic, Claes-Jensen type (MRXSCJ; MIM# 300534; Fig. 1K).The variant was shared by the proband's affected brother (III.1), and one of her healthy sisters (III.3).The variant was predicted to be deleterious by bioinformatic analysis (CADD Phred: 29.7; REVEL: 0.866), and the affected residue mapped to a region that was considered intolerant to variation by MetaDome (Fig. 1L), and conserved from vertebrates to drosophila (Fig. 1M).Another variant affecting this amino acid residue [c.1204 G > T p.(Asp402Tyr)] was previously demonstrated to compromise KDM5C stability and enzymatic activity [27].
MRXSCJ is an X-linked recessive disorder, characterized by DD/ ID with clinical heterogeneity in affected males [28].Recurrent features include short stature, microcephaly, hyperreflexia and aggressive behavior, which were present both in the proband and her brother (III.1).Females with variants in KDM5C, as in case III.2, have only recently been found to be associated with incomplete penetrance and a variable phenotype ranging from mild to severe ID (Supplementary Table 10) [28].The presence of both a male and female in this family initially led us to discard X-linked genes.
By determining the phase of AR alleles and the KDM5C alleles by linkage analysis, we demonstrated that the affected sister (III.2) had a preferentially active mutant allele (90%); conversely, the unaffected sister (III.3) and her mother had a preferentially inactive mutant allele (Fig. 1J).
A genomic ATRX deletion characterized by the Xdrop method in family #236 In family 236, the proband was a 13-year-old boy with a long diagnostic odyssey (Fig. 2A).At 3 years of age, he presented with hypotonia, DD/ID and dysmorphisms.The phenotype was compatible with mental retardation-hypotonic facies syndrome (MRXFH1, MIM# 309580); however, ES was negative for an intragenic ATRX pathogenic variant (MIM* 300032).We found complete XCI skewing (100:0) in the mother, prompting us to reevaluate the genetic data.By visually inspecting the ES reads using IGV [29], we noticed no coverage of exons 3 and 4 of ATRX (Fig. 2B upper panel), suggesting the presence of an intragenic deletion.According to linkage analysis, the X i chromosome in the mother carried the haplotype with the deletion (Fig. 2A).
For in-depth characterization of the deletion, we used the Indirect Sequence Capture (Xdrop technology) [30,31], a powerful method for characterizing specific genomic regions.We enriched for a region of ~100 kb within the ATRX gene, spanning the deletion.This region was subsequently sequenced at high coverage using the Illumina NGS platform.The analysis identified the breakpoints of the deletion (Supplementary Fig. 5), with an uncertainty of 3 bp, identical on both sides of the interrupted region (hg38; chrX:77,697,545-77,703,516; chrX:77,697,542-77,703,513) (Fig. 2B).Remapping of the Illumina reads on the reconstructed sequence demonstrated perfect alignment, without mismatches, thus confirming the correctness of the breakpoints (Fig. 2B, lower panel).The deletion of 5971 bp was confirmed by Sanger sequencing using flanking PCR primers and shown to be inherited from the mother (Fig. 2C, D).

PDZD4: a possible novel NDD gene in family NWM25
In family NWM25, we identified a mother of an affected boy with 90:10 XCI (Fig. 2E, Supplementary Table 2).Since the age of two, the son presented symptoms of DD, followed by the development of kyphoscoliosis with pectus excavatum, hyperelastic skin and joints, persistent hand tremors, facial dysmorphisms and polymicrogyria by brain MRI.Two maternal uncles were reported to be affected by undefined ID.
Re-analysis of the X-chromosome variants led to the identification of a c.2190 G > C p.(Lys736Asn) missense variant in the PDZ domain-containing 4 gene (PDZD4; MIM* 300634) (Fig. 2F), which was inherited from the healthy mother.Lys736 is conserved in vertebrates (Fig. 2G).Using linkage analysis, we showed that the haplotype with p.(Lys736Asn) was located on the inactive X-chromosome (Fig. 2E).
By exploiting GeneMatcher (https://genematcher.org/), we identified a second affected 12-year-old female (II.1; family TF110, Fig. 2H, Supplementary Fig. 6B) with a de novo frameshift c.10_16del p.(Asn4Alafs*12) variant in PDZD4.She presented with an overlapping phenotype, including DD, microcephaly, ID and dysmorphisms.Also in this family we observed almost complete Fig. 1 Pedigree and variant analysis in the three families with XCI-skewed female cases.A, F, J Family trees of families 113, NWM24, and 237.We used X-Chromosome polymorphic microsatellites to reconstruct the haplotypes and to phase the pathogenetic variant on the inactive/active X chromosome (percentage indicated below the symbol of tested females; Xi and Xa indicate the less and the most active X chromosomes).Haplotypes are colored to illustrate the segregation and the presence of recombinants.For each marker, and the gene involved, the physical position in Mb is reported (GRCh37/hg19 reference genome).The hyphen above each symbol indicates whenever DNA was available for genetic testing.B, G, K Sanger sequencing used to confirm the variants in TAF1 (NM_004606.5),PHF6 (NM_01015877.2) and KDM5C (NM_004187.5).Representative electropherograms are shown: wild type (wt); mutant hemizygous (mut); mutant heterozygous (mut/ wt).C, H, M Multiple sequence alignment of the protein amino acid sequences in different species obtained using Marrvel software for the relevant changed aminoacids (highlighted in yellow; http://marrvel.org/)(hs:Homo sapiens; mm: Mus musculus; rn: Rattus norvegicus; xt: Xenopus tropicalis; dr: Danio rerio; dm: Drosophila melanogaster).D, I, L Tolerance Landscape obtained using MetaDome Web Server visualizes regional tolerance to normal genetic variation (https://stuart.radboudumc.nl/metadome/).The position of the missense change is indicated for each gene.The Tolerance Landscape Y-axis is reported as a color scale from blue (position tolerant to variation, T), to yellow (position neutral to variation, N), to red (position intolerant to variation, I).Below the X-axis, a schematic representation of the known protein domains (pink).E Localization of the pathogenic (red) and likely pathogenic (orange) variants reported in the literature for TAF1 gene in male (upper panel) and female cases (lower panel).Our patient's change is shown in black (p.(Gly249Arg)).F, H. Sanger sequencing validation of the identified variants.J In family 234, we sequenced the genomic region (gDNA) and the corresponding transcript (cDNA) in one of the probands (II.1) and their mother (I.2).The wildtype allele only was detected in both cases in the cDNA, showing that the pathogenic variant was not expressed and thus located on the inactive X-chromosome.G Multiple alignment of the protein amino acid sequences in different species as described in the legend for Fig. 1.XCI (95:5) in the proband, although we could not determine if the variant was located on the inactive X chromosome.

OTUD5 a novel recently identified gene in family #234
The probands of family 234 were two brothers, aged 16 and 26 years, with mild ID.Their healthy mother showed complete skewing of XCI (100:0) (Fig. 2I).We identified a missense c.1526 C > T p.(Pro509Leu) variant in OTU Domain-containing protein 5 (OTUD5), a gene that has recently been associated with Multiple Congenital Anomalies-Neurodevelopmental syndrome (MCAND; MIM# 301056) [32].MCAND is an X-linked recessive congenital multisystemic disorder characterized by poor growth, global developmental delay with impaired intellectual development together with variable abnormalities of the cardiac, skeletal, and genitourinary systems.Disease severity is highly variable, ranging from death in early infancy to survival into the second or third decade, suggesting the variant is hypomorphic [32].
We first confirmed that the c.1526 C > T allele was indeed expressed in the patient's blood (II.1, Fig. 2J).Next, we compared the cDNA sequence of OTUD5 from the patient's blood with the OTUD5 genomic DNA sequence (gDNA) from the mother and showed that the c.1526 C > T allele was not detectable, suggesting that the skewed X inactivation preferentially silenced the chromosome with the variant (I.2, Fig. 2J).Bioinformatic analyses predicted the variant to be likely pathogenic (Table 2).The substitution of proline with leucine triggers a structural damage with a local clash score of 33.58 versus a score of 15.21 calculated for the wild type protein (Supplementary Fig. 4C) [24].

ZMYM3: a possible novel NDD gene in family NWM127
In family NWM127, subject II.1 (Fig. 2K) is a 13-year-old male with DD, moderate ID, cryptorchidism, osteoporosis and dysmorphic features.He was the fourth child in a family of European ancestry and had an affected sister (II.2) presenting with severe ID due de novo tetrasomy for 15q11.2-q13.1 (MIM* 608636).He was severely hypotonic in early infancy and showed relevant delay in his gross motor milestones (head control at one year and sitting position at five years).He never developed fine motor skills nor acquired toilet training.Dysmorphic features included long face, tall forehead, thick eyebrows, deeply set eyes, broad nasal tip, and low-set flashy ears with cupped formed ear lobes.Upon reanalysis of the ES data, we found a c.1322 G > A p.(Arg441Gln) variant in the Zinc Finger, MYM-type 3 (ZMYM3) gene.The mother showed completely skewed XCI (100:0), and similar XCI skewing was found in both the II.2 (90:10) and II.3 unaffected sister (85:15), with the mutant allele preferentially inactive.Furthermore, a p.(Arg441Trp) variant was described by Philips et al. in 2014 in three male probands with ID and several dysmorphic features shared with our proband II.1, and was recently confirmed as a recurrent variant in a novel ZMYM3-associated NDD [33,34].Other potentially causative ES-detected variants were excluded by functional analysis [e.g., de novo OSBPL8: c.1535 T > C; p.(Val512Ala)] that did not show altered protein activity (Prof.T. Balla, Bethesda, MD, personal communication).

DISCUSSION
Among the mechanisms that cause deviation from random X chromosome inactivation is selection against cells expressing X chromosomes carrying a pathogenic genetic lesion.We reasoned that we could take advantage of unbalanced XCI and use it as a guide for re-evaluating clinical and molecular data in NDD patients in which previous genetic testing failed to make a diagnosis.To test for XCI, we set up a multiplex fluorescent PCR that simultaneously analyzed the methylation status of three independent polymorphic markers on the X-chromosome.This assay allowed us to increase informativeness to >98%, compared to 80% using standard HUMARA [14].
Analysis of 90 female NDD patients and 186 mothers of male NDD patients, previously undiagnosed by CMA and trio-ES, showed a significant enrichment of subjects with extremely unbalanced XCI, defined as a > 90:10 XCI ratio (28/276, 10%) in line with the results of a similar study [10].The extreme skewing of XCI suggested that some of our undiagnosed cases might be attributable to a gene located on the X chromosome.Proof of principle came from the re-evaluation of available trio-ES data: by focusing on X-linked coding regions, we identified likely pathogenic variants in 7/28 cases, solving 25% of NDDs with skewed XCI.
In our original survey of 575 NDD cases, we had 28 patients with skewed XCI, nine with X-linked variants classified as class 4 (LP) or 5 (P), and 12 with class 3 (VUS) variants.Taking into consideration all these cases, we estimate that X-linked genes account for 6.4-8.5% (9 (LP+P) + 28/575; 21 (VUS+LP+P) + 28/575) of the patients in our survey.Our figures are in agreement with the data from a recent evaluation of the burden of X-linked coding variation based on 11,044 Developmental Disorder patients, which estimated X-linked causes in 6.0% of males and 6.9% of females [7].
We previously missed seven variants on the X chromosome for one of three reasons, namely: (i) the gene was not associated with disease at the time of the analysis (OTUD5, PDZD4, and ZMYM3); (ii) the variant was a structural rearrangement missed by ES (ATRX) or (iii) the variant was overlooked because it was apparently inconsistent with X-linked segregation, since both males and females were affected (TAF1, PHF6, and KDM5C).In the first category, LINKage-specific-deubiquitylation-deficiency-induced embryonic defects (LINKED) syndrome was first associated with pathogenic OTUD5 variants only in 2021 [32].Variants predicted to damage protein structure or function in ZMYM3 have been identified in patients with NDD in December 2022 [33], whereas PDZD4 has been at present only proposed to be disease-causing gene [35].In the second category, the family with a microdeletion in ATRX highlights the importance of searching for genomic rearrangements, exploiting exome data, or performing genome sequencing.In this case the deletion was missed by CMA due to lack of array probes in the deleted tract.The strong clinical suspicion of ID-hypotonic facies syndrome (MIM# 309580) prompted us to analyze the coverage of all ATRX exons on ES data and to finally identify the deletion of exons 3-4.We also chose to locate the precise breakpoints using a novel method based on the enrichment for targeted resequencing by the Xdrop technology, which combines high-resolution droplet PCR (dPCR) with droplet sorting and Multiple Displacement Amplification in droplets (dMDA).This approach proved to be successful in finemapping the deletion breakpoints, narrowing them down from a large putative region of ~20 kb between exons 2 and 5. Given the flexibility of this technology, we expect it to be useful when analysing other similar cases where the large size of the involved region hampers the efficient use of traditional assays for the characterization of structural variations at the single-base resolution.Alternatively, achieving the same results would have required either genome sequencing (more expensive) or a very large sets of PCR-based assays and labor/time intensive work to map the whole 20 kb region, also because the deletion maps within a region rich in repeated sequences.Availability of the deletion boundaries allowed us to set up a simple PCR test to follow segregation of the variant in the family.
The X chromosome is often underestimated in the diagnosis of female NDD patients because of the common misconception that females are less susceptible to X-linked conditions [18].Although many X-linked conditions show a profound sex-linked bias, given the specific mechanism of inheritance, an increasing number of X-linked diseases have been described that occur similarly in both female and male patients [18].For example, in families 113 and NWM24, we identified a missense variant in TAF1 and PHF6 in a female; we overlooked/ignored these variants at the first ES reading because inconsistent with an X-linked recessive disease.However, the literature reported females with phenotypes consistent with variants in those genes.In the case of TAF1, completely skewed XCI is consistent with other recently described cases where the phenotype, which differs in females and males, is uniform within each sex [36].XCI unbalance favors the wild-type allele in both the mother (mild phenotype) and the affected daughter, leaving the pathogenic mechanism unclear.We can speculate that: (i) expression of 5% of the pathogenic allele is sufficient to cause the phenotype or (ii) the XCI pattern is different in affected tissues such as brain, where the pathogenic allele is for some reason more expressed than in blood.In family NWM24, the phenotype associated with PHF6 is consistent with the literature that reports two females carrying the de novo p.(Cys305Phe), just a few amino acids distant from our proband's variant [25,37].Family 237 is another example of X-linked gene complexity: three females carried a missense variant in KDM5C, but we detected skewed XCI towards the deleterious allele only in the individual with the disease phenotype.Segregation analysis showed skewing towards the deleterious allele.KDM5C is known to escape XCI and thus the role of skewing in the phenotype is not clear [38].
Among the various causes of female susceptibility to X-linked conditions, XCI certainly plays a key role at the penetrance level.Although the mechanism of XCI has been known for a very long time, evaluating XCI's influence on phenotype remains challenging.In some cases, the presence of skewed XCI is more easily explained by the selection of cells that inactivate the mutated allele, expressing only the wild-type allele and gaining a selective advantage during the early stages of development [11].Typical examples are mothers heterozygous for OTUD5, ATRX, ZMYM3 and PDZD4 variants who are protected against the deleterious effect of an X-linked pathogenetic variant by skewed XCI.In females with X-linked conditions, XCI can modulate expression of the phenotype [39]; it is likely that there are several mechanisms that underlie disease and skewing that currently escape our understanding and are not always easily identifiable.Finally, in 20 XCIskewed cases, we could not identify any potentially causative variant.We hypothesize that the phenotype might be explained by variants in coding regions not covered by exome sequencing or by noncoding variation, such as deep intronic variants that affect splicing or regulatory regions.
Taken together, our data conclude that XCI testing is a simple, inexpensive and productive means for re-evaluating exome data from the X chromosome.

Fig. 2
Fig. 2 Pedigrees and variants analysis in the three families with XCI skewed mothers of affected males.A, E, H, I, K Family trees of families 236, NWM25, TF110, 234 and NWM127.See legend in Fig. 1.B NGS Coverage of ATRX exons (schematized above) in ES data (upper panel) and with Xdrop enrichment (lower panel) in the II.1 proband from family 236.Xdrop enrichment primers (blue bars below) were designed 5ʹ of the maximum estimated deletion.After enriching DNA for the region, and subsequent Illumina Sequencing, we were able to precisely identify a 5971 bp deletion spanning exons 3 and 4. C. Sanger sequencing validation of the ATRX deletion in II.1 and his mother (I.2) using primers flanking the deleted segment (arrows).The deletion breakpoint is shown in (D).F, H. Sanger sequencing validation of the identified variants.J In family 234, we sequenced the genomic region (gDNA) and the corresponding transcript (cDNA) in one of the probands (II.1) and their mother (I.2).The wildtype allele only was detected in both cases in the cDNA, showing that the pathogenic variant was not expressed and thus located on the inactive X-chromosome.G Multiple alignment of the protein amino acid sequences in different species as described in the legend for Fig. 1.

Table 1 .
Comparison of skewed X-inactivation ratio in adult population, female NDD patients and mothers of male NDD patients.
* Control population data obtained from12.